The problem of induction is not satisfactorily resolved in the light of the different Indian and Western views discussed earlier, as we have seen. So far as contemporary philosophers are concerned some continue to hold that Hume’s critique of induction is justified as the following quotes show: “My primary purpose … is to support a claim that … Hume’s argument is actually correct. … [T]hat argument has stood since it was first presented, a philosophical classic … withstanding all attempts to overturn it…. Hume’s argument is one of the most robust, if not the most robust, in the history of philosophy.” (Hume’s Problem (HP), C. Howson, Clarendon Press, Oxford, 2000, 2, 10, 14–15) We shall examine if Hume’s and Carvaka’s arguments are irrefutable. In the hope of throwing more light on the problem we look at the later or Navya (New) Nyāya position. We do not touch on all of the highly sophisticated viewpoints. Rather, we address some main arguments in later Nyāya.
We first quote a passage from the Tattvacintāmaņi (TC) of Gangesa. [Modern scholars eulogize Gangesa (thirteenth century) as the founder of Navya Nyāya or New Nyāya. The commentarial and supercommentarial literature on TC includes a number of outstanding works, such as Raghunatha’s Dīdhiti (fifteenth century) and runs into thousands of pages.]
The removal of that [the apprehension of deviation] is sometimes through counterfactual reasoning or CR (tarka) countering the opposite thesis and sometimes comes on its own (svatah siddhah). Should there be an infinite regress because CR is based on pervasion? No. CR is resorted to up to the point there is apprehension [of deviation]. Where the apprehension does not arise at all because of conflict, there pervasion is known without CR. Thus: if smoke were produced neither by the aggregate excluding fire nor by the aggregate including fire, it would not have come into being. Here there may be deliberation as to: Could it be that smoke always comes into being without fire or sometimes comes into being without fire as well or comes into being without a cause? If one is apprehensive that the effect could come into being without the cause with which there is known agreement in presence and absence, why should that very person regularly procure fire for smoke, food for nourishment or words for communication with others? For, that [the effect] could come into being without that. Therefore, procurement of those itself is the obstruction to that kind of fear. (TC 187–94)
There are three main points in this passage. (1) The skeptical doubt about induction may be countered by counterfactual reasoning: a sample is included. (2) The counterfactual reasoning leads onto a follow up argument from belief-behavior conflict. (3) Since the conflict obstructs the skeptical doubt, no additional reasoning is called for and there is no infinite regress.
We first take up (what for the lack of anything better we translate as counterfactual or subjunctive reasoning: CR) tarka or ūha. The Jain philosophers, it may be remembered, have given the same name to a kind of hypothetical reasoning. In the Jain version the reasoning uses hypothetical propositions with a true antecedent and a true consequent. The hypothetical forms are “if this, then that” and “if not that, then not this,” where “this” signifies the pervaded or the probans (vyāpta: hetu) and “that,” the pervader or the probandum (vyāpaka: sādhya): “this” and “that” are replaceable by truth-preserving non-empty names like “smoke” and “fire” to generate true conditionals. However, in a wider sense tarka is of five kinds: (1) self-dependence or trying to prove A from A, (2) mutual dependence or trying to prove A from B and B from A, (3) circularity or trying to prove A from B and B from C and also C from A, (4) infinite regress and (5) undesirable consequence where the first four kinds are included in the last (ATV 863). What is an undesirable consequence? As Varadaraja explains, it is rejection of something reliable or acceptable (prāmāņika-parityāgah) or acceptance of something unreliable (tathetara or aprāmāņika-parigrahah) (TR, verse 70). That is, in a tarka a hypothesis is shown to involve rejecting something reliable or accepting something unreliable and is thereby disfavored. In the context of supporting an induction, tarka proceeds by showing that the supposition that the induction is false leads to an undesirable consequence. (Examples of such tarka are given below.) Specifically, in a narrower version, tarka stands for reasoning with a counterfactual hypothetical proposition that is known to have a false antecedent and a false consequent. The Nyāya philosophers operate with an internal realistic, utility-linked version of correspondence to give an account of truth or reliability (yāthārthya, prāmāņya).1 However, they explicitly label the counterfactual proposition as false while granting that it is subservient or conducive (sahāyaka, anugrūhaka, upayogin, prayojaka) to truth.2
In the TS, tarka is explained as “the factitous supposition (āropa) of the pervader due to the factitious supposition of the pervaded” (351). Thus in a counterfactual reasoning both the antecedent and the consequent of the conditional premise are willfully made false assumptions. Further, the assumptions must have an important relationship. The antecedent must be the assumption of the pervaded and the consequent, that of the pervader (vyāpya-āropeņa vyāpaka-āropah). Based on that relationship one can validly make the counterfactual claim that if the antecedent were true so would be the consequent. In other words, the conditional premise is such that it can reasonably be ruled out from what we know that the antecedent is true, but the consequent, false. [Since the conditional premise is explicitly labeled as false, it follows that the Nyāya logicians are not using material implication. If this were a material implication, the conditional would have been true.]
As an example, the TS (351) cites the proposition “if there were no fire, there would be no smoke.” This may refer to the particular inferential situation where it is known that smoke and (therefore) fire are present. Then it would amount to saying that if there were no fire in a given location, there would be no smoke in that location. Alternatively, the conditional may refer to the imagined absence of all fires and all smokes in the universe. On either construal both the antecedent and the consequent are taken to be false. The antecedent contains the absence of fire that is the pervaded and the consequent, the absence of smoke that is the pervader; that is, it is known in each observed case that where there is absence of fire, there is absence of smoke. From this (and the observation of co-presence of smoke and fire and so on) it has been surmised that fire is a necessary condition of smoke, so that absence of fire implies absence of smoke. Since the premise is about the presence of the pervader on the condition of the presence of the pervaded, that the antecedent is true and the consequent is false is ruled out. This is important. Although the conditional is labeled as false, it is still subservient or conducive to truth. If it were a conditional with a true antecedent and a false consequent, its status would have been different. Since, however, both the antecedent and the consequent are taken to be false, each is described as a factitious supposition (āropa). It differs from an ordinary error (viparyaya or bhrama) where the falsity is undetected. Such a factitious supposition is a kind of āhārya cognition where the characteristic of being aharya is explained as “being willfully caused in spite of falsity” (bādha-kālīna-icchā-janyatva).3
Vyasatirtha has criticized this view of tarka by claiming that this overextends to such a false inference as that of fire from mistaking vapor as smoke (Bāşpe dhūma-bhrama-janya-bhramarūpa-anumitau ativyāpteh, TTD 140). That is, vapor is not smoke. Thus this inference proceeds from the mistaken identification of something with something else that is pervaded and seems to fit the account of inferring through the factitious supposition of the pervaded. But this criticism overlooks that tarka differs from an error where the falsity is not known and that in a tarka both the antecedent and the consequent are willfully made assumptions in spite of being known to be false.
In Vyasatirtha’s view tarka is not merely conducive to truth as the Nyāya holds but is a kind of inference. According to Vyasatirtha, tarka is an inference by way of refutation (dūşaņa-anumāna). For example, the tarka that if there were no fire there would have been no smoke (yadi niragnikah syāt tarhi nirdhūmah syāt) refutes that there is no fire. Since refuting that there is absence of the probandum is based on pervasion (vyāptibalena gamakatvāt), tarka should be accepted as a kind of inference (TTD 139–42). [Although Vyasatirtha accepts tarka as an inference by way of refuting that there is absence of the probandum, he still distinguishes it from an inference that proves directly that there is the probandum (sādhana-anumāna, TTD 139).]
The Nyāya does not agree that tarka should be accepted as a kind of inference that is a source of knowing (pramāņa). Although tarka is based on pervasion, it involves a willfully made false assumption. For the Nyāya in an inference that is a source of knowing each premise must be true or reliable. Since tarka includes a premise that is false, it falls short of the norm of a source of knowing.
In the Nyāya view, tarka lends support to such a general proposition as that wherever there is smoke there is fire. It is given that this general proposition is confirmed by positive and negative examples and neither any counterexample nor any adjuncts have been found. The given general proposition serves as a representative of any other confirmed general proposition of its kind. It is consciously chosen for what is (or may be taken to be) backed by the required kind of observational evidence. It still has to meet the skeptical challenge. It is thus granted that the skeptical challenge cannot be met merely by adding more numerous and more varied observational data. [In other words, by way of comparison, the rules of Pascalian induction, the rules of Baconian induction or, for that matter, the rules of Mill’s methods, even when fully implemented, cannot by themselves resolve the skeptical doubt.] This does not imply neglect of observation. On the contrary, the Nyāya emphasizes the value of repeated, intelligent and varied observation as well as the role of relevant hypotheses. Still, it is held that mere refinement and improvement of observational techniques will not answer the skeptic. If the skeptical challenge cannot be met in the given case, since it serves as a model of its kind, the challenge may very well remain unanswered. In that case, the nyāya that is prized as the paradigm of reasoning will lose a needed premise. [The structural affinity between the nyāya that dominates Indian (and Asian) logic and the categorical syllogism, that dominates traditional Western logic, is remarkable.]
Another counterfactual conditional for supporting the induction that all that is smoky is fiery is: if smoke were deviant from fire (i.e., belonged to a locus of absence of fire), it would not be an effect of fire (BP 771). Of the two counterfactual conditionals, viz., (1) if there were no fire, there would be no smoke and (2) if smoke were deviant from fire, it would not be an effect of fire, the former is described as subject matter–refining (vişaya-pariśodhaka) and the latter as pervasion-supporting (vyāpti-grūhaka) (BP 771). The former is so called because it supports what is to be inferred, the subject matter of inference, by countering its negation. Thus it is argued that if there were no fire [in the yonder hills, say], there would have been no smoke [there]. Since the fire in the yonder hills is not observed, one could deny that fire is there. This supposition or hypothesis (kalpanā) that fire is not there, it is pointed out, conflicts with the observed fact that smoke is there. The latter conditional is so called because it is taken to lend support to the generalization directly (sākşāt) by countering its negation while the former is taken to do so indirectly (paramparayā) (BP 771).
One significant difference between these two counterfactuals is that while the justification for the former would require an appeal to observation, the justification for the latter, given the Nyāya analysis of the concept of a causal condition, would not. For the claim that if there were no fire, there would be no smoke can be sustained only after the connection between smoke and fire has been learnt from observation. On the other hand, the claim that if smoke were deviant from fire, it would not be an effect of fire can be sustained simply on the ground that it is a part of the definition of a causal condition that the latter is an invariable (niyata) antecedent of the effect and, therefore, that anything that is deviant from something else cannot be an effect of the latter. This results from mental reflection (mānasa-jñāna) on the contents of the definition. It seems further that the justification of the former would eventually include an appeal to causation. The point is that even if smoke is observed regularly with fire without any exception, it cannot be claimed merely on that ground without begging the question that if there were no fire, there would be no smoke. The brunt of the skeptical critique is that such inductive reasons are no reasons. Hence the claim can be justified only by linking fire and smoke as cause and effect. This is why the latter counterfactual gets priority over the former, for the latter utilizes the causal connection explicitly.
Another counterfactual conditional, cited in the earlier passage from Gangesa, brought in defense of the said general proposition, is the following: if smoke were produced neither by an aggregate including fire nor by an aggregate excluding fire, it would not have been produced (TC 192).4 For this counterfactual, too, like the second one, mental reflection is involved in seeing it as being conducive to truth, but not only because it exploits the concept of cause (as does the second one), but also because of its logical structure [in the Nyāya terminology the latter is describable as a relation holding at the level of contentness (vişayatā)]. If smoke [or anything] were to be produced, it must be produced either by a collocation of causal conditions including fire [something] or excluding fire [that thing]. These two alternatives are not collectively exhaustive but are mutually exclusive and both cannot be true. Thus the truth of this counterfactual depends in a significant way on the logical structure as well as general intuitions about the nature of causation.
In the above conditional, “smoke being produced neither by an aggregate including fire nor by an aggregate excluding fire” is the factitious supposition (āpādaka) and “smoke not being produced” the factitious consequence (āpādya). Since it is known that smoke is produced, the consequence part is false. The falsity of the consequence proves the falsity of the antecedent. Thus it follows that it is not the case that smoke is produced neither by an aggregate including fire nor by an aggregate excluding fire. In other words, it follows that smoke is produced either (a) by an aggregate including fire or (b) by an aggregate excluding fire. Since both the alternatives (a) and (b) are logically possible and also have factual contents, the choice between the two cannot be based on logic alone; we have to go beyond logic to the world of observation. Accordingly, the alternative that is favored by the data from observation is to be preferred. Gangesa has not explicitly stated this epistemic principle but it is without any doubt implied and useful for understanding his answer to the skeptical challenge to induction as well as for justification of empirical truths in general. Thus, it may be laid down as a general epistemic principle (favored by empiricism) that a factual claim that is backed by observation is preferable to one that is not. This may be called the principle of observational credibility—OC for short. [A similar empiricist principle is that a factual claim that has greater observational support is preferable to one that has less observational support. This could be called the principle of greater observational credibility—GOC for short. GOC is not needed in the present context.]
OC is a meta-principle presupposed in the acceptance of particular empirical claims as reliable (prāmāņika). For example, suppose that I have to choose between two particular factual claims such as that this table is green and that this table is yellow. Suppose further that as I look at the table I see it as green and not as yellow. That provides me the basis to say that the table is green and not yellow and in the process I am implicitly relying on OC. Since OC is presupposed in accepting any particular empirical claims as reliable, OC is not an empirical induction—for that would amount to putting the cart before the horse. In the present context, the reliability of particular observations is not in dispute by Carvaka or Hume. Hence a general empiricist principle that is needed to make sense of the reliability of particular observations should also not be in dispute. Needless to say, reliability of particular observations as well as empiricism as a whole may be challenged. But responding to such a challenge is not our task at hand. Our task is to respond to the Carvaka-like and also incidentally the Hume-like challenge to induction. Further, following Gangesa, our discussion is limited to only generalizations in which the pervaded is an effect and the pervader is a causal condition. It may be indeed possible to extend the discussion to other cases of generalization as well as to induction in a broader sense. But that is not our task here.
Given OC that is acceptable to both the Nyāya and empiricists in general, it follows that the first alternative (a), viz., that smoke is produced by an aggregate of causal conditions including fire, is true or reliable. This bestows favor (anugraha) on the induction that all smoky things are fiery. This follows from the definition of a causal condition. [Nyāya arguments for the causal law are briefly stated later.] A causal condition is defined in part as a constant condition. “Constancy” is needed to leave out accidental factors such as a donkey that may happen to be present where an effect like smoke is produced and is not a causal condition. Given this definition, that all smoky things are fiery is true or reliable (prāmāņika).
This argument, it may be noted, implicitly makes use of the rule of double negation, the De Morgan rule, the rule of disjunctive syllogism, and the rule of modus tollens. The argument may be reformulated and the formal structure explained as below. Let p symbolize “smoke is produced by an aggregate including fire,” let q symbolize “smoke is produced by an aggregate excluding fire” and let r symbolize “smoke is produced.”
| 1. (˜p & ˜q) ⊃ ˜r | ||
| But | 2. ˜˜r | |
| Therefore | ˜(˜p & ˜q) | (modus tollens) |
| Therefore | ˜˜p v ˜˜q | (De Morgan) |
| Therefore | p v q | (double negation) |
| But also | 3. ˜q | |
| Therefore | p | (disjunctive syllogism) |
The formal part of this argument also is of historical interest. These Sanskrit works belong to a period long before the rise of modern logic. In that period similar logical acumen making implicit use of the De Morgan law in particular is missing in other logical traditions of the world.5
It is clear that in arguing for the reliability of induction Gangesa has implicitly relied on some logical laws and an epistemic principle called OC. Needless to say, even the logical laws are not above challenge; still, they are as safe as it gets in the world of philosophy. They are also not rejected by either Carvaka or Hume. So far as OC goes, neither Carvaka nor Hume should disown it. Carvaka challenges the rationality of inductive leap but holds that particular observations may be reliable and are the only sources of knowing (pratyakşaikapramāņavāda). Hume also questions if any reason can be given for induction but holds that (impressions and) observations of particulars are the foundations of all knowing. Neither position may be sustainable without OC. So the argument of Gangesa is right on the target.
Another way of seeing the point following Bhavananda is that the skeptical supposition that smoke is sometimes caused by something other than fire is uneconomical (prayojakāntara-kalpane … gauravāt, TCDP 600). The lack of economy is based on cognitive link (upasthiti). That is, of two suppositions the one with a closer link to something known is to be preferred. Since smoke is observed to arise where there is fire, the supposition that smoke is sometimes caused by something other than fire is more removed from what is observed than that smoke is caused by fire. Thus the rejection of skeptical doubt need not be based on animal faith or instinct but could be based on a principle of reason such as OC or even the law of parsimony.
It may be noted that one point of exploring these counterfactuals seems to be that these help to show a degree of continuity and affinity between (by borrowing modern terminology) deduction and induction. [From a typical Indian point of view the distinction between nigamana—or extracting what is implied from something given—and āgamana, vyāptigraha—or moving from the particular to the general, as also between what is para or independent of experience and apara or dependent on experience—must be drawn. But that does not warrant the conclusion that any given knowledge claim is exclusively deductive or inductive or a priori or a posteriori. Nothing here should of course be taken to suggest that the relevant Indian and Western concepts are quite the same.] Although some passages taken out of context may suggest otherwise, the Nyāya logicians have never tried to show that induction is at bottom deductive (as Aristotle is alleged to have done) or replace induction with the hypothetico-deductive model (like Karl Popper) or defend induction on purely a priori necessitarian grounds (like early Pierce and D. C. Williams). But the skeptic presumably does not dismiss deduction as irrational and is not also begging the question and equating openly rationality with deducibility. [If the skeptic does claim that being rational is synonymous with being deducible, no meaningful debate, from the Nyāya point of view, is possible, for induction is, admittedly, not formally valid. In other words, in order to have a meaningful debate, it must be possible for the inductionist to show that induction is rational without having to reduce it to a valid deduction.] From this point of view the above discussion is relevant. What that shows is that our run-of-the-mill general propositions have counterfactual implications the truth of which depends in a significant way on their logical structure (or, in the Nyāya terminology, is discernible [partly] by mental reflection and [significantly] dependent on the relation within contentness). Deduction and induction then are not diametrically opposed, as it may appear in the beginning, but are analogous and kindred in an important way. Because of the analogy and kinship the inductionist may now plausibly claim that induction too is a rightful candidate for being rational.
The skeptic may retort that arguing analogically falls short of demonstrative proof and amounts to having recourse to a species of induction to vindicate induction. But Nyāya logicians would refuse to fall into the trap of having to prove rationality of induction deductively on purely noninductive ground. That is an impossible task. If that is what the skeptic dogmatically insists on, there can be no real debate, as already said, for there is not enough common turf and, therefore, not enough room for resolving the differences.
For a better understanding of what is at stake here, let us think of the scenario where a conservative and a liberal try to discuss and resolve their differences over cultural diversity. It may soon transpire that the conservative has already made up his mind about defining culture through certain criteria, say a, b and c, which apply only to his chosen model. On the other hand, the liberal does accept a, b and c as cultural criteria but also adds certain others, say d, e and f, which are somewhat analogous but still significantly different from a, b and c. The liberal may try to persuade the conservative that something (say, with the features a, b and d) other than the latter’s chosen model should also be accepted as (an advanced) culture and the latter may try to persuade the former that it should not be so accepted. But assuming that both will stick to their positions, this is a dispute that cannot be fruitfully resolved. Similarly, the dispute over induction cannot be fruitfully resolved if the skeptic has already conceived rationality in such a way as to fit only deduction and the inductionist flatly asserts that inductive reasons are rational in their own right.
One of course assumes for the health of philosophy that both are willing to reconsider their positions and continue the debate. But that would require fulfilling at least two conditions. The skeptic must refrain from assuming that deducibility and rationality always go together and allow, at least provisionally, the inductionist to bring in some nondeductive considerations to make it possible to show that induction is rational. The inductionist too must allow the skeptic to show why induction is still irrational on such grounds as circularity. It is in this spirit that the Nyāya logicians appear to stay in the debate.
Nyāya logicians are not claiming that it is rational to infer directly from the way the world was or is to the way the world will be. They agree with the skeptics [and this shows how close they are to skepticism and what a major concession they have made without being skeptics] on the following: the fact that smoke has always followed fire in the observed past or present does not by itself give the rational ground to infer that it will be so tomorrow, or that it will not be so tomorrow. Thus external experience alone does not provide such a ground although for knowledge of external things the mind (more accurately, the inner sense: manas) is totally dependent on the external senses [in modern terms, there are no a priori connections in the realm of experience]. They also agree that the ground requires the office of mental reflection (mānasa-jñāna).
Since no direct inference from experience to the future is justified, either the skeptic wins or some indirect way must be found. Accordingly, they introduce the counterfactual reasoning for the latter purpose. This deals with counterfactuals and with what would have been. This is not surprising, for our beliefs about the future are not merely beliefs about what will actually happen in the future. They also include beliefs about what would have or could have happened. For example, “whoever jumps off a tall building (and crash lands without protection), dies” includes the belief that if I were to do that I would meet the same fate. This is why I do not do that and make sure to the best I can that it does not happen. This is more patent for general statements that are, by borrowing modern terminology, vacuously true. For example, consider: “an eternal entity that is independently productive is productive for ever.” Nyāya philosophers accept this, although there are, in their view, no eternal entities that are independently productive. This still makes sense, for what is implied is that if there were to be any such thing it would have been so. [One may, if one wishes, change the example to something more modern: for example (Newton’s First Law) that if no force is exerted on a body, its acceleration is zero.] Further, and equally importantly, what would have been is not what was or is or will be. What was or is or will be belongs to the real world, what would have been does not. [Nyāya philosophers do not subscribe to the realism of possible worlds as David Lewis and others do.] The counterfactual situation, by definition, will not be realized and observed. Since we are not dealing with future external events about which the mind must learn from the external senses when the event will take place and since all relevant information is already available, the world of what would have been is a realm where the mind has its legitimate sway. Thus, by resorting to counterfactual reasoning and exploring what would have been, Nyāya philosophers seek to justify the claim about what will be and about all unobserved cases.
An accidental and false generalization does not hold up when we explore its modal character and try to support it by counterfactual reasoning. Consider: wherever there is fire there is smoke. Suppose that we argue like this: if there were no smoke, there would have been no fire. This is patently falsified by the counterexample of a red hot iron ball where we see that there is fire but no smoke. But consider: wherever there is smoke there is fire. It does hold up when we explore it subjunctively. This then is a significant difference between the two generalizations that is brought to light by probing counterfactually. The counterfactuality is conveyed by the formulation of the tarka in the subjunctive mood and the explicit labeling of both the antecedent and the consequent of the conditional as factitious.
The skeptic could retort: how do we know that there is a significant difference? He might argue that using counterfactual language does not really change anything. The difference between “If there were no smoke, there would have been no fire” and “If there were no fire, there would have been no smoke” is equivalent to, he might say, the difference between a straightforward causal statement that has known counterexamples and one that does not.
But such a retort would from the Nyāya point of view overlook the peculiar nature of counterfactuals. Since counterfactuals deal with what would have been and since the latter is not a part of the real world, the realm of what would have been, as said, can be justifiably explored by the mind. This has been brought out earlier by the exploration of the counterfactual argument that if smoke were produced neither by an aggregate including fire nor by an aggregate excluding fire, it would not have been produced. Thus the difference in the epistemic values of the two above counterfactuals is shown through mental reflection. [The recognition of this role of mental reflection does not in any way compromise the basic empiricist position that the mind is totally dependent on the external senses for information about the external world.]
Again, if one questions the above generalization, one must also question that smoke is caused by fire and invariably preceded by the latter. Then one should suppose further that possibly smoke is produced by an aggregate of causal conditions that does not include fire. But such a supposition is no more than a mere speculation and has no empirical evidence to back it up. It is thus no better than such an idle speculation as that possibly there are crows having teeth. This latter supposition is not self-contradictory and is logically possible. Nevertheless, there is nothing in what we have observed about crows and teeth that gives the slightest credence to the supposition. Similarly, although it is logically possible that smoke is sometimes produced by a sum total of causal conditions that excludes fire, there is nothing in our observation of smoke and fire that supports such actually being the case. [Further, the supposition is uneconomical due to a relative lack of a cognitive link (upasthiti) compared to that smoke is caused by fire.]
Thus, by exploring the consequences of the skeptical doubt about induction, it is shown to involve claims about possibly observable situations that are empirically baseless as well as uneconomical. Factual possibilities in the external world are determined not merely by a priori speculation but also, additionally and more importantly, by observations of what there is. [This is a corollary of OC.] Under the circumstances, a skeptic who would persist with the possibility of inductive deviation is no better than someone who would persist with the possibility of examining, say, the teeth of crows and only deserves to be ignored (upekşanīya). This does not show that skepticism about induction is logically impossible. [We have already said that Nyāya philosophers are not up to that, which, incidentally, shows their difference from analytical rationalists like Strawson or Ayer.] Still it shows that such skeptics do not qualify either as commoners (laukika) or as experts (parīkşaka) whose opinions are, to the Nyāya, the prime sources of philosophical material. A philosopher is entitled to evaluate common as well as expert opinions. But the results of his evaluation must find acceptance among commoners or experts. Otherwise, if a philosopher does not exercise some judgment and attaches an equal weight to any and every opinion (such as that possibly crows have teeth, or that this thing which everybody else in the room says is a table, is not a table but an elephant), he cannot even get started. Accordingly, if the skeptic’s (empirically baseless) opinion about the possibility of deviation fails to coincide with either common opinion or expert opinion, it only deserves to be rejected.
Needless to say, the inductive claim about all cases (observed and unobserved), though justifiable, is also falsifiable and would remain so. Gangesa and other Nyāya philosophers are very clear on this. But that is very far from saying that the skeptical doubt about induction is justified. As the exploration of counterfactual conditionals shows, the skeptical doubt involves claims about possibly observable situations that are empirically baseless and uneconomical and, therefore, unjustified.
The skeptic could again try the old rejoinder that the above proves only that there are as yet no known counterexamples to some inductions. He could reiterate that the inductionist is still committed to assuming that the future will be like the past and insist that observations about what is tells us nothing about what will be. Thus the skeptical doubts are no more empirically baseless or unjustified, he could say, than the inductionist’s claim to knowledge.
But such a defense for the Nyāya would amount to conveniently bypassing the points made by the inductionist above without trying to meet them. Since the observations about what is, for the skeptic, tell us nothing rationally about what will be, the skeptical claim about possible unobserved counterexamples is a claim about an observable situation, which claim is merely speculative. But the inductionist is neither indulging in assuming that the future will be like the past nor merely speculating about what the future holds for us. Instead he is basing his claim about the future on the mental exploration of what would have been as brought about by the counterfactual argument. Thus, while the inductionist is able to utilize the crucial bridge of what would have been between what is and what will be, the skeptic is unable to do so. Since the latter’s claims about future observable situations cannot be justifiably based on merely claims about what is logically possible, the latter’s position seems to be significantly weaker than that of the inductionist.
Nyāya philosophers emphasize that an exploration of the law of causation is useful for a resolution of the problem of induction. Thus Raghunatha says: “Knowledge of the cause-effect relation, too, must be investigated, for knowledge of pervasion is dependent on that” (GD 680). He says further:
When one sees the co-presence and co-absence of smoke with the aggregate including fire, donkeys, etc., one comes to the conclusion that one of these must be the cause of smoke…. There of those belonging to the aggregate that without which smoke is found to be produced is ascertained not to be the cause, such as the donkeys. That without which smoke is found not to be produced in spite of the presence of all the others in the aggregate is ascertained to be the cause, such as fire. (GD 676)
As already said, the smoke-fire case serves as a paradigm and “smoke” and “fire” play the roles of quasi-variables with smoke representing any effect of its kind and fire any cause of its kind. Using the paradigm Raghunatha is in so many words recommending the joint method of agreement and difference for the purpose of eliminating connections that are accidental, such as that between smoke and donkey, and for finding connections that are causal, such as that between smoke and fire. [This is of historical importance considering that Raghunatha (fifteenth century) is long before Mill and also before Bacon. Raghunatha is probably the first philosopher to have stated the joint method explicitly. It remains true, of course, that anticipations of the methods of agreement (anvaya, sādharmya) and difference (vyatireka, vaidharmya) are found in Indian writings (as well as Western writings) from early times.] He points out that in every observed case where smoke is produced, fire is included in the collocation of things immediately preceding it. Thus fire is a uniformly common factor in each such collocation (agreement in presence). Further, in every observed case where smoke is not produced in spite of the presence of all other factors in the collocation, fire is found to be absent (agreement in absence). This establishes fire as a cause of smoke. Once fire is known to be a cause of smoke, the suspicion that smoke may deviate from fire in unobserved cases is removed. Gadadhara has observed:
Being an effect is opposed to being deviant…. Cognition of being an effect removes the apprehension of deviation by way of putting forth the counterfactual argument (tarka) that if smoke were deviant from fire, it would not have been a product of fire. (GD 681)
The Nyāya thus has been drawn into defending the causal relationship against skeptical onslaughts. Since, however, this is a large topic in its own right and would require a great deal of space even for a preliminary discussion, we shall look at it very briefly and only in outline (though we do summarily present some powerful Nyāya arguments for upholding causality below). The Nyāya philosophers have argued at great length to try to show that such views as that things originate without any cause (ahetuka) or that things come into being merely by chance (ākasmika) or that the origin of things can be explained merely by an appeal to their own nature (svabhāva) do not survive sustained and tenacious philosophical criticism. They also reject after a prolonged examination the Sāmkhya view (somewhat similar to Aristotle’s view) that there is an essential identity and continuity between the cause and the effect so that the latter is potentially contained in the former. They are further averse to the idea that a cause has the power (śakti) to produce the effect. The Mīmāmsā philosopher Kumarila Bhatta has championed the doctrine of causal power (a similar view is held by Locke); the Nyāya has subjected it to a detailed and careful examination and refutation.
In defending causality the Nyāya is not subscribing to a necessitarian view of nature (shared with some differences by both Platonists and Aristotelians) that has dominated traditional Western philosophy. From the latter point of view cause and reason are very closely linked. Knowledge is of first principles and what is deduced from them. Hence scientia, according to medieval scholastics, must get at the essence of things and proceed by the demonstration of effects from first causes. But the influence of this view by no means ended with medieval scholastic philosophy. Even Bacon, the father of Western inductive logic, held that knowledge is derived from common notions and that we seek true axioms and real notions that eventually produce knowledge and not opinion. Descartes tried to demonstrate the laws of planetary motions, the laws of refraction of light and even that the blood must be red. For Leibniz there is a sufficient reason for any truth and it can be proven a priori. Scientists of this period aspired for demonstrative knowledge of primary qualities although they could perform experiments only on secondary qualities. Causes were thought to be the domain of respected sciences like optics, astronomy or mechanics where demonstration seemed to be achievable. Inferior sciences like geology or medicine had to be content only with signs that relied on observed association without backing of demonstration.
The wedge between knowledge and opinion was retained by Hume, for whom knowledge was confined to mathematics and the like and evidence short of deduction, like the medieval thinkers, was not really evidence at all. Knowledge was still of first principles in a sense and what can be demonstrated from them. Only no scholastic causes and necessary connections were to be found in nature, as Robert Boyle and company were persistently driving home. The way out was to dissolve the longstanding marriage between cause and reason. This Hume did and in the process collapsed the division between cause and sign. Causes exhibiting nothing more than regular association could not to be closeted with reasons and, therefore, had to be closeted with signs. In other words, demonstrative knowledge advertising a priori reasons on the one hand and irrational opinion smuggling in associative signs on the other were the only choices. Since the first was ruled out for causes, the second had to be the case.
There is much in Hume’s crusade against a priori necessities in nature that the Nyāya would share. There are no logically necessary connections between causes and effects for Hume; the Nyāya does not dispute that. The latter too argues, against the Sāmkhya, that cause and effect are distinct existences and would agree with Hume that the ideas of a cause and its effect are distinct, so that a particular cause is always conceivable without its effect, and a particular effect without its cause. Thus the drive to get at the a priori essences of natural phenomena and proceed by the demonstration of effects from the first causes is foreign to the Nyāya thought. The latter further agrees that all that is observed for causation is constant conjunction and dismisses causal power. Thus there is no power that if we found it in a cause would tell us at once that the cause would bring about the effect. There is also agreement between the Nyāya and Hume on the foundational role of causation for inferences concerning matters of fact.
But the Nyāya would not give a psychological explanation for causation and conclude like Hume that causality is only in the mind. Although regular succession is what we observe and there is no causal power, causation is still objective and not projected by the mind onto things. The Nyāya would also disagree that the experience of constant conjunction does not provide materials for any rational inference from cause to effect (or vice versa) in a new instance.
Why this difference between the Nyāya and Hume? One main reason is that the former does not regard demonstrative knowledge and irrational opinion as the only choices, a vestige of scholasticism in Hume. The former also thinks that deduction is only one way of giving reasons and does not hold that evidence falling short of deduction is no evidence at all.
What are [while looking at it very briefly] some of the Nyāya reasons for upholding causation? First, causal and accidental connections are separable by exploring the corresponding counterfactual conditionals. Thus, if smoke is claimed to be caused by donkeys, the corresponding conditional is: if there were no donkeys, there would be no smoke. When we observe in a kitchen that there is smoke but no donkeys, the conditional is found to have a true antecedent but a false consequent. This is never the case when the connection is causal. Thus when a counterfactual conditional is found not to have a true antecedent and a false consequent, it is accepted as being conducive (anugrūhaka) to truth. Nyāya philosophers do not claim such a conditional to be true. Still its epistemic value must be different from some other conditional that has a true antecedent and a false consequent. The difference in the epistemic values of the two sorts of conditionals provides a ground for separating causal from accidental connections.
[Further: (1) Nyāya philosophers have added absence of obstruction as a general causal condition. (2) While each causal condition is regarded as a necessary (niyata) condition, Nyāya philosophers regard the sum total of causal conditions as the sufficient (phalāyogavyavacchinna) condition. (This notion of a causal aggregate (kāraņa-sāmagrī) is similar to that of a causal field introduced by John Anderson to resolve difficulties in Mill’s account of causation.)6 (3) They distinguish between triggering (phalopadhāyaka) causes and predisposing (svarūpayogya) causes. (4) They elaborately study causal irrelevance (anyathāsiddhatva). While discussing these is beyond the scope of our inquiry, one hopes that various difficulties that may crop up in the course of the conditional analysis may be resolved in the light of these.]
Second, the observed fact of the occasional nature (kādācitkatva) of the effect (i.e., that the effect is produced only on the occasion the cause is there) points to the dependence (sāpekşatva) of the former on the latter, which in its turn, points to causation.7 While other explanations are not logically impossible, no other explanation in the Nyāya view gives more economy (kalpanālāghava) or does a better job. The inference from occasionality to causality is justified, because no better explanation is available (ananya-gatikatayā: literally, “because there is no other reasonable gati or explanation”). In particular, as Udayana argues, if the effect is not dependent for its origin on the causal condition, why does it not come into being anywhere and anytime (NK 1.4 and 1.5)? Effects do not happen at all places and all times. They happen only at particular places and particular times. These particular places and times are where and when certain conditions are fulfilled. These conditions put limits (avadhi) to the possible places and times for such happenings and such limits are constant (niyata). There are no effects that happen anywhere and anytime: it is impossible for an effect to happen anywhere and anytime.
An effect is that which comes into being and was nonexistent before (prāgabhāva-pratiyogin). It follows necessarily that there are times and places when and where the effect is nonexistent. The nonexistence comes to an end only under certain conditions that accordingly provide the limits to the possible times and places for the thing’s existence. Without such limits the effect could exist anywhere and anytime and could not be an effect: something that exists anywhere and anytime (sadātana) is not an effect. Thus the occasional nature of effects cannot be explained without the acceptance of limits and that is tantamount to the acceptance of causal conditions. In other words, occasionality presupposes limits and limits are nothing other than causal conditions. This defense of causality seems to be satisfactory and no sound rebuttal from the skeptic seems to be available.
An important part of the concept of causation is that effects depend on their causes. Nyāya philosophers often express this dependence in terms of counterfactual conditionals. Since the causal condition is a constant antecedent (niyata-pūrvavŗtti), if f is a causal condition of s, s would not have taken place if f had not taken place (tat-asatte tat-asattā). Such counterfactual analysis seems to be natural and has been adopted by some contemporary philosophers just as it has been criticized as well. (See, for example, G. Bjornsson, “How Effects Depend on Their Causes,” Philosophical Studies, 133/3, 2007). It is remarkable that such criticism has been implicitly anticipated and addressed by Nyāya philosophers by making subtle distinctions. Due to the limitation of space we cannot discuss the issues in detail but shall mention only one point. Suppose that two archers shoot arrows at the same prey and one arrow hits the prey first and kills it. Had not the first arrow killed the prey, the second arrow would have killed it. Nyāya philosophers have distinguished between the two arrows by calling the first a triggering or immediately productive causal condition (phala-upadhāyaka-kāraņa) and the latter a predisposed or inherently capable causal condition (svarūpa-yogya-kāraņa). In this way, if we dig deep, we may find, though no philosophical position may be above criticism, an adequate defense of causation in the Nyāya.
The Nyāya has argued in part against a skeptic like Carvaka; but the argument could also be extended to a skeptic like Hume or anyone who would not leave the origin of things to causes but to chance. The point may be brought further out as follows. While leaving to chance is logically possible, how does that throw any additional light on the matter? The only explanation that such a skeptic has at his disposal is that anything or everything happens by chance. Why does the flower bloom? Because of chance. Why does water flow downward? Because of chance. Why do people get malaria? Because of chance. Such a skeptic fares no better than a theist who would leave anything and everything indiscriminately to God. Why does the flower bloom? Because of God’s will. Why does water flow downward? Because of God’s will. Why do people get malaria? Because of God’s will. All that such a skeptic thus has done is to replace God’s will with chance. The irony of the whole thing is that while such skeptics dismiss the theists as dogmatists (historically both Hume and Carvaka have rejected theism as dogmatism) and the theists dismiss the skeptics as charlatans, neither may fare any better than the other. While both the appeal to chance and the appeal to God are logically possible, neither has any additional explanatory value when the same old explanation is offered for anything and everything.8 By contrast, assigning different causal conditions for different effects gives us the needed order and control over the virtually endless empirical data: such order and control is the hallmark of rational inquiry. Accordingly, a causal explanation is preferable to an appeal to chance.
A skeptic may not advocate chance in all cases and may accept causation in observed cases and allow chance in some future cases. The skeptic then could appear to avoid the above difficulty. Still, such a skeptic would be hard put to explain why the effect produced by chance in future is of the same kind as the effect now (kāryaikajātīyatvānupapattih). Clearly, that they are of the same kind cannot be due to having the same kind of cause. That would violate the skeptic’s claim that the future effect is due to chance. So the only consistent position for the skeptic is to say that they are of the same kind due to chance. But if being of the same kind is left to chance, how does that throw any additional light (apart from being logically possible) on the question about why they are of the same kind? A little reflection then shows that the old difficulty in the above view crops up in the present view as well. Indeed, if being of the same kind is due to chance, why allow causation at all? That is, what is the difference between cases of chance and cases of causation that makes it necessary to say that while the former is due to chance the latter is not? Further, if someone challenges the claim that the present effect caused by something and the future effect due to chance are of the same kind, the skeptic does not have the resources to answer the challenge. That is, the skeptic cannot produce any cogent reason to show that they are of the same kind. Under the circumstances, the skeptic’s claim that the future effect produced by chance is of the same kind as the present effect is an idle speculative factual claim that may have nothing but being merely logically possible to recommend for it.
The critic may, again, complain that nondeductive reasons have been introduced to defend causation and, thereby, induction. If so, Nyāya philosophers would plead guilty to the charge (işţāpatti). What justifies the introduction of such nondeductive reasons, of course, is that no better explanation is available. This does not amount to begging the question, for the Nyāya claim is based in part on showing that rival explanations fare worse than that provided by the acceptance of causation.
It may be added that Hume’s psychological explanation of causation is inadequate for the purpose. It may be readily granted that the repeated observation of contiguity of two things could produce, as Hume suggests, the habitual expectation of those two being causally connected. But this does not address the prior and more fundamental question raised by the Nyāya. The Nyāya does not ask about what could result from the repeated observation of contiguity, but about what could provide an explanation of the repeated observation of contiguity itself. The question, to repeat, is: why is one thing, say, smoke, never found to come into being without the presence of another thing, say, fire? The best available explanation, the Nyāya claims, is that the presence of one is required for the origin of the other—which points to causation.
A number of considerations may go into determining why one explanation is better than another. One obvious consideration is whether the explanation can be tied to a general truth. For example, the hypothesis (kalpanā, arthāpatti) that a particular person who is fat and does not eat during the day, eats, though unobserved, at night becomes reliable (prāmāņika) if it is true in general that whoever is fat and does not eat during the day eats at night (BPP 552–53). The facts in the situation present an incongruity: the person is fat and yet does not eat during the day. The hypothesis is offered to resolve the incongruity. The general proposition offers an explanation of why the hypothesis holds in the particular case. With it added as a premise the hypothesis may be validly deduced as follows: “Whoever is fat and does not eat during the day eats at night. [The wider general truth lending support to this premise, as it is pointed out, of course is that no one can stay fat without eating. The process of finding more and more general truths will eventually lead to the fundamentals of the system.] Rabi is fat and does not eat during the day. Hence he eats at night.” Thus a hypothesis that is validly derivable by adding an accepted general law as a premise is reliable as opposed to another (such as that Rabi has spiritual power and can get nutrition without eating) that is not.
Another consideration is to apply one or more of the three basic laws of economy (or simplicity: lāghava), viz., economy in cognitive link or order (upasthiti), economy in relationship (sambandha) and economy in constitution (śarīra). The first enjoins the following. Of two necessary antecedents (or two equally matched hypotheses) the one that is more directly related to the effect (or the explanandum) in the cognitive order is more economical. For example, when the smell of a mango changes its color too changes. Thus prior absence of the new smell and prior absence of the new color are both necessary conditions of the changing smell; but only the former and not the latter is accepted as a causal condition of the new smell, for that is more directly related to the effect in the cognitive order. In other words, previous absence of something is more immediately relevant than previous absence of something else as an explanation for the origin of something and should be recognized accordingly.
The second is as follows. Of two necessary antecedents (or equally matched hypotheses) the one that is more directly related to the effect (or the explanandum) is more economical. For example, a wheel is accepted as a causal condition of a pot but not wheelness (the common feature of all wheels) although both are necessary antecedents, for the latter’s relation to the pot is established through the former and, therefore, is more indirect.
The third law implies that of two necessary antecedents (or equally matched hypotheses) the one that is analyzable into fewer constituents is preferable. For example, for a substance to be perceptible (in the Nyāya view) it should have intermediate magnitude (that is neither the biggest nor the smallest possible magnitude) and should also be made of many substances. Although both are necessary conditions, only the former is accepted as a causal condition of perception of a substance on the ground of economy of constitution. These principles of simplicity do not imply that in the Nyāya view the world is simple. On the contrary, the favorite ontology (that the Nyāya has adapted from the Vaiśeşika) is highly structured and elaborately worked out to the minutest detail. But it does mean that a theory or an explanation that is unnecessarily complex is inferior, other things being equal, to another that is not so. This follows from the very nature of a theory or explanation one purpose of which is to give a clearer understanding. If no cap is put on avoidable complexities, the explanation could be indefinitely long and hinder rather than contribute to a clearer understanding.
Besides the basic principles of economy there are also numerous auxiliary ramifications. One ramification is that, other things being equal, an explanation that conflicts with fewer observations (or accepted truths) is preferable to one that conflicts with more. Thus in SL (58) a theory conflicting (apalapa) with two experiences (anubhava) is found to be at a disadvantage compared to another theory conflicting with only one experience. The other side of this is that, other things being equal, an explanation that applies to a greater number of relevant situations is preferable to one that applies to a fewer number of relevant situations (BPP 36). Another ramification is that an explanation that is equally matched (tulyabala) by a rival explanation is not reliable (NS 1.2.7). Yet another ramification is that an explanation that relies on mere (random) similarity (sādharmyamātra) or mere (random) dissimilarity (vaidharmyamātra) is not reliable (NS 5.1.2). Still another ramification is that an explanation that leads to the addition of something unfavorable (utkarşasama) or the deletion of something favorable (apakarşasama) is not reliable (NS 5.1.4). Indeed, inference to the best explanation (ananyagati, anyathānupapatti, prayojakakalpanā) is a well developed and widely used technique in Nyāya logic.
Returning to causation, a skeptic may not dispute that there is causation in particular cases. [For Hume’s “rules by which to judge of causes and effects,” see Treatise, I, iii, 15; Selby-Bigge 173–74.] For example, when I light up fire and see something being burnt and smoke coming out, I know (after removing superfluous factors, if appropriate) that smoke is caused by (an aggregate including) fire here and the skeptic may accept that. [If the latter disagrees, it suffices to point out that general skepticism about particular observations is not the subject of discussion here and is out of place. Our concern here, as already said, is to see if induction is justified assuming that particular observations are true or reliable.] The skeptic, of course, claims that this and other such additional observations do not give us a rational ground for thinking that smoke will be caused by fire in a new case or that smoke is caused by fire in all cases. Still he does not dispute that smoke is an effect, is caused by something and is caused by fire in the observed cases. But then in order to be able to question that smoke will be caused by fire in new cases, the critic must court the doctrine of plurality of causes. Thus one possible skeptical challenge to the above solution comes from the doctrine of plurality of causes. This is the doctrine that the same effect may be produced by more than one sum total of causal conditions. For example, death may be caused by drowning, taking poison, starvation and so on. Fire may be fueled by grass, wood, coal and so on. Could it then be that although smoke is produced by an aggregate that includes fire, it may also be produced by an aggregate that excludes fire but includes something else?
The doctrine of plurality of causes, however, has been examined and rejected by Nyāya philosophers like Udayana. Briefly stated the Nyāya position is that the so-called cases of plurality of causes boil down upon careful scrutiny to one of two situations. The seemingly different causal aggregates may be found to have common traits so that they can all be said to be of the same kind. For example, all cases of death from drowning, poisoning and so forth. may be found to involve the common factor of stoppage of flow of oxygen to the brain and this may justify the conclusion that all cases of death are caused by the same kind of cause. Alternatively, the seemingly same effects may be found, when closely examined, to have significantly different features so that they can be said to be of different kinds. For example, fires fueled by different kinds of materials burn differently. Some fires burn for a short time and some for long. Some produce more heat and some less. Some produce more light and some less. All these may justify the conclusion that these are different kinds of fire caused by different kinds of causes. So, either the seemingly different effects are found upon examination to have a common nature and then the seemingly different causal aggregates are also found to have a common nature or the seemingly same effects are found to have different natures and then they are also found to be caused by things of different natures. Thus, the admission of plurality of causes can be avoided either by showing that the effects are of different kinds (kārya-vaijātya) or by showing that what appear to be different kinds of causes have a common nature (kāraņaikajātīyatva).9 If the above reasoning is sound, the skeptical objection from possibility of plurality of causes fails.
Another possible skeptical challenge comes from the doctrine of accidentalism (āksmikatāvadā) that things either routinely or at least sometimes happen by chance. If this is accepted, it cannot be claimed that, say, smoke is invariably caused by an aggregate that includes fire, for at least accidentally it may be caused by an aggregate that excludes fire.
The doctrine of accidentalism too has been examined and rejected by Nyāya philosophers like Udayana as already noted. In brief, the Nyāya position is that there are no accidents in nature. The so-called cases of accident show upon examination uniform causal connections with common effects and common causes. For example, one may be said to have died accidentally from drowning. But then the usual causal connections were surely not violated in such a case. That is, one who died did get into water, did not stay afloat, got submerged in the water, could not breathe after being submerged and consequently died. It may be thought perhaps that the one who died from drowning still got into water accidentally, such as that he/she may have been leaning on the railings of the deck of a ship, the railings suddenly gave away and he/she fell into the water, did not know how to swim and drowned. But even then no causal laws were breached. Perhaps what happened was that the railings were in disrepair and rusted and the man’s weight was too much for those railings to bear. Indeed, the search and discovery of causal connections where such connections are not apparent is one of the foundations of scientific inquiry.
But where is the evidence, the skeptic may persist, for this fundamental principle that causal laws are universal and uniform? If this evidence is merely from observation of every known case of an apparently accidental happening as being eventually tied to accepted causes, it presupposes the rationality of induction and, in the present context of justification of induction, is circular. So Udayana gives a different answer: the evidence comes from the occasional (kādācitka) nature of effects as said before. Occasionality is best explained by admission of limits that point to causation and disfavor accidentalism. The skeptic’s challenge from the standpoint of accidentalism seems then to fail.
Further, in order to argue for the plurality of causes the skeptic has to show that the different cases of death are all of the same kind (and not merely similar). Now, the notion of being of the same kind, the Nyāya philosophers argue, cannot be ultimately defended without the admission of universals (jāti). This is, again, a large and difficult topic in itself and cannot be fully discussed here. Still we note here that Nyāya universals are not transcendent ideal exemplars like the Platonic forms, but are (sometimes observable) common characters inherent in the particulars and that both particulars and universals are real [a loose Western analogue is David Armstrong’s theory of universals]. A key Nyāya argument for this is as follows. We speak of natural classes like lions, tigers and so on. We put all lions into the same class for a reason. The reason is that individual lions are found to share a family of similar features. Similar features are particular features. But particularity itself is not the basis of similarity, for then any two particular things could be similar. So some other basis for similarity is needed. If that basis is something particular, the same question about what makes that similar comes back and continues infinitely as long as only something particular is offered as an answer. Since infinite regress can be avoided only by admitting nonparticular identical features shared by different particulars, these should be admitted as real. Such identities called universals are needed not only for class inclusion but also for class exclusion. We not only put all lions, for example, into the same class, but we also exclude all tigers from that class. The reason for that is that tigers do not share the same features. Since once again similarity without identity will generate infinite regress, universals are needed to make sense of class exclusion as well.10 Though Nyāya philosophers accept universals, they do not subscribe to either the older Platonic essentialism or the recent new essentialism of Kripke, Putnam and so on. (For a brief account of new essentialism and criticism see S. Mumford, “Kinds, Essences, Powers,” Ratio (new series) XVIII 4, 2005, 420–36.) Nevertheless, the Nyāya supports natural kinds as corollaries of universals.
But if universals are admitted, why not admit causation and induction as well? For then the unobserved cases could be viewed as being of the same kind as the observed cases and causation and induction upheld accordingly.
The appeal to causation may invite the old and familiar charge of circularity. Since pervasion (induction) presupposes causation and causation presupposes pervasion (induction), no real progress, the critic might say, has been made. Hume, in particular, argues as follows. The reason for moving from observed to unobserved cases would have to rely on the principle of uniformity of nature that unobserved instances resemble observed ones. But this principle is not necessarily true, for its denial is not self-contradictory. It cannot also be shown to be probable, for any such attempt would have to rest on the very presumption of the principle of uniformity—which would be circular.
All this makes sense only if it is assumed that nothing falling short of a valid deduction constitutes a reason or a rational exercise, an assumption that the Nyāya does not buy. In particular, the principle of uniformity must be presumed as a premise while showing that induction is reliable only if such showing must be a deductively valid argument. But clearly, the Nyāya has not tried any such thing. Hence they are not obligated to presume or add as a premise the law of uniformity while arguing for rationality of induction. On the contrary, the Nyāya has offered counterfactual reasoning to argue for the reliability of induction. The principle of uniformity does not appear as a premise in that reasoning. Further, the Nyāya has argued at length for the law of causality and not merely taken it for granted as we have seen. If these arguments make sense, the charge of circularity is not in order.
The charge of circularity, in the highly developed Nyāya view, is justified only when the conclusion is (or can by analysis be shown to be) identical with a premise (or a part of a premise) brought in support of the conclusion or the truth or reliability of the premise is indispensably dependent (sāpekşa) on that of the conclusion. The Nyāya position is not circular in this sense. The crucial premises (the remainder being formal operations) in the counterfactual reasoning discussed earlier are (1) that if smoke were produced neither by an aggregate that includes fire nor by an aggregate that excludes fire, smoke would not be produced and (2) that smoke is produced. The conclusion is that smoke is produced by an aggregate that includes fire. Thus clearly the conclusion is never identical with the premises. So the only remaining question is whether the truth or reliability of the premises is materially dependent on that of the conclusion. We have here two premises. First, take the premise that smoke is produced. Is it indispensably dependent on the conclusion that smoke is produced by an aggregate that includes fire? No, for the evidence for that premise comes directly from observation. Smoke is observed to come into being where it was nonexistent before. This suffices to show that the premise is reliable. [Being produced is analyzed by the Nyāya to mean being the negatum of a prior absence (prāgabhāvapratiyogitva)—that is, coming into being after being nonexistent before.]
Now take the other premise that is a counterfactual proposition both the antecedent and the consequent of which are known to be false. Since this premise is the willful articulation of a known counterfactual situation, it is not true or reliable for the Nyāya.11 Given the Nyāya theory of truth or reliability, the claim that truth or reliability of that premise is not indispensably dependent on that of the conclusion is vacuously true, for the premise is not true or reliable in the Nyāya view (though the premise is a part of an argument that contributes (anugrūhaka) to the truth or reliability of induction).
A theory of truth or reliability cannot be discussed in a short space and we cannot properly discuss the Nyāya theory of truth/reliability here. Still, it is clear that the epistemic structure of the premise is significantly different from that of the conclusion. The latter is an indicative proposition. The former is a conditional with a false antecedent and a false consequent. It will take an adequate theory of counterfactual conditional and a substantial argument to show that the truth or reliability of such a conditional depends on such an indicative proposition. Neither Carvaka nor Hume has provided that.
Further, many would agree that the acceptability of the counterfactual premise is not dependent on the conclusion that smoke is produced by an aggregate that includes fire. This may be seen if we realize that the premise is not critically about smoke and fire. The main point of the premise is that if something were not produced by an aggregate that includes a particular kind of thing or by an aggregate that excludes that particular kind of thing, that something would not be produced. No reference to smoke being produced specifically by an aggregate that includes fire is materially relevant for this. The epistemic task of the premise is accomplished by relying on logical laws like the law of excluded middle and the obvious truth that if something is not produced by any aggregate, it is not produced. It thus appears that neither premise of the counterfactual reasoning is indispensably dependent on the conclusion and that the reasoning is not circular.
A skeptic may point out that the Nyāya case for induction involves at least the induction that the best available explanation is reliable. Accordingly, the Nyāya is guilty of what has been called rule-circularity.12 When one relies on the same rule for which one is arguing, there is rule-circularity. The skeptic may add that whatever reasoning is offered in support of induction would inevitably involve some induction and be invariably circular.13
But the assumption behind this objection is that if the same rule is involved in the justification of a given rule, the reasoning is circular. This assumption is questionable. Suppose that one has to argue for the rationality of deduction. One has no choice but to rely in part on deduction to do so. Similarly, if a skeptic denies that there are any sources of knowing, there is no choice but to rely on some sources of knowing to refute the skeptic. So the above kind of circularity, if recognized as a defect, would threaten the status of not only induction but that of all knowing. If accordingly the assumption is rejected to allow for the possibility of knowledge, the objection would fail.
Sometimes the point of the distinction between rule-circularity and premise-circularity is misunderstood. The point is that just as one has no choice but to use memory to check trustworthiness of memory in general or just as one has no choice but to use deduction to check trustworthiness of deduction in general, so also one has no choice but to use induction to check trustworthiness of induction in general. Howson has argued in rejecting that rule-circularity is not a flaw that there is nothing circular in testing another person’s memory with my own or somebody else’s memory or testing the soundness of a particular deductive rule like modus ponens that does not involve that particular rule itself (HP 25, 28). But this is based on confusion. Those who hold that rule-circularity is not a flaw are not arguing from testing one particular memory with another particular memory or testing one particular deductive rule with another deductive rule. Rather, they are arguing from the general faculty of memory being tested or the general method of deduction being tested. And then rule-circularity is unavoidable. (For more discussion of epistemic circularity and rule-circularity, one may see B. Reed, “Epistemic Circularity Squared? Skepticism about Common Sense,” Philosophy and Phenomenological Research, 2007, 186–97; M. Bergman, “Epistemic Circularity and Common Sense: A Reply to Reed,” Philosophy and Phenomenological Research, 2007; and N. Tennant, “Rule-Circularity and the Justification of Deduction,” The Philosophical Quarterly, 55/221, 2005.) Further, the skeptical claim that no argument can show the reliability of induction without presupposing that reliability itself involves induction and cannot be sustained without presupposing its reliability. Thus if rule-circularity is a flaw, the skeptical objection is flawed too. In other words, if the reliability of induction cannot be challenged without presupposing that reliability, the challenge is futile. In particular, a skeptic can ill afford to disown inference to the best explanation. The skeptical position should be argued for and the explanation offered by the skeptic should be better than that of the opponent and, therefore, qualify as the best explanation.
The skeptic, again, may revive the complaint that the Nyāya arguments brought in defense of causality and induction are not deductively valid. But so what? The community of scholars does not grant the skeptic the exclusive right to decide what counts as a reason (hetu, gamaka, apadeśa, liñga, sādhaka, upapatti, etc.) and does not endorse that all rational performances are deductively valid. This applies to the world of Sanskrit scholarship as well as the world of contemporary scholarship. In the Nyāya view, it is the community of scholars that is the custodian of the world of learning (vidyā, śāstra) and it is that community that preserves and revises it. As long as the larger scholarly community does not enjoin the narrow-minded view that all reasons must be deductively valid, the Nyāya is under no obligation to produce only deductively valid arguments in defense of causality and induction. (This is not to suggest that Nyāya logicians have neglected deduction, for Nyāya logic includes an advanced formal logic.) The skeptic may retort: why then embark on the project of justifying induction in the first place? But self-examination, self-defense in the face of opponents’ objections and refutation of opponents’ views are all parts of the ongoing scholarly activities. This is precisely what the Nyāya is doing while trying to defend induction.
While on the charge of circularity, it is useful to add that Gadadhara has carefully distinguished between pervasion involved in causation and pervasion required for inference. The former is: not being the negatum of an absence belonging to the locus of the effect in the moment immediately preceding its origin. To explain: if something is the negatum of an absence in the locus of the effect immediately before its origin, it is absent where and when the effect is produced. Since the effect has come into being without it, it cannot be regarded as a cause that is a necessary condition. Hence a cause must be different from that—that is, it must not be the negatum of such an absence. For example, fire is not the negatum of an absence belonging to the locus of smoke immediately before its origin. On the other hand, pervasion for inference is the following: not being the negatum of an absence that is co-located with the probans and that is not co-located with the negatum. Thus the two pervasions are different. [The differences are substantial, as a detailed study of inferential pervasion that will be lengthy, difficult and that cannot be undertaken here, will inevitably show. One obvious difference, of course, is that causal pervasion, unlike inferential pervasion, includes a reference to the time of origin.] Not that causal pervasion will be brought in as a premise while arguing deductively for inferential pervasion and vice versa. Nyāya logicians are not up to that. Still, it would be a serious mistake to confuse one for the other and Gadadhara is putting that on notice. In the words of Gadadhara:
It cannot be said: since effect-hood involves pervasion, cognition of effect-hood is itself cognition of [inferential] pervasion…. For pervasion involved in effecthood is different from pervasion leading to inference: while the former is: “not being the negatum of absence belonging to the locus of the effect in the moment immediately preceding the effect,” the latter is: “not being the negatum of absence that is co-located with the probans and that is not co-located with the negatum.” (GD 681)
The Nyāya response to the charge of circularity, it may be noted, is quite different from that of Braithwaite mentioned earlier. The former is not attempting to show that the inductive justification of induction meets the criteria of subjective validity or the criteria of subjective and objective validity— which was found to be unsatisfactory. The Nyāya response that may have more promise is fourfold: (1) It is questionable if rule-circularity is a flaw; at least rule-circularity cannot be construed to mean that the justification of induction must be on exclusively noninductive grounds. [This is not, as already said, any worse than having to fall back on methods of knowing in order to refute some skeptics who might claim that there are no methods of knowing or, for the matter of that, having to rely (partly) on deduction in order to justify deduction.] (2) The principle of uniformity or causality or any other such principle is not needed as a premise in arguing for induction, because the argument concerned is not required to be deductively valid. (3) Causal pervasion and inferential pervasion are substantially different. (4) The conclusion of the counterfactual reasoning brought in defense of induction is not identical with the premises or a part of them nor are the premises indispensably dependent on the conclusion.
We now move on to the argument from belief-behavior conflict. This may remind one of an argument of G. E. Moore to defend what he called “common sense propositions.” Moore argued that a philosopher who denies or doubts such common sense propositions as that material objects exist or that time is real inevitably engages in behavior that conflicts with such claims—which refutes the denial or the doubt. The interesting point in Gangesa’s case is that a similar argument has been brought in support of causal propositions like fire is a causal condition of smoke and, thereby, in support of induction.
The Carvaka skeptic claimed that one engages in practical activities always on the assumption of probable causes. Hume also said that although the skeptical doubt is beyond resolution, this should not interfere with practical activities; all that one needs to do is to switch from the theoretical to the practical standpoint. Hume held further that although skeptical considerations lead to philosophical melancholy and delirium, it does not persist, for lively impressions or other thoughts or feelings divert us to other things. Additionally, although reason alone cannot overcome skepticism, reason supplemented by our natural desires, inclinations, instincts and habits allow us to recommit ourselves to rational activities including induction that follow our propensities. Thus Hume offers a psychological explanation of how inductive inferences are caused—an exercise in cognitive psychology—that is quite different from the justification of induction that is an epistemological exercise.14
Gangesa would agree with Hume in part on the psychological thesis. It is not the intention of Gangesa to deny that one is motivated to action by probable opinion. It is easy to think of situations (e.g., a scientist trying out a tentative hypothesis or a detective pursuing not so clear a clue) where this actually happens. Nyāya philosophers also acknowledge that skeptical doubt as a psychological state is routinely replaced by other states without needing any argument and also recognize the roles played by habits and inclinations. But what Gangesa definitely wishes to reject, if such is offered as a thesis in cognitive psychology, is that one is always motivated to action by probable opinion. Clearly, any Hume-like critic of induction who wishes to argue for such a thesis would have to rely on induction and thus forfeit his case. So a skeptic is not in a position to offer proper evidence for the claim. But further there is counterevidence (bādhaka) from uniform and unwavering action. That is, one does not remain doubtful about something being a cause of something else and still continue uniformly and unwaveringly (nişkampa-pravŗtti) to procure the former in order to produce the latter. To doubt that something is a cause of something else amounts to endorsing the possibility or sometimes even the probability that the latter is produced without the former. If one were truly doubtful about something being a cause, one would try other alternatives, as the scientist or the detective in our examples would. Hence the very action of procuring something regularly and unwaveringly to produce something else reliably (though fallibly) shows the absence of any actual doubt. This then is a thesis in cognitive psychology that Gangesa offers: uniform and unwavering voluntary action is prompted by doubt-free belief.
The issue cannot be avoided by advocating a cleavage between the theoretical and the practical points of view. The Nyāya does not dispute that it may be certain for all practical purposes that fire is a causal condition of smoke and still be theoretically possible that this is not so. This is implied in acknowledging the fallibility of induction. But such fallibility does not make induction doubtful. [We assume that particular observations like there is a cat on the mat are often reliable and doubt-free in spite of being fallible. Some skeptics do question this. But, as said, we are not here dealing with such a skeptic but only with someone who denies the justifiability of induction without denying the reliability of particular observations.]
But further, Gangesa is also building an epistemological argument from belief-behavior conflict for reliability of induction. That is, one’s action provides the epistemic ground for rejecting the actual presence of doubt. [This does not imply that my believing x and acting on it makes x true; rather, the epistemic ground is provided, as explained below, by the success of the effort.] For example, the action of a honey gatherer to light fire to drive away bees from the honeycomb, say, is prompted by the cognition or belief that fire is a cause of smoke. Since smoke is the intended fruit (phala) of the action, when smoke is produced, the action becomes successful. The success of the effort (pravŗtti-sāmarthya) shows (i.e., gives the epistemic ground to the effect) that the said cognition or belief is reliable (prāmāņika) and dislodges the claim that it is doubtful.15 [Needless to say, the Nyāya does not hold that certainty or strong belief is knowledge. For example, one may be absolutely certain about seeing a snake in front while the thing in fact is a rope. Rather, the point is that success of the effort is a reliable (though fallible) sign for inferring the reliability of the (fallible) belief prompting the effort.] The issue here is not merely psychological, but epistemic. Gangesa is not claiming that it is psychologically impossible to harbor the doubt under the above circumstances, for it is not. Nobody can force the skeptic to free his mind of unfounded doubts. The Nyāya recognizes that desire (icchā) is a sufficient stimulant (uttejaka) to enable someone to hold on to even a glaring contradiction. Still, the point is that since there is epistemic ground from one’s own action to show that induction is reliable, the skeptical doubt is out of place. In other words, uniform and unwavering action prompted by a belief is a reliable epistemic ground for reliability of that belief. This is why Gangesa says that such action is an obstruction to the doubt. It is an obstruction partly in the sense that it provides reliable evidence for absence of the doubt (that is quite different from the routine absence of the doubt when the doubt is replaced by some other psychological state). Unless the skeptic is able to refute this evidence the skeptical claim of presence of the doubt is hollow (more on this below). When the doubt is eliminated (that does not involve that doubt is logically impossible) on reliable epistemic grounds, no additional reasoning is called for. Thus the skeptic’s charge that there must be either an infinite regress or circularity in the justificatory process is untenable.
Gangesa took over this argument from Udayana. In a famous verse the latter said (NK, chapter 3, verse 7): if there is doubt, there is inference (for the fear of deviation with reference to a future time or place has to make use of inference); if there is no doubt, there is inference; doubt is removed by subjunctive reasoning; conflict is the limit of doubt. Udayana was replying to the charge of infinite regress. Could the doubt be revived after it is removed by subjunctive reasoning? No, said Udayana, as long as one acts unwaveringly. Unwavering action is in conflict with doubt and sets the limit to doubt—that is, does not allow one to have the right to doubt.
Sriharsa (twelfth century), an Advaitin skeptic, made some marginal changes in the wording of the verse and came up with a crushing rejoinder: if there is conflict, there is doubt; if there is no conflict, there is doubt all the more; how can then conflict set the limit to doubt and how can subjunctive argument set the limit of doubt (KKK 364)? His point is that the claim that conflict is the limit of doubt itself incorporates a generalization, viz., whenever there is conflict there is no doubt. Now he is not pressing for the irrationality of this or that induction, but of any and every induction. Since the argument from conflict is itself relying on an induction, albeit a different one, the skeptical doubt will inevitably haunt it and keep the regress alive, for where is the (noninductive) reason to show that this induction will hold in a new case?
Gangesa is replying directly to this rejoinder. He thinks that Sriharsa has misunderstood Udayana’s argument. The latter is not first generalizing empirically that whenever there is conflict, there is absence of doubt and then arguing deductively after adding that as a premise that since there is conflict, doubt is gone. Rather, the point is that the unwavering action obstructs doubt.
Several things should be considered while interpreting this argument. First, in Gangesa’s view, introspections (alone) in a limited sense are self-certifying. It never happens, he says, that I am not aware of anything and still believe that I am aware of something, nor that while I am aware of a pot, say, I introspect that I am aware of a piece of cloth. Even when I misperceive a shell as silver, I introspect unfailingly that I am aware of silver (TC 284–85).
A proper discussion of this view that introspection is in some sense incorrigible will take a lot of space and must be left out. Still, we note that Gangesa’s view is similar to the view of Descartes that although we can call into question what we perceive by means of our senses, we cannot call into question that the ideas or thoughts of whatever is perceived hover before our minds.16 For example, I may be entirely wrong in claiming that what I see is a horse, but I cannot be wrong about claiming that I have the idea of a horse in my mind. So Gangesa rules out that when we act unwaveringly, we still have a lingering doubt in our minds. Of course we may be doubtful about fire being a cause of smoke and may be hesitant while acting upon it. But if we are certain about it, introspect it so and act upon it unwaveringly, we are not doubtful about it any longer, for the introspection that we are certain about it is reliable. In other words, if I am certain about something and my introspection says so, it is reasonable to accept that. This is an item of personal experience (anubhava) that (though fallible) is on its own ground reliable. If this is rejected, the price to pay will be much higher than merely rejecting induction.
Second, apart from the evidence from introspection, unwavering action by itself reliably proves absence of doubt. Udayana has distinguished between contradiction in language (svavacanavyāghāta) and contradiction in action (svakriyāvyāghāta).17 The former is illustrated by “son of a barren woman” (bandhyā-suta) [similar to “married bachelor,” familiar in the West]. Since barrenness stands for childlessness, this expression is a patent contradiction in terms. The latter is illustrated by someone actually saying “I am dumb.” This sentence is not self-contradictory and there will be no conflict if the person merely writes it down. But the very act of articulating it aloud brings out the contradiction and falsifies it. [A solution to the age-old liar paradox may be worked out along these lines.] Just as the very fact of someone speaking out falsifies the claim of his being dumb that involves the lack of the ability to speak, so also the very fact of someone acting unwaveringly falsifies the claim of his being doubtful that involves indecision and inability to act unwaveringly. Needless to say, one may pretend to act unwaveringly and may not get caught and, whether someone is pretending or not, we may be wrong in judging that someone is acting unwaveringly. Still, it remains true that unwavering action is a reliable ground for lack of doubt.
In other words, Gangesa makes the general claims that whenever there is unwavering action, there is lack of doubt and that success of effort is a reliable sign for inferring reliability of cognition or belief that prompts effort. With regard to inference of reliability of prompting cognition from success of effort, Vacaspati Misra (ninth century CE, a great philosopher and author of masterpieces on the Nyāya, Advaita and Sāmkhya-Yoga) held that while an average cognition is not self-certifying (svatahpramāņa), such an inference is.18 While other Nyāya philosophers do not go as far as that, they recognize, as we have seen, a class of cognitions that are discernible as true or reliable by mental reflection. This is due to the special logical, semantic and epistemic relations holding within the contentness (vişayatā)—that is, among the different contents (vişaya) of the cognition. For example, that no barren women have sons is discernible as true by mental reflection although that no rabbits have horns is not. Both are true general propositions; but an important difference between the two is that while the denial of the former involves a contradiction in language (svavacanavirodha), the denial of the latter does not.
Now, a part of what Gangesa is implying in rejecting the charge of infinite regress and circularity is that the skeptic is overlooking the difference in the epistemic status of the pervasions involved. When we deal with pervasions like “whenever the effort is successful, the cognition prompting it is true or reliable,” mental reflection suffices for the purpose.19 [This in no way denies that our expectations can seem to be satisfied even if our perceptions are false. For example, someone mistaking a rope for a snake could succeed (in his mistake) in avoiding the snake by running away from it. But even in such a case there is a factual core (albeit mistakenly interpreted). For the said person does as a matter of fact succeed through his effort in establishing some distance between himself and the thing in front. The crucial question, for the Nyāya, is whether an external sense organ plays a causal role for the false perception. If so, there will have to be a factual base even in our worst hallucinations. However, if no external sense organ plays a causal role, as it happens in a delusion, the experience cannot be regarded as perceptual (pratyakşa). The underlying issues are once again deep and difficult and a proper discussion is beyond the scope of this work. But it may be noted that the Nyāya has worked out a highly developed and complex epistemology to back up its viewpoint. Some skeptics will no doubt insist that we can never get beyond appearances. But how does the skeptic know that? Doesn’t he have to use induction and thus forfeit his case?] Since we are here no longer dependent on external sense organs, as we are in the smoke-fire case, the skeptical doubt arising from unobserved cases is not relevant. Once again, if the skeptic denies this and disowns the role played by mental reflection, the price to pay will be much higher than merely degrading induction. Consider, for example, that nothing is both blue and not-blue. Are not we sure through mental reflection that this is true? And, if we cannot be sure about this, how can we be sure about anything? So, unless a more sweeping skepticism is adopted, the role played by mental reflection should be admitted. Then it can be seen, through mental reflection, that the skeptical doubt (though logically possible) is not appropriate for pervasions like “whenever the effort is successful, the cognition prompting it is true or reliable.”
It may be noted that the Nyāya does not subscribe to the dichotomy between analytic truths and synthetic truths (if the issue is framed in modern terms) that has been popular in modern philosophy since the time of Kant. From the Nyāya viewpoint there is continuity between what in modern philosophy are called analytic truths and synthetic truths and even logical truths have a minimal factual content, for they too are true of the world. However, the Nyāya recognizes the distinction between beliefs the reliability of which can be discerned by mental reflection alone and beliefs that are not so. What are called analytic truths in modern philosophy would come under beliefs the reliability of which can be discerned by mental reflection alone in the Nyāya scheme. Once some data have been received from the external senses the inner sense can analyze and discern the connections at the level of contentness. In this way the Nyāya recognizes that given some premises a conclusion may follow logically or analytically from them and thus acknowledges the role of deduction. Similarly, once we have concepts like that of a bachelor, we can also have sentences like “a bachelor is married,” the falsity of which can be discerned by mental reflection alone as a case of contradiction in language (svavacana-virodha). It is in this sense that the Nyāya speaks of beliefs that are reliable (or unreliable) by mental reflection alone.
This role for mental reflection, it should be noted, is directed primarily toward concepts that are complex (sakhanda) where it can be shown by analysis what is included in or excluded from the contentness. [The mind (or more accurately, the inner sense: manas) also has the function of making possible direct awareness of our internal states like pleasure and pain.] Thus it does not amount to endorsing an intuitive reason in the rationalist sense and does not compromise the traditional Nyāya perspective that perception is the leader among the sources of knowledge. [Gotama put perception first in his list of sources of knowledge and added that inference is preceded by perception (NS 1.1.4, 1.1.5). Nyāya philosophers have interpreted this to imply that while reasoning can expose errors in sensing and also allow us to extend our knowledge to imperceptibles, reasoning alone cannot override the testimony of the senses. For example, a reasoning to prove that fire is cold will be set aside because it is contrary to the perception that fire is hot, if for no other reason.]
While there is general (but not universal) support among Nyāya philosophers in advocating the counterfactual argument and the argument from belief-behavior conflict, some (including Gangesa) additionally hold a view (opposed by Raghunatha and others) involving what may be seen as an enhanced role for the external senses. This consists in admitting an extraordinary kind of external perception called sāmānyalakşaņa, with common characters (sāmānya) providing the sensory connection (pratyāsatti), as a source of the awareness of pervasion. Perception, in the Nyāya view, cannot take place without sensory connection. That sensory connection is a necessary condition for perception follows from considering such as the following: while we can see things in front of a wall, we cannot see things behind that wall. But if sensory connection is needed for perception, how can there be sensory connection with all the particulars of a kind—past, present and future—that are covered in the awareness of pervasion? Under the circumstances as an answer to this question, it is proposed that common characters, among which some are admitted to be eternal on independent grounds and to be perceptible (if belonging to perceptible particulars), provide the sensory connection.
To explain: When an ordinary (laukika) perception of a particular takes place, the common character inherent in that particular may also be perceived. After that there may take place the extraordinary (alaukika) perception of all the particulars sharing that character, it being the qualifier (viśeşaņa) of the particular with which there is ordinary sensory connection. The individual features of the particulars do not become the contents of such extraordinary perception. Further, such awareness can take place only when there is an ordinary sensory connection with a particular having that character.20
Thus the induction that all smokes are caused by fire is, in this view, a case of external perception and not inferential at all. The import of the induction is that smoke as qualified by the common character smokeness is caused (in part) by fire as qualified by the common character fireness. Smokeness and fireness are, in this case, the limitors (avacchedaka) or specifiers of respectively the characteristic of being an effect and the characteristic of being a causal condition. The so-called inductive leap may take place through an extraordinary perception when there is an ordinary sensory connection with a particular smoke and a particular fire. Thus:
Awareness of pervasion that comprises all smokes, etc., takes place through the sensory connection called sāmānyalakşaņa. Since otherwise the smoke in the hill is not known to be pervaded [by fire], how is there inference [of fire] from that serving as the ground? (TC 230)
In a second version, not the common character itself, but its cognition, is said to be the sensory connection. This is because both eternal and noneternal entities serve as common characters and may pave the way for this kind of extraordinary perception of all their substrates. But a non-eternal character may cease to exist. If the common character itself is held to be the sensory connection, no such extraordinary perception can take place when that character is nonexistent. This contingency is avoided by holding that not the character but its cognition supplies the sensory connection. Obviously, cognition of the character, say in the form of a remembrance, may be there when the character is nonexistent. Thus:
If by sāmānyalakşaņa is meant what is of the nature of a common character, the character itself is the sensory connection; but if what is meant is that of which the common character is the specifier, cognition of that [is the sensory connection]. (GD 773)
It may be noted that in the first version, too, the cognition of the common character is required, for the character must be featured as the qualifier of that with which there is an ordinary sensory connection. It should also be noted that a character may be simple or complex. If the character happens to be complex, the simple character at the bottom should be held to provide the sensory connection through an appropriate indirect relationship. Thus: “Common characters are of two kinds, simple or complex. If the character is complex, it is still a simple character that functions as the sensory connection through an indirect relation” (GD 779).
It is also held that the skeptical doubt that smoke, etc., may deviate from fire, etc., too is a judgment of external perception. This kind of extraordinary perception should be admitted, it is argued, as the source of the skeptical doubt as well. For the doubt is not about cases observed in the ordinary way, but about cases unobserved in the ordinary way. The only way in which these could be (perceptually) presented before the mind is through such extraordinary sensory connection provided by the common characters (TC 235–36).
Many Nyāya philosophers do not regard the skeptical doubt or the inductive generalization as cases of external perception and are not persuaded that this kind of extraordinary perception should be admitted. However, there may be no overriding difficulties in admitting this kind of extraordinary perception. Since the internal criteria for settling what should count as external perception [apart from the innocuous truism that an external sense organ should function as an instrument (karaņa) for such perception] are far from noncontroversial, this view remains an interesting option in epistemology. Needless to say, the view is not an invitation to any kind of mysticism or esotericism, but is promoted to address specific epistemic concerns. In fact, one motivation for it is not to allow any enhanced role for mental reflection, to promote instead a dominant role for the external senses and to avoid questions that could in the long run spell trouble for an empiricist and common sense realistic point of view. [This is not going to impress the skeptic, but may impress those leaning toward empiricism and common sense realism.]
To sum up, the skeptical doubt about induction involves doubting such beliefs as that fire burns, food nourishes or language is a tool of communication with others. The said doubt is untenable because it inevitably leads to belief-behavior contradiction that is an instance of contradiction in action as distinguished from contradiction in language. Thus if one doubts that food nourishes and thinks that food may not be indispensable for survival, one would not continue to eat uniformly, day after day and without any hesitation in order to be able to survive. Similarly, if one is doubtful about language being a means of communication with others, one would not continue uniformly to use language for that purpose. For, if one were doubtful about the outcome of an action, one would explore other alternatives and would not pursue that course of action uniformly and unwaveringly. Thus, an important thesis in cognitive psychology that emerges is that uniform and unwavering action is caused by doubt-free belief. Such action is also an obstruction to doubt in an epistemic sense. The fact that one acts uniformly and unwaveringly to get certain results proves in a reliable (though fallible) way absence of any actual doubt about the success of the action. Since the skeptical claim about doubt is obstructed by a reliable argument, that claim is not tenable until the given argument is refuted. Thus belief-behavior conflict yields both an important thesis in cognitive psychology stated above and a powerful objection to the skeptical claim about doubt. We also learn from introspection that when we act unwaveringly, we are not subject to doubt. Such an introspection testifying to absence of doubt is reliable. Further, whenever an action prompted by an anticipation of what is to be achieved is successful, cognition prompting the action is reliable. No skeptical doubt arising from the possibility of (externally) unobserved cases is relevant here for reliability of this general claim is discernible by mental reflection alone. Since in this way absence of doubt may reliably (though fallibly) be asserted in cases of belief-behavior conflict, the skeptical charge of infinite regress or circularity in justificatory arguments is unfounded.
Moreover, factual generalizations are vindicated by the counterfactual argument exploring the implications that show the affinity and continuity between deduction and induction. For example, “fire burns” implies that if I were to put something I treasure into fire, it would be burned. Since I am certain about this, I ensure that this does not happen. Past observations do not directly provide the rational ground for making general claims about the future, but they do so indirectly by way of the counterfactual argument. The mind is dependent on the eyes, the ears and so on, for information about the external world concerning what was, what is or what will be. But what would have been is different from these and can be legitimately explored by the mind itself on the basis of what has been learnt from experience. A celebrated counterfactual in defense of the stock induction that all smoky things are fiery is that if smoke were produced neither by an aggregate that includes fire nor by an aggregate that excludes fire, smoke would not have been produced. We observe that smoke is produced. So the consequent is false. This logically implies falsity of the antecedent and given OC we should hold that smoke is produced by an aggregate that includes fire. This favors the induction that all smoky things are fiery.
Finally, neither the doctrine of causal power nor the doctrine that the cause potentially contains the effect is acceptable. But the observed fact that some kinds of things come into being only on the occasion when some other kinds of things are present is best explained by supposing that the former is dependent on the latter for its origin—which justifies the law of causation (in the sense of a cause being a constant and indispensable antecedent) and thereby induction. It is this powerful defense of causation that may turn the tables against the skeptic and in favor of rationality of induction. Regarding inference to the best explanation, it should be noted that a hypothesis that is validly derivable by adding an accepted general law as a premise or fulfills one or more of the criteria of economy is superior to another that is not so. Further, it should be noted that the evidence for plurality of causes is unsatisfactory as also that the notion of being of the same kind may not be explicable without admitting universals that, in their turn, may help to justify induction. Both Carvaka and Hume have questioned the causal law. Quite appropriately the Nyāya has offered substantial arguments outlined earlier for the causal law, arguments to which followers of Carvaka or Hume have given no adequate response. The principal rational grounds for induction, then, come from the counterfactual argument, the argument from belief-behavior conflict, the principle of observational credibility (OC), the principle of inference to the best explanation, refutation of the charge of circularity, the doctrine of universals and the defense of the law of causality.
Such in outline is the later Nyāya justification of induction as we have understood it. Many compromises and simplifications had to be made while borrowing modern terminology for the ease of communication and much of the rigor of the extremely precise Nyāya technical language had to be sacrificed. Still our effort may be a small step toward serious comparative and systematic study. It should, however, be clear, given the serious difficulties facing various contemporary views already discussed, that the Nyāya view is undoubtedly of current philosophical interest. While the Nyāya theories of universals and causality have their perennial place among great philosophical theories, what we find specifically attractive in the later Nyāya justification of induction is the exploitation of the counterfactual conditionals, the notion of contradiction in action, inference to the best explanation, a sophisticated view of circularity and recognition of the value of hypotheses. It is this thorough and comprehensive approach to the problem with a series of modal, epistemological and ontological moves that gives the later Nyāya theory its distinctive appeal.
While Aristotle, the Stoics and the Epicureans made great contributions to the study of induction, there is no firm evidence to show that in the Western tradition the problem of induction was explicitly recognized and elaborately discussed as a serious problem before Hume. But clearly the Indian logicians have done that long before that time. Again, in the Western tradition (notwithstanding the good work done by Whewell, Herschel and Mill earlier in the nineteenth century) it was left to Pierce in the late nineteenth century to bring out the value of the method of hypothesis (calling it abduction and distinguishing it from deduction and induction). Even after that philosophers in this century took time to warm up to the idea as can be gathered from the relative lack of any substantial discussion of this method in the first decades of the twentieth century. The same is true of the link between causation and the counterfactual conditionals. Although some traces are found in Hume, no detailed and systematic study of them is found in any Western writing before the twentieth century. The same, further, applies to the principle of economy. While the principle is very old and sometimes called the Occam’s razor, no Western philosopher has systematically and explicitly studied different kinds of economy before the twentieth century. Similarly, a systematic study of inference to the best explanation is emerging only in some recent publications. As an epistemological theory Nyāya empiricism, though older, appears to be more developed than the modern European empiricism of Locke, Berkeley and Hume. The powerful defense of causality, the careful analysis of circularity, the sophisticated arguments from counterfactual conditionals and belief-behavior conflict appear to give to Nyāya empiricism the decisive edge. No doubt, for some philosophers, skepticism will remain a more attractive position. But for those with a different inclination, the Nyāya position offers a viable option.
The main advantage of the Nyāya view is its well-balanced and multipronged approach to the problem. So far as the analytical justification of induction (seeking virtually to show that the problem of induction is a pseudo-problem) is concerned, it is unlikely that the skeptic will ever be persuaded that the criteria of inductive acceptance are rational in their own right. The Nyāya avoids this kind of head-on collision with the skeptic and quick shortcut to the solution. Instead, the Nyāya has recognized the genuineness and seriousness of the problem and, with great patience and understanding, has sought to expose the various questionable strands around which the skeptical case is built (such as that the skeptical doubt arising from the logical possibility of deviation amounts to making dogmatic claims about empirical possibilities without any shred of empirical evidence, so that the skeptic does not qualify as either a commoner or an expert). Again, an exclusively or predominantly inductive justification of induction will inevitably invite the crushing charge of circularity. No doubt the Nyāya does not concede, even if the skeptic so demands, that induction must be justified on purely noninductive grounds. This is an impossible task. If this is what the skeptic dogmatically insists on (while the larger community of scholars does not endorse this), there can be no worthwhile philosophical debate. Just as it is proper to (partially) rely on a shining lamp to reveal it, so also it is proper to partially rely on induction to reveal its legitimacy. But the Nyāya acknowledges that a justification of induction on mainly inductive grounds is ineffective and also unnecessary. That is why it brings in additional arguments from counterfactual conditionals, etc., and recognizes the role of mental reflection (that serves [among other things] the purpose of what a priori reasoning in modern terminology would sometimes seek to achieve). The pragmatic justification utilizing the straight rule and counterfactual analysis is the closest, among the modern views, to the Nyāya view. But as it stands, the pragmatist position is open to defeating objections such as that the argument for the straight rule applies equally to an infinite number of inductive rules, so that there will be no objective basis for our choice among competing predictions and that the straight rule cannot justify the accuracy of predictions in the short run. However, such objections would have no force against the Nyāya position, for considerations from belief-behavior conflict and inference to the best explanation would ensure that such predicaments do not arise. Surely the principles of inference to the best explanation, such as considerations of economy, can justify the accuracy of predictions in the short run. (If considerations of economy are thrown out as irrelevant or inadequate, the price to pay will be much higher than losing induction.) Similarly, conflicting predictions cannot all and always equally satisfy considerations of economy and cannot all and always have the same implications for our behavior. (This is not to deny that more than one divergent predictions may satisfy [at least some and maybe all] considerations of economy at the same time. There are certainly no a priori arguments to show that this could not happen. But if this does happen, there is no choice but to pile up more inductive evidence to see if the field could be narrowed down further.)
In the same way, although great strides have been made in the investigation of probability in recent times, we are not convinced that this by itself would produce a satisfactory solution to the problem of induction. The natural setting of the calculation of probability is for situations where the terms for predictions can be drawn up in such a way as to allow a conceivable chance of settling their accuracy on the basis of specific observations. This may happen with particular hypotheses, such as that it will rain tomorrow. (A favorite area for the study of probability, quite naturally, is that of betting where it can be determined who wins the bet.) But if the hypothesis is a factual generalization over an unlimited domain, no one can (assuming that only confirming instances are available) decisively settle the accuracy of the prediction and the measure of probability, for no one can know about the truth of each singular conditional deducible from the hypothesis. Hence we are not hopeful particularly in the light of the many difficulties in the major theories of probability discussed earlier that an adequate theory of probability will solve the problem of justifying factual generalizations over indefinitely large domains. We are probably better off in trying to build the bridge between inductive evidence and reliability by exploring counterfactual conditionals and so on, as the Nyāya does.21
Before concluding we look briefly at the so-called new riddle of induction introduced by Goodman.22 Carvaka and Hume tried to show that past and present observed confirmation of a hypothesis does not provide any rational ground for upholding the hypothesis in the future. Goodman’s new riddle highlights the problematic nature of the relation between observed evidence and future prediction in a different way.23 Suppose that all emeralds observed so far are green. This seems to confirm that all emeralds are green and permit the prediction that the next emerald to be seen will be green. But now consider the concocted predicate “grue.” Something is grue if it has been found to be green whenever it has been observed so far or it is not yet observed and will be observed to be blue. Clearly, the observed evidence that seems to confirm that all emeralds are green also seems to confirm that all emeralds are grue. But then we seem to have two conflicting predictions equally confirmed by the same inductive evidence. If all emeralds are green, the next one should be green, but if all emeralds are grue, the next one should be blue. It can be easily seen that we can concoct an indefinite number of grue-like predicates and the same difficulty will arise in each case. That is, if we want to, we can always come up with new, fabricated predicates incorporated into empirical hypotheses that will lead to predictions conflicting with those based on commonly accepted empirical hypotheses while both sets of hypotheses seem to be equally consistent with the observed data. Can induction be rational when it seems to produce such contradictory results?
Goodman’s own solution is that the riddle does not invalidate induction or the generalization formula as such but presses home the need for criteria to separate projectible predicates like green from cooked-up, nonprojectible predicates like grue. The projectible predicates are essentially the well-entrenched ones. What makes a predicate better-entrenched? Essentially that it has a longer history. In Goodman’s own words:
we must consult the record of past projections…. Plainly, “green,” as a veteran of earlier and many more projections than “grue,” has the more impressive biography. The predicate “green,” we may say, is much better entrenched than the predicate “grue.”24
Some critics have complained that such an account of entrenchment leaves the progress of science to luck. Is it merely a stroke of luck that “green” has a longer and more impressive history and biography than “grue”? If so, there is the danger that growth of science may be stultified for excluding hypotheses with unfamiliar or new predicates. Goodman has responded to the criticism by arguing that entrenchment and familiarity are different concepts.25 An unfamiliar predicate may turn out to be well entrenched if the coextensive or parent or comparable predicates are already in frequent and wide circulation. Although Goodman’s theory is more elaborate than our sketch suggests, the criticism, however, has a point. First, Goodman does not show how the danger of excluding new predicates that are not coextensive with or derived from or comparable to other predicates that are already in circulation can be avoided. Second, projectibility and entrenchment are, because of the emphasis on the history, overly dependent on past projection. But for Carvaka or Hume past regularity alone fails to provide rational ground for future regularity and Gangesa, as we have seen, concurs with that.
Quine has offered to explain the distinction between projectible and nonprojectible predicates by saying that while the former are true of things of a kind the latter are not.26 Being of a kind depends on similarity. The more similar things are the more reason that they are of the same kind. Accordingly, a kind is a set of objects that are more similar to a paradigmatic member of the set than they are to something else (called a foil) that is not a member of the set and is too dissimilar to the paradigm. But the difficulty in this view centers round the basis of choosing the paradigm. Is the paradigm chosen because it has certain features or not? If the first, objects should become members of a set by virtue of having most or all of the paradigmatic features which, then, are the family of common features that account for membership of the set. Projectibility then depends on sharing some common features. But objects in a nonprojectible set too may be said to share some common features, such as (at least trivially) that they are grue. So unless we have some reasonable criteria to separate the “right” kind of common features from the “wrong” ones (and none are provided by Quine), the division between projectible and nonprojectible predicates would collapse. If the second, someone may have chosen the paradigm for no reason and others may have followed suit merely for personal reasons. Projectibility then may not have any rational foundation and even inductions with projectible predicates may be irrational. So once again it needs to be shown that although the paradigm is chosen not because it has certain features the set still has a rational foundation but Quine has not done that and it is unclear that a rational basis can be provided. Undoubtedly, a skeptic would like to utilize the situation to press home the irrationality of induction.
The new riddle of induction has generated considerable debate in recent decades and many other solutions and their criticisms have been offered. It would take a whole book to discuss the merits of these solutions and we must skip that. We, however, look briefly at a similar development in Sanskrit logic in the hope of throwing some light on this recent controversy.
Take the stock inference of fire in a hill from smoke. As pointed out earlier, the hill is the inferential subject (paksa) wherein a typical case smoke is observed and fire is not. That there is fire in the hill is open to doubt; the doubt is removed by the inference of fire in the hill. The inference is based in part on the general premise or pervasion (vyāpti) that all smoky things are fiery. The general premise is supported by observation of instances where smoke is found with fire and/or observation of instances where absence of fire is found with absence of smoke. Here smoke is the pervaded (vyāpya: something the extension of which is not wider than that of the pervader, anatirikta-deśavŗtti) and fire is the pervader (vyāpaka: something the extension of which is not smaller than that of the pervaded, anyūna-deśavŗtti). The pervasion is supported by observation of instances where smoke is found with fire and/or observation of instances where absence of fire is found with absence of smoke. The former are positive instances (sapakşa) and the latter are negative instances (vipakşa). The inferential subject is neither a positive instance nor a negative instance, for presence of the probandum is reliably known in a positive instance before the inference and absence of the probandum is reliably known in a negative instance before the inference and neither the presence nor absence of the probandum is in a typical case reliably known in the inferential subject before the inference. [However, in atypical cases presence or absence of the probandum may be known in the inferential subject.]
Now take the cooked-up property of “not being either the inferential subject or a negative instance” (disni: pakşa-vipakşa-anyatara-anyah) cited by Gangesa.27 This property in a typical case is true of any positive instance: a positive instance (being where presence of the probandum is reliably known) is not either the inferential subject (where neither presence nor absence of the probandum is reliably known) or a negative instance (where absence of the probandum is reliably known). In the above stock example a case of disni is not being either the hill or a lake: the latter is true of a fiery kitchen hearth that is neither the hill nor a lake. It should be clear that by definition in typical cases disni is present wherever the probandum is reliably known to be present before the inference. [Disni is also not true of any negative instance: it is not true of any negative instance that it is neither the inferential subject nor a negative instance, for it is a negative instance and if something is a negative instance, it is also either the inferential subject or a negative instance (i.e., an inclusive disjunction is true if either disjunct is true). Thus, by definition, in typical cases wherever there is absence of disni there is absence of the probandum.] It seems to follow that there is warrant for the generalization that wherever there is the probandum there is disni or that disni pervades the probandum. In the above stock example, then, there seems to be warrant for the generalization that no fiery things are either the inferential subject or a negative instance.
At the same time disni cannot be true of the inferential subject. If disni pervades the probandum, absence of the probandum in the inferential subject then follows from absence of disni. But we also reliably know that the probans belongs to the inferential subject (in the stock example: the hill is smoky) and that wherever there is the probans there is the probandum (in the same example: all smoky things are fiery). Thus the above set of facts seems to warrant both inference of the probandum and its absence in the same thing at the same time—a contradiction. We symbolize this by replacing the inferential subject with “this,” the probans with “M,” the probandum with “P” and being either the inferential subject or a negative instance with “Q” as follows.
All M is P.
This is M.
Therefore, this is P.
But also
No P is Q.
This is Q.
Therefore, this is not P.
The problem is mainly due to that the same generalization formula that permits the induction that wherever there is the probans there is the probandum also permits the induction that wherever there is the probandum there is disni. It is in this respect that this problem is similar to the new riddle of induction. In Goodman’s example the observed facts seem to support both that all emeralds are green and that all emeralds are grue. If all emeralds are green, the next emerald should be green. But if all emeralds are grue, the next emerald should be blue. This is a contradiction pointing to some possible gap in the generalization formula. In Gangesa’s example the observed facts seem to support in a typical case both that all smoky things are fiery and that no fiery things are either the inferential subject or a negative instance. If all smoky things are fiery, then (since the hill is observed to be smoky) the hill is fiery. But if no fiery things are either the inferential subject or a negative instance, then (since it is true of the hill that it is either the inferential subject or a negative instance), the hill is not fiery. Here too is a contradiction pointing to some possible gap in the generalization formula.
As another example (freely coined by utilizing Nyāya views) take the inference that this mango is colored because of being a fruit. Here an instance of disni is “not being either the mango or an air molecule.” [In the Nyāya view air is colorless.] “Not being either the mango or an air molecule” is true of any reliably known colored thing, such as a banana or a pebble. [As already said, the inferential subject is not included in the class of positive instances in a typical case.] So it seems to be permissible to generalize that nothing colored is either a mango or an air molecule. If this is accepted, it follows that this mango is not colored. But at the same time it is reliably known that this mango is a fruit and that all fruits are colored. Given these premises it follows that this mango is colored. Once again we have a contradiction pointing to some possible gap in the generalization formula. [Needless to say, in this example as well as in the previous example of inference of fire from smoke in a hill, the singular conclusion could be replaced by a universal or a particular statement requiring appropriate changes in the rest of the argument as well.]
The following is a solution (among others) mentioned in Gangesa’s TC.28 Induction does need the support of observation of positive instances or negative instances. Further, there should be nonobservation of any counterexample. Thus the generalization formula so far comprises observation of positive instances or observation of negative instances and nonobservation of any counterexample. The assumption that this is the whole story, however, leads to the problem. To solve the problem it needs to be added that a reliable induction must also have the support of additional reasoning to counter the doubt that the induction may be false. The doubt that an induction may be false is reasonable; an induction includes a claim about future countless cases based on favorable observation of a limited number of past or present cases. But sometimes an induction confirmed in a large number of cases is found later to have a counterexample. So it is reasonable to suppose that a counterexample may be found in other cases where none has been found so far. Such doubt should be countered by additional reasoning that explores the consequences of supposing that an induction is false and shows that an undesirable consequence results from that. Such additional reasoning is called tarka that we translated as counterfactual reasoning: CR. It includes a counterfactual conditional the antecedent and the consequent of which are false. The following reasoning as noted earlier has been offered in support of that all smoky things are fiery.
If smoke were produced neither by an aggregate that includes fire nor by an aggregate that excludes fire, smoke would not have been produced.29
We develop following the earlier discussion the argument as follows. Gangesa has offered a counterfactual conditional to back up the induction that all smoky things are fiery. The conditional is: if smoke were produced neither by an aggregate that includes fire nor by an aggregate that excludes fire, smoke would not have been produced. But we observe, it is implied, that smoke is produced. So the consequent is false. It follows (by applying the implied law of modus tollendo tollens) that the antecedent is false. So we derive (by applying the implied De Morgan law) that smoke is produced either by an aggregate that includes fire or by an aggregate that excludes fire. Now we have two opposed factual claims, viz., (1) that smoke is produced by an aggregate that excludes fire and (2) that smoke is produced by an aggregate that includes fire. It is again implied in Nyāya empiricism (and empiricism in general) that of two factual claims the one that has observational support is preferable to the one that does not. This may be called as noted the principle of observational credibility (OC). Given OC, it is then accepted that smoke is produced by an aggregate that includes fire. But to say that smoke is produced by an aggregate that includes fire is to say that fire is a constant antecedent of smoke, for a causal condition (kāraņa) is defined in part as a constant (niyata) condition. [“Constancy” is added to separate a causal condition from an accidental factor such as a donkey that happens to be present where smoke is produced and is not a causal condition of smoke.] The argument thus bestows favor (anugraha) to the induction that wherever there is smoke there is fire by showing that its denial leads to the undesirable consequence (anişţaprasañga) of conflict with reliably accepted views.
The above argument implicitly utilizes logical laws like modus tollendo tollens as well as OC. While even logical laws are not above challenge, they are as safe as it gets; proponents of inductive skepticism like Carvaka or Hume accept them as well. Accordingly, it is not absolutely necessary to argue for them here. So far as OC is concerned, critics of induction like Carvaka or Hume should not reject it. Although Carvaka rejects an inductive leap into the future as unreasonable, he holds that particular observations may be reliable (pramanika) and are the only sources of knowing. Similarly, Hume labels induction as questionable but holds impressions or observations of particulars as the ultimate epistemic foundations. Neither the position of Carvaka nor the position of Hume can be sustained without OC.
One may object that appealing to OC does not quite get the job done. The point may be elaborated by looking again at the two claims that (1) smoke is produced by fire and (2) that smoke is produced without fire. (1) may involve simply claiming (1A) that smoke in the past and the present has been produced by fire and (2) may involve simply claiming (2A) that smoke in the past and the present has been produced without fire. As between these two empirical claims OC clearly favors (1A). Neither Carvaka nor Hume would also object to accepting (1A) in preference to (2A). However, (1) may also additionally involve claiming (1B) that smoke will be produced by fire and similarly, (2) may additionally involve claiming (2B) that smoke will be produced without fire. Clearly, (1B) does not logically follow from (1A). So a skeptic who accepts OC and also accepts (1A) does not thereby necessarily commit to accepting (2A) over (2B). Accordingly, the skeptic may still maintain that acceptance of (2A) over (2B) is on such irrational grounds as custom or habit. Thus, Carvaka or Hume may be interpreted as implying that knowledge claims should be confined to past and present observations (for Carvaka knowledge claims more strictly speaking should be confined only to present observations) and should not be extended to the future. Claims about the future are not instances of knowledge but anticipations based on repeated observation.
But the skeptical claim is not simply that claims about the future are based on habit but also additionally that claims about the future cannot be possibly based on rational grounds. So a critic may agree with the skeptic that claims about the future are sometimes based on habit and still disagree with the skeptic that it is impossible for such claims to be based on rational grounds. An inductionist is certainly under no obligation to restrict OC to only factual claims about the past or the present. To remove any possible ambiguity and include factual claims about the future under the umbrella of OC the following strengthened version of OC or OC’ may be offered: of two factual claims the one that has observational support is preferable to one that does not and in case of factual claims about the future the one that is homogeneous with an accepted factual claim is preferable to one that is heterogeneous. What is homogeneous and what is heterogeneous? A factual claim about the future is homogeneous with an accepted claim if the former can be derived from the latter merely by changing the tense of the latter from the past or the present to the future. On the other hand, a factual claim about the future is heterogeneous to an accepted claim if in order to be derived from an accepted claim at least an additional logical operator is needed over and above changing the tense. Such a heterogeneous claim is also homogeneous with a rejected claim. Thus factual claims that are homogeneous with accepted claims are preferable to ones that are heterogeneous and are homogeneous with a rejected claim. OC or OC’ fit with the majority opinion of the scientific and scholarly community. The burden of proof is on Carvaka or Hume to show why these should be rejected and why it is preferable to have doubts based on claims that are homogeneous with rejected claims.
Does OC or OC’ involve induction so that the Carvaka-Humean charge of infinite regress or circularity can be brought back? No. OC or OC’ involves only mental reflection (mānasa-jñāna) on the nature of empiricism. Since no external observation is involved in mental reflection or analysis or unpacking of conceptual contents (vişayatā), the charge of circularity or infinite regress is groundless.
Accordingly, claims about the future may be chosen on the basis of such a general principle of reason that is consistent with empiricism. This is irrespective of the question of the principle’s fit with the particular (and self-refuting, as already argued) brand of empiricism of Carvaka or Hume. A healthy empiricism need not confine knowledge claims to only past or present observations. If the present is continually becoming the past and the future is continually becoming the present and no line can be drawn between the past, the present and the future, can any observational reason be given for making exclusive knowledge claims about only the past or the present? In other words, if time is a continuum and if divisions within a continuum are relative, are there any consistently empiricist reasons to justify absolutistic claims with reference to a division within the continuum? If not, since future-oriented claims can be based on such a principle as OC or OC’, the skeptical claim that such choice can only be based on anticipation or habit is refuted.
If this makes sense, the above argument of Gangesa is relevant so far as the critique of induction goes.30 This does not make induction infallible (infallibility may be claimed for logical truths but is ruled out for induction in any case), but it (together with observation of positive instances or negative instances and nonobservation of any counterexample) does make it reliable (prāmāņika). In Nyāya epistemology reliability is inferred from successful action (saphala-pravŗtti) prompted by a cognition (such as when a thirsty person looks for water, finds it, drinks it and the thirst is quenched) or its fit or coherence (samvāda) with other accepted truths.31
A skeptic may point out that the above reasoning involves at least the induction that a supposition that conflicts with accepted views is not reliable. Accordingly, the reasoning is circular, the skeptic may object. The skeptic may add that whatever reasoning is offered in support of induction would inevitably involve some induction and be invariably circular.32
But the assumption behind this objection is that if the same rule is involved in the justification of a given rule, the reasoning is circular.33 This assumption is questionable, as argued earlier. Further, the skeptical claim that no argument can show the reliability of induction without presupposing that reliability itself involves induction and cannot be sustained without presupposing its reliability. Thus, if rule-circularity is a flaw, the skeptical objection is flawed too. In particular, if the reliability of induction cannot be challenged without presupposing that reliability, the challenge is futile.
Another response to the above objection involves a distinction between a meta-induction such as that a supposition which conflicts with accepted views is not reliable and a proper induction such as that all smoky things are fiery. The former is a second-order proposition relying primarily on conceptual analysis and mental reflection. The latter is a first-order proposition relying primarily on external observation. While the charge of circularity may be relevant if one relies on a proper induction in the process of justifying induction, it loses all force when applied to a meta-induction. Just as what applies to meta-logic does not necessarily (and in fact is not according to many expected to) apply to logic, what applies to meta-induction does not necessarily apply to induction.
If one overlooks the distinction between meta-induction and induction, one may be persuaded by Howson’s argument that an inductive rule may, by using a variant of the grue case, be shown to prove its own unreliability (HP 30–31). Suppose that an induction based on an inductive rule is right if it implies something true and wrong if it implies something false. Now suppose that an induction is “ring” if it is checked and right or not checked and wrong. It follows that an induction that is found to be right is also “ring.” It also follows that if most checked inductions are “ring,” most inductions are also “ring.” But only a finite number can be checked, leaving the remaining potential infinity unchecked and, therefore, wrong. Thus most inductions based on an inductive rule turn out to be wrong. But the above argument is without teeth, for a grue-like exercise involving meta-induction proves nothing about induction proper.
The main point of the solution then is that a reliable induction should have the support of observation of positive instances or negative instances and nonobservation of any counterexample and also have the support of counterfactual reasoning so that the denial of the induction would lead to an undesirable consequence. The undesirable consequence may be a contradiction in action (such as if I speak aloud in so many words that I am dumb) or a practical conflict (such as belief-behavior conflict utilized in Gangesa’s example discussed earlier) or conflict with something reliably accepted or accepting something that is uneconomical (guru).
Now let us look at the grue case. So far as the support from observation of at least one positive instance and nonobservation of any counterexample is concerned, both that all emeralds are green and that all emeralds are grue appear to have that support. But there is a difference when it comes to the support from counterfactual reasoning. Suppose that “all emeralds are green” is false and that the next emerald to be seen is not green. Then that emerald will not complement red, for only green complements red. But the next emerald may be observed to complement red and that would conflict with the supposition that it is not green. Thus the assumed denial of the induction that all emeralds are green has the undesirable consequence that it invites the risk of conflict with what may be observed in the next case. Now suppose that “all emeralds are grue” is false and that the next emerald to be seen is not blue. No undesirable consequence follows. Even if the next emerald is observed to complement red, there is an incongruence: something not blue may complement red. Hence “all emeralds are grue” fails to qualify as a reliable induction.
Further, compared to grue green is simpler with respect to constitution (sarira)—that is, grue appears to contain more concepts than green. Compared to grue there is also greater economy in the cognitive link/order (upasthiti) so far as ordinary discourse is concerned: in ordinary discourse awareness of grue cannot take place without awareness of green; but awareness of green can take place without awareness of grue. [In the light of Nyāya ontology there will be moreover greater economy in relationship (sambandha). But exploration of Nyāya ontology will take much space here and must be skipped.] So green is preferable to grue in terms of the principles of economy (lāghava) as well.
It may be noted that Gilbert Harman has argued that “all emeralds are green” is preferable to “all emeralds are grue” because the green hypothesis is more economical or simpler than the grue one.34 Harman has proposed a computational or pragmatic theory of simplicity according to which ease of computation is the basis of preference among competing and equally relevant hypotheses. That is, theories that are easier to use in getting results in which scientists are interested are preferable to those that are harder to use in getting those results. This is somewhat similar to a part of the Nyāya solution. But Harman does not explicitly recognize the subtle distinction between the three kinds of simplicity as the Nyāya does.35 Further, the Nyāya does not rely on simplicity alone and holds, as already said, that an undesirable consequence may be due to belief-behavior conflict or conflict with something reliably accepted (e.g., that fire is cold is liable to be rejected on the ground that fire is directly observed to be hot) and so on. This is an important difference between the Nyāya and Harman and other proponents of a simplicity solution. Although simplicity may sometimes help to determine what is reliable or true, relying on simplicity alone may not suffice to show that we are getting any closer to truth or reliability. That is, what still needs to be argued for is that the fact that one theory is simpler than another is a good reason for saying that the former is more likely to be true or reliable than the latter.36 This issue is implicitly addressed in the Nyāya solution. That the denial of a hypothesis conflicts with something reliably accepted or conflicts with the way one regularly behaves may (among others) be offered as good reasons to think that the hypothesis is likely to be true or reliable.37
Further, unlike Quine’s position, the Nyāya solution does not hinge on deciding which predicate represents a kind and which predicate does not—a task that appears to be fraught with difficulties to say the least. Again, unlike Goodman’s position this solution is not pinned down to checking the past history of how often a particular predicate has been projected and does not leave the choice between two predicates to counting which predicate (along with coextensive or parent or comparable predicates) has been projected more often in the past. So the test that an induction is not reliable unless the assumed denial leads to an undesirable consequence, does not leave the progress of science to luck and does not forbid the introduction of new predicates. The undesirable consequence may also result from future developments. An induction that passes the test now may fail it in the future. There are no guarantees in nature. Since Gangesa is a fallibilist, he does not also try to find one.
Finally, David Sanford has argued that since grue is a disjunctive predicate and green is not, a part of the solution may be found in a clearer, objective and semantic (as distinguished from a merely syntactic) analysis of disjunctiveness that Sanford has offered.38 He also notes that while in the predicate grue there is a link between a color term and a temporal term, terms that are semantically disconnected, there is no such linking of what are semantically disconnected in the predicate green. It is remarkable that Gangesa too has cited a problematic property that includes disjunction of semantically disconnected terms like the hill and where absence of the probandum is known.
But it should be added that Gangesa has also shown (TC, chapter on Upādhi, 301) that other properties that are not disjunctive such as “not being the inferential subject (bois: pakşetara)” turn out to be equally problematic and may be handled in the same way explained above. Thus bois appears to pervade the probandum, for no positive instance where the probandum is known to be present is the inferential subject. Bois cannot be also true of the inferential subject and it seems to follow that the probandum does not belong to the inferential subject no matter what is offered as the probans. In inference of unobserved fire in the hill from observed smoke, bois amounts to “not being the hill.” Since fire is not observed in the hill, “not being the hill” appears to pervade fire. At the same time since “not being the hill” cannot be true of the hill, it appears to follow that fire does not belong to the hill no matter whether smoke or something else is found in the hill. Gangesa discusses the problem at length but one of his main points is as follows. Bois does not reliably pervade the probandum for the lack of CR that would obstruct the doubt over that induction. Since the probandum may be present in the inferential subject and since bois is necessarily missing in the inferential subject, there remains the lingering doubt that bois may not pervade the probandum.39
If this makes sense, since bois is not disjunctive, disjunctiveness may not have a crucial role in the present issue. But Sanford’s point is also that perverse predicates like grue are formed by linking terms that are semantically disconnected. This holds of the overtly nondisjunctive properties cited by Gangesa. For example, bois links by implication terms like “the hill” with terms like “where typically neither presence nor absence of the probandum is known” and, therefore, are formed by linking terms that are semantically disconnected. Since grue-like predicates link terms that are semantically disconnected, they would also be more complex with respect to constitution (śarīra), cognitive order (upasthiti) and relation (sambandha) compared to predicates like “green” or “fire” that do not link terms that are semantically disconnected. Further, since grue-like predicates link terms that are semantically disconnected, the hypotheses concerned would be without support from counterfactual reasoning and their denial would not lead to an undesirable consequence.40
1. See K. Chakrabarti, “Some Remarks on Indian Theories of Truth,” Journal of Indian Philosophy, XII, 1984, 339–55.
2. TS 378. It may be noted in this connection that some recent philosophers like Nelson Goodman in his Fact, Fiction and Forecast (FFF) (Harvard University Press, Cambridge, fourth ed., 1983) have regarded the problem of induction and the problem of the justification of counterfactual statements as merely two complementary ways of looking at the same thing. This is analogous to the Nyāya view that exploration of counterfactuals is relevant to the rationality of induction. In the Nyāya view the falsity of a counterfactual is no bar to its being conducive to truth. Further, while a false induction could fail an empirical test, a counterfactual by its nature can never be subjected to any empirical test by realizing its antecedent. However, a proper discussion would require making deeper inroads into Nyāya epistemology that is beyond the scope of this study.
3. TS 69.
4. An alternative translation requiring a change, mutatis mutandis, in the reformulated version given below is as follows: if smoke were not produced by what is not collocated with fire and were not also produced by what is collocated with fire, it would not have been produced.
5. An alternative reformulation of the above argument with complex terms is also possible. For a discussion of the relevant ambiguity of Sanskrit texts, see K. Chakrabarti, “Some Non-Syllogistic Forms in Early Nyāya Logic,” Proceedings of the Fifth International Congress of Logic, Philosophy and Methodology of Science, Ontario, 1975, section 12, 9–11.
6. John Anderson, “The Problem of Causality,” Australasian Journal of Psychology and Philosophy, 16, 1939; cited in J. L. Mackie, The Cement of the Universe, Clarendon Press, Oxford, 1974, 35, footnote.
7. NK 9–12.
8. Udayana allows the admission of nonempirical causal conditions like God only if empirical causal conditions are not available.
9. NK 13–22. Hume too says that where several different things produce the same effect, it must be by means of some common quality in them; he also holds that the difference in the effects of two resembling things must proceed from that particular in which they differ (Treatise I, iii, 15).
10. See K. Chakrabarti, “Nyāya-Vaiśeşika Theory of Universals,” Journal of Indian Philosophy, III, 1975, 363–82.
11. For the Nyāya theory of truth, see “Some Remarks on Indian Theories of Truth,” op cit.
12. See, for example, Nicholas Rescher, Induction, University of Pittsburg Press, Pittsburgh, 1980, 119.
13. This objection has been raised by Sriharsa (twelfth century) in the Khaņdanakhaņdakhādya, Chowkhamba Vidyabhavan, Varanasi, 1992, 386. More discussion on this follows later. Needless to say that Hume would agree with Sriharsa.
14. Don Garrett makes these points in Cognition and Commitment in Hume’s Philosophy, Oxford University Press, Oxford, 1997, chapters 4 and 10.
15. See “Some Remarks on Indian Theories of Truth,” op cit.
16. Descartes, Meditations on First Philosophy, translated by D. A. Cress, Hackett Publishing Co., Indianapolis, 24.
17. ATV 533.
18. See “Some Remarks on Indian Theories of Truth,” op. cit.
19. Ibid.
20. TC 230–51.
21. Useful discussion of the Nyāya view and some other Indian views on the classical problem of induction is found in S. S. Bagchi, Inductive Reasoning, University of Calcutta Press, Calcutta, 1949 (which, though somewhat dated, gives the best coverage of a vast body of Sanskrit material); E. A. Solomon, Indian Dialectics, vols. 1 and 2; Gujarat Vidya Sabha, Ahmedabad, 1976; and Raghunath Ghosh, The Justification of Inference, Bharatiya Vidya Prakashan, Delhi, 1990.
22. Fact, Fiction and Forecast, abbreviated as FFF, chapter III.
23. Goodman himself holds that the classical Humean problem of induction has generated much fruitless discussion and should be dissolved (FFF, chapter III). But we shall see that Goodman’s own solution to the new riddle is vulnerable to the Humean critique. However, see note 39 below for the point that elsewhere Goodman comes close to the Nyāya solution.
24. FFF, 94.
25. FFF, chapter IV.
26. Quine, W. V., “Natural Kinds,” in Grue! (GR), ed. Douglas Stalker, Open Court, Chicago, 1994, 41–56.
27. Pakşa-vipakşa-anyatara-anyah yathā prasiddha-anumāne parvata-jalahŗadaanyatara-anyatvam: “not being either the inferential subject or a negative instance” (is a pseudo-adjunct or pseudo-property, upādhyābhāsa), for example, “not being either the hill or a watery lake” with reference to the stock inference (of fire from smoke in a hill), Tattvacintāmaņi (TC) of Gangesa, ed. K. N. Tarkavagisa, vol. II, Chowkhamba Sanskrit Pratisthan, Delhi, 1990, 403–4.
28. Tatra anukūla-tarka-abhāvena sādhya-vyāpakatva-aniścayāt sahacāra-darśānadeh tena vinā samśāyakatvāt: due to the lack of supportive CR pervasion of the probandum (by the said property) is uncertain; without that (CR) observation of co-presence and so on is subject to doubt (i.e., is a doubtful base for the inductive claim) (TC 355).
29. Dhūmo yadi vahni-asamavahita-ajanyatve sati vahni-samavahita-ajanyah syāt, notpannah syāt (TC 219). The argument has another important part utilizing in particular belief-behavior conflict that we have discussed earlier.
30. For a discussion of difficulties in various attempted solutions to the classical problem of induction and how this solution differs from them and avoids those difficulties, see preceding discussion.
31. For further discussion see Chakrabarti, Kisor, “Some Remarks on Indian Theories of Truth,” op cit.
32. This objection has been raised by Sriharsa (twelfth century) in his Khaņdanakhaņdakhādya, Chowkhamba Vidyabhavan, Varanasi, 1992, 386.
33. This kind of circularity is sometimes called rule-circularity. See Rescher, Nicholas, Induction, University of Pittsburgh Press, Pittsburgh, 1980.
34. “Simplicity as a Pragmatic Criterion for Deciding What Hypotheses to Take Seriously” in GR 153–72.
35. Harman does mention simplicity of representation that is similar to what is called economy with respect to constitution in the Nyāya. He also distinguishes computational simplicity from semantic simplicity advocated by Elliott Sober in Simplicity, Oxford University Press, Oxford, 1975. According to Sober, a hypothesis is simpler and preferable to another if less information is needed in the light of the former compared to the latter to answer questions that matter.
36. Another kind of undesirable consequence, as already mentioned, is due to contradiction in action, (e.g., in speaking aloud that I am dumb). Yet another kind of undesirable consequence is due to conflict between what one asserts and the way one regularly behaves.
37. Harman tries to show that computational simplicity is an indicator of verisimilitude. He, however, assumes in this process that there is no difference between believing something and believing that it is true. This assumption is questionable: believing something without believing that it is true seems to be quite possible. For example, a man of science may participate in religious rituals in the belief that this is needed for afterlife although he has also reason to believe that science rejects life after death. At least the Nyāya claims that it is psychologically possible to hold on to a contradiction in spite of being aware of that it is a contradiction.
38. “A Grue Thought in a Bleen Shade: ‘Grue’ as a Disjunctive Predicate,” in GR 173–92.
39. D. H. H. Ingalls, who taught at the Harvard University where Goodman too taught, was trained in Nyāya philosophy by a traditional pundit in Kolkata. Goodman makes no reference to the Nyāya anywhere. Still, it is possible that Goodman and Ingalls had some philosophical conversations. It is also possible that Goodman met B. K. Matilal, a leading specialist in Nyāya philosophy trained by pundits in Kolkata, who spent a few years at Harvard and that Goodman was influenced by Nyāya ideas in coining perverse predicates like the grue. Goodman holds: “A hypothesis is projectible if and only if it is supported, unviolated, and unexhausted, and all such hypotheses that conflict with it are overridden” (Problems and Projects, Bobbs-Merrill, Indianapolis, 1972, 393). This is analogous to the Nyāya view that an induction (or a hypothesis: kalpanā) is not reliable unless it is supported by subjunctive reasoning that shows that the denial leads to an undesirable consequence. As already said, the Nyāya theory of undesirable consequence is highly developed. An undesirable consequence may be a contradiction or a belief-behavior conflict or a conflict with something reliably accepted or acceptance of something uneconomical.
40. In an interesting article Stephen Hetherington has offered a solution of the Goodman paradox by drawing attention to the psychological nature of the evidence on which induction is founded. More specifically, Hetherington draws a distinction between experience-of and experience-as and suggests that one’s experience could be of a thing being green without one’s experiencing that thing as being green. Accordingly, whether one would generalize from the observation of emeralds that all emeralds are green or rather that all emeralds are grue would turn on whether one experiences emeralds as green or rather as grue. This avoids, Hetherington suggests, having to say that for a particular subject the same observation may be evidence for both that all emeralds are green and that all emeralds are grue. (S. Hetherington, “Why There Need Not Be Any Grue Problem About Inductive Evidence As Such,” Philosophy 76, 2001, 127–36.) But then the reliability of a scientific law or induction would have to hinge on the psychological nature of a particular subject. While a given subject may experience emeralds as green another subject may experience them as grue. Accordingly, for the former subject all emeralds may be green and for the latter subject all emeralds may be grue. Further, there is nothing to rule out the possibility that even the same subject may experience emeralds as green until now and then from now on experience emeralds as grue. For such a subject there would then be evidence for both that all emeralds are green and that all emeralds are grue. The underlying idea seems to be that a normal subject would experience emeralds as green rather than as grue. But then one would need a criterion to distinguish a normal subject from an abnormal subject. If normalcy is defined in terms of agreement with most observers, induction is presupposed; the charge of begging the question then looms large.