From the perspective of addiction as dysfunctional decision-making, gambling and other behaviors can also be seen as “addictions.” These nonpharmacological addictions tend to depend on vulnerabilities in the support systems (e.g., motivational systems, situation-recognition systems) more than in the action-selection systems themselves, but they do also depend, of course, on how action-selection systems use those support systems.
An interesting question that has arisen over the past few years is whether addiction is fundamentally related to drug use or whether one can be addicted to something that is not drug-related.1 From the old “dependence” theory of addiction—that drugs change your internal balance of some chemical2—it is very hard to describe nondrugs as an addiction. But from the “maladaptive” theory (that you put up with high costs to get your hit) or the similar economic “inelastic” theory (that use doesn’t decrease as much as one expects when costs increase),3 there are lots of things that one can get addicted to. But there are also lots of things that are inelastic that we don’t want to say are bad (like breathing!). However, working from the “malfunctioning decision-making machinery” theory described in this book, we can see that behaviors too can become overselected given the underlying correct valuation they should have.
In the clinical world, the most common non–drug-related addiction is problem gambling, which affects between 1% and 4% of the U.S. population, depending on the specific criteria used to define it.4 Problem gambling is usually identified because the maladaptive consequences (losing money, family, stability of life) are often so clear. From the decision-making perspective, problem gambling is definitely an addiction. Our decision-making system is not optimal; it can be fooled into making poor decisions under the wrong conditions.5 As with our description of multiple potential ways that pharmacological substances (drugs) can access the decision-making system and can drive incorrect decisions, there are multiple ways that sequences of wins and losses (gambling) can drive incorrect decision-making. We discussed some of these potential vulnerabilities in the last chapter, which concentrated on the pharmacological addictions.
In addition to these motivational effects, humans are very poor at recognizing random sequences as random and are always looking for underlying causes.6 This means that we can be fooled by limited experience with a probabilistic event. Because we do not see the correct distribution of possibilities, or the correct distribution of the likelihood of those possibilities happening, we think we have more control over the system than we actually do. (This is most likely a failure in the situation-recognition system7—we are unwilling to accept coincidence and are always constructing narratives to explain why things turned out the way they did.8)
In particular, humans are very susceptible to the “near-miss,”9 where we are told that we “almost got it,” which leads us to think that we are making progress on figuring out the complex system that is the world in which we live. This is something that slot-machine makers have learned to exploit.10 Over the past several decades, slot machines have gotten more and more complex until they are almost like videogames these days. Of course, casinos have also learned how to reduce one’s ability to recognize these complexities by providing an abundance of lights and cues all around the casino to draw your attention away, and by providing alcohol to diminish your ability to plan and think.
Imagine taking the 10 digits (0 through 9) and ordering them. Which sequence of numbers seems more random to you: 4037981562 or 0123456789? I’ll bet it’s the former (unless that’s your Social Security Number). Actually, they are each equally likely. If you are selecting a random sequence of 10 digits (without replacement), the probability of seeing any specific sequence is 1 in 3.5 million.A
Several researchers have argued that the reason animals evolved brains is prediction: the better you can predict what’s going to happen, the better you can prepare for it, the better you can react to it, the better you are going to be at surviving and procreating your genes into the next generation.11 We are evolved to recognize patterns, even if they arise from true randomness. This has several effects on our decision-making abilities. First, it means that we are terrible at generating random sequences.12 Second, it means that we find patterns even in randomness.13
Ask a hundred people to generate a string of five random digits and I’ll bet none of them will say “00000” or “99999.” (Unless they’re neuroscientists or psychologists who study this stuff and will come up with those numbers just to screw with your experiment—but of course, then those numbers aren’t random either.) In general, people will start off with a number (say 7) and then try to find a number that is different from that, and then come back to another number. This is a process called the “inhibition of return” and is suggested to be a means of improving foraging.14 In a sense, one is “foraging” for numbers,15 and, having found 7, one wants to find the next number elsewhere.
My favorite demonstration that randomness is hard to do is that computer programs can learn to beat us at rock-paper-scissors.16 This is the game where you and your opponent count to three and, on three, place your hand as a fist (rock), as a flat palm (paper), or with two fingers making a cutting motion (scissors). In this game, each choice beats one of the other choices and each loses to one choice in a ring: rock beats scissors, which beats paper, which beats rock. If you can predict what your opponent is going to do, then you can always win. But if you can’t predict, then the optimal strategy is to play randomly. Unfortunately, humans are terrible at playing randomly. As part of his graduate thesis work, Steve Jensen at the University of Minnesota developed an algorithm that basically remembers sequences that it has observed. It uses these sequences to predict future outcomes. Because it has a very large memory, this program will eventually beat a human opponent. If it plays only one game against you, then who wins will be random. But if you play a hundred games against it, it will win more than half the time (a lot more than half!). It’s a remarkably frustrating experience, because even if you know you have to play randomly and you try to play randomly, the computer keeps guessing correctly what you’re going to do!
Predicting what your opponent is going to do is a key factor in most competitive sports, whether it be a baseball batter predicting which pitch the pitcher is going to throw (and where he is going to throw it), a football coach predicting whether a pass play or a running play will be coming up next, or a tennis player predicting where her opponent will serve the ball next. Even a small improvement in knowing the next option will play out big over the course of a match or a season. And, of course, the ability to generate randomness (or its converse, the ability to predict what an opponent is going to do) has important implications for other pursuits as well, including business, politics, and war.B
The other implication of our predilection for prediction is that we recognize patterns that are not there.17 From the man in the moon to religious icons on moldy bread to patterns in marble tiles, our brains look for patterns in the world. Some of these patterns are real. Mushrooms and flowers tend to appear after a big rainstorm. A fist coming at your face is likely to hit you if you don’t block it and likely to hurt if it hits you. Other patterns are not real. The man in the moon is an illusion caused by an interaction between the specific pattern of dark lowlands and bright highlands on the moon’s surface with the facial-recognition systems genetically hardwired into our brains.
The recognition of patterns that don’t exist becomes a real problem when it leads to the illusion of control.18 In a casino, playing a game, sometimes you win and sometimes you lose. Let us assume, for now, that these games are truly random. But imagine that you notice that you broke a long losing streak just after you touched your hat and wiped your brow. Maybe that’s worth trying again. Of course, it probably won’t work the second time, but now we’re in a probabilistic world—trying to figure out whether winning is just a little bit more likely in one condition or another.
Of course, logically, it doesn’t make sense why wiping your brow would change a slot machine’s outcome, but what if, instead, we think of how you push the button, or when you push the button in the course of a spin of some video effect on the machine? Now, we have a potential logical cause that could affect the machine’s outcome. Imagine a slot machine with a simple switch. The switch could even have a sign on it that says “this switch does nothing.” Yet someone playing the machine will likely try the switch, and will be more likely to win with the switch in one position or the other. (Although flipping a coin will come up tails half the time on average, the chance is very low that any specific sequence of coin flips will be half heads and half tails. Similarly, the number of wins you observe with the switch in each position is very unlikely to be exactly the same.) So, the player will believe (incorrectly) that the switch has an effect. This is the illusion of control.
This is effectively a form of superstition.19 B. F. Skinner (whom we met as a prominent behaviorist in our discussion of reinforcement in Chapter 4) showed that pigeons would eventually do interesting, individual random behaviors if they had done them a couple of times before getting a reward, even if the reward was actually being delivered randomly.
A large part of our intelligence is dedicated to trying to predict the world. Several researchers have argued that learning is about recognizing contingencies—whether an event is more (or less) likely to occur in the context of another event.20 As we saw in our discussion of situation-recognition and the construction of narratives (Chapter 12), much of our cognitive effort is spent on recognizing causal structure within the world. This has been a hypothesis for the origin of mythologizing21 since Giambatista Vico.C
So what happens when your causal hypotheses make mistakes—when you recognize differences between two situations that are not really different? This leads to the illusion of control—if you think that there are two situations (one in which you win and one in which you lose) and you can find a way to get yourself into the winning situation, then gambling makes sense. Of course, this is where cognitive errors can lead to problem gambling.23 Think of the stereotypical gambler in the movies—“I know what’s different. Loan me the money; it’s a sure thing this time. I’ll pay you back when I win.” That gambler isn’t hooked on randomness. That gambler isn’t looking to play the game. That gambler thinks he understands the causal structure of the world well enough to control it. That gambler has succumbed to the illusion of control.
We can also see that incorrectly recognizing when a situation has changed can lead to gambling problems. If a situation has changed, but you do not recognize the change, then you will be performing the wrong actions in the wrong situations. Again, cognitive errors lead to incorrect decision-making. Here, again, we can see this effect best in classic literature of tragic inabilities to recognize when the world has changed—an athlete who no longer has the abilities of his youth but can’t quit, the businessman unable to recognize that the new competitors are using different tactics, the generals fighting the last war.24
This is something that casino operators and slot-machine makers understand. Over the past several decades, slot machines have become more complicated, not less.25 In her book Addiction by Design, Natasha Dow Schüll describes slot and video machine gamblers as succumbing to a particularly procedural deficit, getting drawn in to the “flow” of a Procedural system, and describes how slot and video game designers have focused the machines to access that procedural (non-Deliberative) learning system. Video gambling machines include more and more complex decisions and controls, providing more of a “game-like” experience, and are less and less about simply putting money in and pulling the lever. More complex options provide more potential options for the illusion of control to grab hold of.26
In a classic study from 1975, Ellen Langer and Jane Roth trained 90 Stanford undergraduate students to play a coin-flip game.27 Students who guessed correctly early were much more likely to decide that they were correctly predicting the outcomes than students who guessed incorrectly early. That is, a sequence of wins followed by losses led the students to imagine they understood the task and needed to try to figure out what was going wrong, while a series of losses intermixed with wins led the students to think it was random. A sequence of four correct early guesses was enough to lead people to believe they could correctly guess the sequence “even for sophisticated subjects.” This is how the casinos hook people—a quick series of wins leads you to think that you can beat the game, even though (as we’ve seen) WWWWWLLLLL is just as random as WLWLLLWWLW.28
Another potential breakdown is that wins and losses are represented differently in the decision-making system. We’ve already seen how people value wins and losses differently.D We’ve already seen how the differences between punishment and disappointment imply that there must be different signals for wins (gains, rewards) and losses (punishments), because disappointment (lack of delivery of an expected gain) is not a simple punishment, and relief (lack of a delivery of an expected or potential punishment) is not a simple win (see Chapter 4).
An interesting example of this can be found in Parkinson’s disease, which occurs when a subject has lost his or her dopamine cells, particularly those in the dorsal aspects of the dopamine areas of the midbrain called the substantia nigra.30 These are, in fact, the dopamine cells that we’ve seen train up the Procedural system (Chapters 4 and 10). One of the primary treatments for Parkinson’s disease is to increase the amount of dopamine in the system, to allow the surviving dopamine cells to work harder (usually by providing a dopamine precursor such as levodopa).31 This can cause a problem because Parkinson’s patients still have some of their dopamine cells intact, particularly those in more ventral aspects, which seem to be involved in the more cognitive, deliberative, and impulsive (particularly the emotional and evaluative) decision-making systems, leading to impulsivity and manic, emotional problems.32 This is what happens to Leonard in the movie Awakenings.33 Leonard’s Parkinsonianism is cured by levodopa, but over time, the dopamine builds up and Leonard becomes psychotic and his treatment has to be stopped.
If these dopamine cells are involved in learning, then Parkinson’s patients should have trouble learning new things from rewards. Both Michael Frank and Mark Gluck have run separate experiments training patients to learn from rewarding or punishing signals.34 They both find that normal subjects can learn from positive examples (get rewards if they choose the correct thing, get nothing if they choose wrong) and from negative examples (lose nothing if they choose the correct thing, lose something [usually money] if they choose wrong), but Parkinson’s patients have difficulty learning from the positive examples (rewards). They can, however, learn from the negative examples (punishments). If the key to learning is actually dopamine, then one might expect this to reverse under levodopa treatment—with levodopa boosting the overall dopamine levels, one might predict that it would be possible for Parkinson’s patients to learn from positive rewards, but no longer from negative punishments. That is, in fact, exactly what happens.
As noted above, slot machines and other casino games have increased the number of near-misses (I almost won!) that turn a true loss into a sensed win.35 If part of the effect of these drugs is to increase a patient’s sensitivity to rewards while decreasing the sensitivity to losses, one would expect Parkinson’s patients on levodopa and other similar drugs that increase dopamine’s efficacy to become highly susceptible to impulse disorders such as problem gambling, where they learn from the positive examples (I won!) and not from the negative (I lost). That is exactly what happens.36
There have also been descriptions of people “addicted” to a number of behaviors, such as food, sex, porn, shopping, video games, and even surfing the Internet.37;E A number of these behaviors have been shown to release dopamine, which is often referred to (unfortunately in the scientific literature as well as the popular news media) as the key to addiction. (Remember that dopamine is one access point, but not the only one, to addiction.38 See Chapter 18.) Even video games lead to the release of dopamine in player’s brains.39 But, as we’ve seen, this is exactly what should happen in the normal learning system (Chapter 4), which uses dopamine to learn appropriate behavior. The question (the answer to which is not known) is whether that dopamine signal becomes correctly compensated as you learn the game.
For example, is obesity an addiction? There are some data showing that obese people have some of the homeostatic compensation and other dysfunctional dopamine mechanisms seen in long-term drug addicts.40 Some of the data suggest that the problem with obesity is related to the high fat and sugar content of junk food. Remember that sugar may well have some of the misattribution problems we’ve seen in addictive drugs.41 So it is not completely out of the realm of possibilities that obesity is a sign of a form of food addiction, but it is still an open question being discussed in the neuroscientific and medical literatures.
Many of these behaviors are internally driven needs and desires that get out of control. As we saw in our discussion of intrinsic reward functions (Chapter 13), evolution only requires that the behaviors have the correct outcome, not that they be “for the right reason.” There are a number of behaviors that have evolved to be pleasurable for themselves, but were evolved because they were really useful for their consequences in the environment they were evolved in. The classic example is, of course, sex. The reason that sex is necessary evolutionarily is, of course, to produce the next generation. But the reason that our brains want to have sex is because it feels good and it satisfies an internally driven need, not because it produces children. Dogs love to chase things (cars, sticks, squirrels). In the wild, dogs were pack hunters that chased down big game. So dogs evolved to like to chase things. The chase itself became pleasurable. The human male’s propensity for porn is presumably a similar effect. It can be easily seen how this can lead to overselection of these behaviors in ways that can interfere with one’s life.
So do we want to say that one is “addicted” to these behaviors? What about other behaviors, such as shopping or setting fires?42 Are these behaviors one can become addicted to? From the decision-making perspective, the term “addicted” is not the important scientific question. From the decision-making perspective, the scientific question is why the person is continuing to do the behavior, whether that behavior is taking drugs, gambling, or watching porn. Fundamentally, whether or not we should call that person “addicted” is a policy question, which determines whether we should encourage or discourage that behavior, and how much we are willing to commit to that policy. But if we understand how the decision-making process is working to drive the person to continue the behavior, we can understand better what one needs to do to strengthen or weaken, to enable or disable, to encourage or discourage it.
• Constance Holden (2001). “Behavioral” addictions: Do they exist? Science, 294, 980–982.
• Natasha Dow Schüll (2012). Addiction by Design: Machine Gambling in Las Vegas. Princeton, NJ: Princeton University Press.
• Mark G. Dickerson and John O’Connor (2006). Gambling as an Addictive Behavior. Cambridge, MA: Cambridge University Press.
• Willem Albert Wagenaar (1998). Paradoxes of Gambling Behavior. Hillsdale, NJ: Lawrence Erlbaum Associates.