It is difficult to argue, with conviction, that digital minds will never exist, unless you are a firm believer in dualism. In fact, even though today we don’t have the technologies necessary to create digital minds, it is difficult to believe that such technologies will never be developed. I will not try to predict when such technologies will be developed, but in principle there is no reason to doubt that some sort of sophisticated digital minds will become available during the twenty-first century.
If a system is recognized as a digital mind, the thorny problem of civil rights will have to be addressed. We take for granted that existing legislation recognizes each human being as having a special status, the status of a person. Personhood, with the rights and responsibilities that come with it, is nowadays granted virtually to every living human in almost every society. However, the notions of personhood are not universal. Furthermore, the universal recognition that every human being, independently of race, sex, or status, has the right to personhood is historically recent. Slaves weren’t recognized as persons before the abolition of slavery, a lengthy and slow process that took many centuries. In the United States, the nineteenth amendment to the Constitution, not ratified until 1920, granted women the right to vote in every state.
The concept of personhood varies somewhat with the legal system of the country. In the United States, as in many other countries, personhood is recognized by law because a person has rights and responsibilities. The person is the legal subject or substance of which the rights and responsibilities are attributes. A human being is called a “natural person,” but the status of personhood can also be given to firms, associations, corporations, and consortia.
It seems reasonable to assume that discussions about the personhood status of digital minds will begin as soon as rights and responsibilities are attributed to them. Digital agents—software programs that take significant decisions on behalf of individuals and corporations—already exist and are responsible for many events affecting human lives. The most visible of them—automated trading agents, whose behavior is based on complex sets of rules and algorithms—affect the economy enormously and have been blamed for a number of economic crises and other troublesome events. Until now, however, the individuals or corporations that own those agents, rather than the agents themselves, have been held responsible for the consequences of their actions. This is simply a consequence of the fact that personhood has not, so far, been granted to any purely digital entity.
However, as digital agents and their behavior become more complex, it is natural to shift the responsibilities from the “owners” of these agents to the agents themselves. In fact, the concept of ownership of an intelligent agent may one day become a focus of dissension if digital minds are granted full personhood. The whole process will take a long time and will generate much discussion along the way, but eventually society will have to recognize digital minds as autonomous persons with their own motivations, goals, rights, and responsibilities. The question whether a digital mind has personhood will be unavoidable if the digital mind in question was created by means of mind uploading, a technology that may become feasible before the end of the century. Society will have to address the personhood status of such digital minds, providing them with some set of civil rights and responsibilities. I will call these entities simply digital persons, and will apply the term to any kind of digital mind that is granted some sort of personhood status. To simplify the writing, I will use the pronoun it to refer to digital persons, but this doesn’t mean these persons are less human than actual humans, nor does it mean they do not have gender. Whether or not they will view themselves as male, female, or neither will depend on many factors, which I will not address here because so many variables are involved. I could as well have decided to address digital persons as him, her, or it, depending on the particular circumstances, but that would have been more awkward to write and to read.
The most central issue is probably related to the rights and responsibilities of digital persons. Suppose a program passes a full-fledged Turing Test by using advanced voice and visual interfaces to convince judges systematically, permanently, and repeatedly that it behaves like a human. To achieve that, a program would have to display the reasoning, memory, learning abilities, and even emotions that a human being would display under similar conditions. This program, a digital person, would have acquired much of its knowledge by means of techniques that enabled it to improve its behavior and to become more human-like as time passed. It is hard to imagine that a program could pass a full-fledged Turing Test without having some form of consciousness, but for now we can assume that it could simply be a very, very human-like zombie.
Digital persons might play many important roles in society. They might, in principle at least, be able to produce any intellectual work that could be produced by a human. They might drive cars, fly airplanes, write news, or service customers. In fact, today many of those tasks are routinely performed by programs. Robot journalists, working for Associated Press and other agencies, write many of the stories released today, autonomous cars are in the news every day, and automatic piloting of airplanes has been in use for many years. Virtual assistants that can schedule meetings on your behalf are offered on the Web by a number of companies, including Clara Labs and x.ai. A virtual assistant, such as Siri or Cortana, is included in the operating system of almost every new cellular phone and can be used by anyone, even though virtual assistants still have many shortcomings and even though there are many requests that they aren’t capable of processing.
Even without a physical presence outside a computer, a digital person can write letters, buy and sell stocks, make telephone calls, manipulate bank accounts, and perform many other activities characteristic of humans that do not require physical intervention. Therefore, there is a strong economical motivation to develop digital persons and to use them to foster economic activity. In many respects, digital persons will not be at a disadvantage in relation to natural persons. They will be able to launch media campaigns, address the press, appear on TV, and influence public opinion much as a natural person might.
What should be the rights of digital persons? Should they be granted personhood and all that comes with it? Could they be owned by individuals or corporations, like slaves, or should they be masters of their own destiny? Could they own money and properties? Could they vote? Could they be terminated—erased from memory—if deemed no longer useful, perhaps because a new model or an upgrade, more intelligent and flexible, has become available?
The questions are not any easier if one thinks about the responsibilities of digital persons. Suppose a digital person does something illegal. Should it take the blame and the responsibility, and go to jail, whatever that may mean in the specific context? Or should the blame be attributed to the programmers who, perhaps many years ago, wrote the routines that enabled the digital person to acquire the knowledge and the personality that led to its current behavior? Should a digital person be sentenced to death if, by accident or by intent, it has caused the death of other persons, natural or otherwise? At the time of this writing, the US National Transportation Safety Board is analyzing the conditions under which a Model S Tesla automobile operating in its Autopilot mode failed to brake before impact with a white truck it hadn’t detected, thereby causing the death of its human occupant. Other similar cases are likely to follow as the technologies become more widespread.
If a program passes a full-fledged Turing Test, convincing everyone that it reasons as a human does, should the prize be given to the programmers who created it, or to the program? The programmers would probably be glad to receive the prize, but the program might disagree with this option. After all, if a program is human enough to pass a Turing Test, is it not human enough to keep a share of the prize?
Although these questions seem outlandish, they are not idle musings about hypothetical entities that will never exist. Intelligent agents—digital minds—are likely to exist sooner or later, and, if they do, discussions about whether or not they are digital persons will be unavoidable. If true digital persons should come into existence, we can be sure that the challenges that will be posed to society will be formidably more complex than the ones that are at the center of our most contentious issues today.
One major difference between the digital world and the physical world is that digital entities, which exist only in computer memories, can easily be copied, stored, and retrieved. In the physical world, people place significant importance on the distinction between originals and copies. A bank teller will willingly cash a signed check, but will not be happy if presented with two copies of the same check, or even with two different checks of which the second bears a scanned copy of the signature on the first. A restaurant will treat you very differently if you pay for dinner with a $50 bill than if you pay with a copy (even a good copy) of a $50 bill, and there is a huge difference in value between a painting by Picasso and a copy of it.
This state of affairs is consistent with the inherent difficulty of perfect duplication of objects in the physical world. Until very recently, exact duplication of physical objects was either impossible or very hard, and the way society works rests on the assumption that perfect copies of physical objects are difficult to obtain. Forgeries are possible, but are usually detectable and almost always illegal. Technologies for handling currency, signatures, fingerprints, and all other means of personal identification are based on the fact that exact copies of physical objects are hard to create. In particular, it is very difficult to make exact copies of human bodies, or of parts of human bodies, such as fingers or retinas, and thus biometric methods can be used to verify the identity of a natural person with almost absolute certainty. In fact, one of the most common means of identification—a handwritten signature—is based on the difficulty of reproducing the sequence of movements that a human hand makes when producing a physical signature.
But this situation, in which authentication and identification mechanisms are based on the difficulty of copying physical objects, is changing rapidly, for two reasons. The first reason is that duplicating physical objects exactly is not as difficult as it once was. The second and more important reason is that in the digital realm duplication is easy.
Technologies already under development will enable us, in the near future, to perform high-precision scans of physical objects and to assemble (perhaps with the use of nanotechnologies) any number of copies, using so-called universal assemblers. Nanotechnologies may one day make it possible to assemble objects, atom by atom, from specifications stored in digital form. The practical challenges such technologies will bring to our physical world are interesting; however, they are not at the center of the present discussion, if only because it is likely that similar and much more complex problems will show up in the digital world well before universal assemblers become available.
In the digital world, exact copies are easy to obtain. We are all familiar with the challenges that existing digital technologies create today. Software is easily duplicated, as are all sorts of media stored in digital format, such as books, music, videos, and photographs. All the industries based on creating, copying, and distributing media content, including the newspaper, book, movie, and music industries, are facing the need to adapt to a world in which exact copies of a digital object are easy to make, obtain, and distribute widely. Some business models will adapt and some will not, but the changes in these industries are so profound that very little of what was in place just ten years ago will survive, in its present form, for another ten years. Newspapers, booksellers, movie studios, songwriters, and singers are having to deal with profound changes in the way people produce and consume content.
In the digital world, there is no difference between an original and a copy. The very essence of digital entities makes it irrelevant whether some instance is an original or a copy. In some cases—for instance, when you take a digital picture and store it in the camera card—it is possible to identify exactly where the sequence of bits that constitutes a digital entity was first assembled. In other cases, it is more difficult or even impossible to identify where a digital entity came in existence—for instance, when you transmit a voice message by phone and it is recorded somewhere in the cloud, to be reproduced later, after having been sent through the air in many dispersed packets. But in none of these cases is the “original” sequence of bits distinguishable from copies made of it. The copy of a digital picture you store in your computer is as good as the version first recorded in the camera card, and any copies that you send to your friends are not only of the same quality, but, in a sense, as original as the one stored in the camera card. In these considerations I am ignoring the accumulation of errors in digital entities. These errors are rare and can be easily corrected.
The possibility of duplicating digital persons raises a plethora of complex philosophical questions. In fact, one may harbor genuine doubts about the eventual feasibility and existence of digital minds. However, it is completely safe to assume that, if digital minds ever come into existence, it will be practicable to back them up, store them for future retrieval, and, in general, copy them exactly an arbitrary number of times. Independently of the exact techniques used to create and develop digital minds, the underlying computational support will probably be based on standard digital storage, which by its very nature allows exact copying an arbitrary number of times.
Dualists will always be able to argue that the essence of a mind cannot be copied, even if the digital support of the system that generates the mind can. Believers in the theory that quantum phenomena play a significant role in the emergence of consciousness will take solace in the fact that the quantum configuration of a system cannot be copied without interfering with the original, whereas a digital entity can be copied without disturbing the configuration of the original. However, there is no evidence that any of these beliefs holds true. It is highly likely that digital minds, if they ever come into existence, will be trivially easy to copy and duplicate. It is hard to grasp even a partial set of the implications of this possibility, since many scenarios that are not possible in the physical world become not only possible but common in the digital world.
Consider the possibility of a direct copy of a digital person. Now imagine that a technology that makes non-destructive mind uploading possible will be developed, sometime in this century or the next, and that an uploaded mind obtains the status of a digital person.
First, it will be necessary to deal with the question of how to treat the digital person and the biological human who served as the template for the uploading. Immediately after duplication, the behavior of the two persons—the digital one and the biological one—will be identical or at least very similar. The copy, which will live in some sort of virtual environment, will want to have access to some resources that will enable it to carry on with its life. Maybe it will want money to buy computer time to run the emulation, or to pay for other resources, physical or digital. The biological human, having (presumably) agreed to be scanned and copied into the memory of a computer, may view this desire with some sympathy and may even be willing to share his or her resources with the digital version. Both are exactly the same person, with the same memories, the same feelings, and the same history, up to the moment of duplication, at which time they will begin to diverge. Divergence will occur slowly and accumulate over time. One can imagine that duplication, in this case, can be similar either to a self-divorce, with the possessions divided between the original and the copy, or to a marriage, with the possessions jointly owned by the original and the copy.
Maybe non-destructive mind uploading will be legally possible only after a suitable binding contract is established between the two future copies of the person who will be uploaded. There is, of course, the small difficulty that these two copies do not exist at the moment the upload operation is initiated; however, that isn’t a serious concern, because both copies will be indistinguishable from the original. Immediately after the uploading operation, both copies will have a common past but divergent futures. This contract could then be written and registered, before the uploading, in effect binding both future versions of the person signing the contract. Such a contract, signed by the natural person before the uploading, would enforce the future distribution of assets, and would presumably be binding for the two (identical) persons after the duplication, since their past selves agreed and signed. It is difficult to anticipate whether or not such a mechanism would work, but it seems to be the most logical way to ensure that a discussion about the rights and assets of the two entities doesn’t ensue. One may counter that such a contract may be deemed unacceptable by one (or both) of the future persons, but legal mechanisms may be developed to ensure the right framework for this type of procedure, that was, after all, agreed to by both parties.
It is also possible that mind uploading, should it come to pass, will have to use a destructive technology. Owing to the limitations of scanning technology, it may be necessary to destroy the original brain in order to obtain a detailed copy that then can be emulated. This case would be less complex than the preceding one; there wouldn’t really be duplication, and deciding who owns the assets would presumably be simpler. Heirs might claim that their predecessor is now dead; barring that possibility, however, it is arguable that the uploaded digital person should retain (or perhaps inherit) the rights and assets of the original biological person. In writing this, I am assuming the legal rights and responsibilities of digital persons would have been established by the time the technology became feasible.
Once a digital mind exists, obtained either by uploading or by some other method, duplication of digital persons will be easy. Duplication and instantiation of a new copy of a digital person is a fairly straightforward process; it would be possible even with present-day technology if a computer-executable version of a digital person could be provided. The challenges are similar to the challenges with mind uploading, but in this case it makes no sense to assume the original is destroyed in the process. Again, a legal contract, binding both instances of the digital person, may be required before duplication takes place. The difference between duplicating a digital mind and uploading a biological mind is that in the former case there exists a complete symmetry, since neither of the two copies of the digital mind can claim to be more original than the other. In the case of mind uploading, the biological person who was the basis for the copy may have a case that he or she is the original, and should be allowed to decide about the distribution of assets and even about termination of the copy.
The questions are not much simpler if, instead of mind uploading, one considers the other possibilities for mind design. If digital persons have rights, handling duplication processes will always be tricky. The essential difficulty stems from one crucial difference: There is no branching in the line that takes one biological person from birth to death, so the definition of personhood stays with the same individual from the moment of birth until the moment of death. Even though the atoms in a human being’s body are constantly being replaced by new ones, and even though most of a human being’s cells are replaced every few years, a human being retains a personality that is essentially unchanged, and there is no doubt, to society at least, that he or she is the same person he or she was a few years ago. Someone who committed a crime and who later argues in self-defense that he is now a different person would not get easily away from his responsibilities. The question becomes different, however, if one considers digital minds, because in that case the flow of time is less predictable—in fact, it is possible to make time go backward or forward in a limited but important sense.
The possibility of time travel has been a subject of many novels, movies, and science-fiction plots. Although some people still believe time travel may one day, by some mechanism, become possible, we are stuck with the fact that, except for relativistic effects, the only way we can travel through time is at the normal rate of one second per second. This is to say that, in the world we know and experience, what is possible is to travel forward in time at exactly the same speed as everyone else and the same speed as the rest of the physical world that surrounds us.
The theory of relativity predicts that, when an object moves in relation to another object, time will slow down for both objects—an effect that will be stronger when the objects are moving at very high speeds relative to one another. This time dilation has been observed and measured many times and, in fact, leads to different rates between global positioning satellites (GPS) clocks and Earth clocks, a difference that requires correction in order to make the GPS system work. By the same token, an astronaut traveling in a spaceship at a speed close to that of light might see many centuries elapse in the outside world during his or her lifetime. Barring such a mechanism for time travel (which is still beyond the reach of science and technology), we are stuck with the fact that traveling in time is not a reality in our daily lives in the physical world. Things are a bit different when one considers the possibilities open to digital persons whose lives and minds are emulated in a computer.
The emulation of a digital mind need not proceed at the usual pace of physical time. Depending only on how much computational power is available, one may decide to run the emulation of a virtual world in real time, or at a speed faster than real time, or at a speed slower than real time. To facilitate the connection with the real world, it is reasonable to assume that emulation of digital minds will usually be run at the same speed as real time. This would enable an easier interaction with the physical world, something that probably will be valuable for a long time.
For a digital person (particularly one who has experienced the physical world), it would be strange to observe time pass, in the physical world, at a speed very different from subjective personal time. Such a difference in speeds, however, may be necessary, or even desirable, for a number of reasons. One may imagine, for instance, that lack of computational resources will require the emulation to be run at 1/100 of real time. In such a case, for each subjective day elapsed, an entire season would elapse in the physical world. In a week, the digital person would see two years go by. Conversations with natural persons in the physical world, and other interactions too, would have to be slowed down to a rate that would make it possible for both sides to communicate. Other phenomena, too, would feel very awkward for the digital person. Subjectively, the digital person would be traveling forward in time at the rate of 100 seconds for each elapsed second. This would be sufficient to observe children, in the outside world, grow visibly every day and friends getting old at an uncommonly high speed.
The opposite might also be true. If sufficient computer power were available, digital minds could be emulated at a rate faster than real time, in effect slowing down external time from the point of view of the digital person. In an emulation ran 100 times the speed of real time, the digital person would see two years of his or her life go by while in the physical world only a week elapsed. Again, conversations with external persons, and other interactions, would be awkward and difficult, requiring long pauses between sentences so that the biological persons in the outside world would be able to process what was said by the digital person, who would be talking at a much higher speed.
More radical forms of forward time travel are also possible. For instance, a digital person might decide to jump forward in time ten years instantaneously. All that would be required would be stopping the emulation for the specified period of time, which would require no more than the flip of a switch. For the digital person, no time would have elapsed at all when, ten years later, the emulation was restarted. In the blink of an eye, ten years would have gone by in the outside world, from the subjective point of view of the digital person whose emulation was stopped.
The existence of digital minds and digital worlds also make possible a somewhat limited version of time travel to the past. I have already alluded to the fact that digital minds, as any other digital entities, can be copied at will. This means that it is possible to copy the status of the digital mind to some storage medium at any given time and to recover it later—perhaps many years later. This rather simple action makes it possible to perform regular backups of a digital person and to recover, at any time in the future, the digital person exactly in the same state it was when the backup was made. In this way, it becomes possible to bring back anyone from any time in the past, if a backup from that particular time is available.
As anyone familiar with computer technology knows, it is quite feasible to perform regular and inexpensive backups of a computer by means of a technique called incremental backups. More sophisticated than normal backups, incremental backups store only the changes from the previous backup and can be used to recover the state of a computer in any specific point in time. If something like this technology is used to guarantee the recovery of a digital person from any point in time in the past, then it is not only possible but easy to go back in time, from the subjective point of view of the digital person.
This possibility has a number of interesting consequences. First, the actual death of a digital person will never happen unless it is specifically desired, because it is always possible to go back in time and reinstate a younger version of the person from a backup. In Charles Stross’ novel Glasshouse, fatal duels are common occurrences, culturally and socially accepted, because the duelists always have the possibility of being restored from a backup performed before the duel. The worst thing that can happen is to lose the memories from the time between the last backup and the duel—something that is seen as inconvenient but not dramatic.
Other possibilities now thought of as in the realm of science fiction become possible. One of these possibilities is for a digital person to have a conversation with a younger or older version of itself. For instance, if a digital person desires to let a younger version of itself know a few useful facts about its future life, all it need do is reinstate the younger version from a backup and arrange for a private chat with it, perhaps over a few pints of beer. From then on, things get a bit more confusing. If the younger version is to be able to make use of that information, it will have to go on with life in the possession of this valuable piece of advice from a uniquely informed source. But then there will be two copies of the same person in existence, with different ages. Or perhaps the older person will regret so much the course taken by its life that it will choose to be terminated and let the younger version become the new person, now enlightened with the wisdom received from an older self. In practice, this represents a way to correct decisions made in the past that a digital person regrets later in life. For instance, after a failed romantic relationship, a digital person may decide to terminate itself and to reinstate a previous backup copy, starting with a clean slate. In the movie Eternal Sunshine of the Spotless Mind, a sophisticated procedure is able to selectively delete memories that no longer are wanted. Unlike in the movie, restoring from a previous backup is equivalent to removing all the memories from the intervening period (and not a selected few), between the moment of the backup and the instant the digital person goes back in time.
The questions and challenges raised by these strange possibilities don’t end here; in fact, they are almost endless. Suppose that a digital person with a clean slate commits a serious crime and is sentenced to termination. Should a backup of that person, obtained before the crime, be allowed to be restored? If restored, should it be liable for the crime its copy committed in its subjective future, which is, however, the past for the rest of the world? One should keep in mind that the nature of serious crimes will change significantly if digital persons come to exist. In the presence of backups, murder of digital persons would not be very serious, as long as the murdered person has a relatively recent backup. In fact, murder might be seen as a quite common event, with minor consequences—perhaps the equivalent of a slap in the face in today’s world. And suicide would be ineffective, as a digital person can be easily reinstated from an earlier backup.
All these strange new possibilities, and many I have not addressed, will remain open to discussion in the decades to come, as the technologies needed to create digital persons are developed. However, it will be useful if advances in our understanding of these questions are achieved before we are faced with the actual problems—something that may happen sooner than is now expected, at least in some particular aspects.
At this point, it is worthwhile discussing in some detail how exactly digital minds interact with the physical world, if they do so at all. So far, I have glossed over the practical difficulties involved in such interaction. I have described some of the technical issues related to the emulation of brains, or to the design of other computational agents that may give rise to digital minds, but I have not addressed the important question of interaction with the outside world. This is an important question for uploaded minds and for neuromorphic systems, because a precise and realistic emulation of a brain will require detailed interaction with the physical world. Without such interaction, it will not be practicable to establish the appropriate connections between neurons that create experiences, memories, and emotions.
Obviously, inputs from the outside world could be fed to the optic and auditory nerves, through cameras and microphones, in order to stimulate the senses of vision and audition. Other sensory nerves, responsible for other types of sensations, could also be stimulated to provide a digital mind with the senses of touch, smell, and taste, so as to make life in a virtual world similar to life in the physical world. The engineering challenges that would have to be surmounted to endow a digital person with a complete set of stimuli are probably as complex as the challenges that would have to be surmounted to create a digital mind, but they are essentially technological challenges. Although the basic five senses are the best-known senses, humans have other senses, which provide them with varied inputs from the physical world. Arguably, however, a detailed emulation that includes complete versions of the five senses is already a sufficiently big challenge to deserve discussion.
In the other direction, a digital mind would have to be able to influence the outside world by controlling the muscles that generate speech and movement. Those muscles are controlled by motor neurons, which receive inputs from other neurons (typically in the spinal cord) and transmit signals to the muscles to produce movement.
The technologies required to achieve an effective simulation of this complex process are the subjects of active research, since they will be also useful to help disabled people control prosthetic limbs and for many other purposes. Brain plasticity (the ability of the brain to adapt to new stimuli) will come in handy when a newly created digital mind is provided with inputs from an unfamiliar source, such as a digital camera or a microphone. This plasticity will enable the brain to adapt and to learn how to generate the right patterns of nerve firing. Brain adaptation has been extensively observed and used in experimental settings in which blind patients have had their neural cortices stimulated directly by visual inputs (Dobelle, Mladejovsky, and Girvin 1974) and deaf patients have their cochlea stimulated directly by auditory inputs (Crosby et al. 1985). There is still the challenge of using this information in the (simulated) motor neurons to interact with the outside world and to move the sensorial devices to the places where the action is.
The more obvious possibility is that a digital mind inside a computer will interact with the outside world through a robot. Advanced robotics technologies will be able to reproduce faithfully the intended movements and interactions of a digital mind, much like the artificially intelligent robots in the movie Ex Machina. Humanoid or not, robots will enable digital persons to interact with the physical world. An array of technologies required to enable robots with a complete set of sensors and actuators that will mimic the sensors and actuators humans have will take many decades to develop, but there is no intrinsic technical difficulty that couldn’t be addressed by advances in robotics. Researchers working on brain-machine interfaces (BMIs) have already achieved many significant results, including devices that enable quadriplegic patients to control limbs using only thoughts. Such technologies will one day be used to enable normal humans to control computers, perhaps replacing existing interfaces. They can also be used by digital minds to control robots in the physical world. BMI researchers have also developed technologies that can be used to perform direct stimulation of brain areas in order to create sensations equivalent to touch and other senses. Such technologies could be used to provide digital minds with realistic sensations of immersion in the physical world.
The second possibility is less obvious but significantly more flexible. A digital person could live in a completely simulated digital world in which the inputs to the senses and the actions caused by the muscles were simulated in a virtual environment. Most readers probably are familiar with many types of virtual reality commonly used in computer games. Today’s most advanced videogames already have an impressive level of realism. Videogames that simulate soccer or basketball already achieve levels of realism that are comparable to real-world images. When all the interaction occurs on a computer screen the sensation of immersion is incomplete, but soon more immersive interfaces will come closer to fooling the senses completely.
Virtual reality has been hailed by many as the future way to interact with computers. Virtual-reality glasses, for instance, have been proposed by a number of companies, but so far they have failed to become a significant way to interact with machines. That is due in part to the fact that getting a virtual-reality device to simulate visual inputs accurately is difficult. The brain’s sensitivity to inconsistencies in movements and in rendered images leads to sickness and disorientation. However, as the technology of virtual reality develops, the quality of the interaction has improved. Oculus VR and some other companies now promise virtual-reality experiences that are essentially indistinguishable from real visual experiences. Google’s innovative Cardboard project puts virtual reality in the hands of anyone who has a smartphone. For less than $20, one can acquire a Cardboard kit with which one can, in conjunction with a smartphone application, experience an immersive and very convincing virtual-reality environment.
So far, these immersive interfaces have addressed only vision and audition. Addressing other sensations, such as touch or movement, is much more difficult when dealing with physical bodies, which would have to be moved and/or put into immersive special environments. The actuators and sensors that would be required would be physically large and difficult to implement. This challenge, however, is much easier to address if there is no physical body present, but only a digital mind running in a computer. In that case, the physical sensations of movement, including acceleration, could be caused by simply acting on the right sensory nerves. The resulting sensation would be of complete and realistic immersion.
I believe that, with the advances in image rendering technologies and with a somewhat better understanding of the operation of motor neurons and sensory neurons, highly realistic versions of digital worlds will not only be possible, but will ultimately become the mechanism of choice to enable digital persons to interact with the outside world. Of course, the outside world here is a very particular one, since it will be entirely simulated within a computer. Cities filled with people, grass-covered fields, beaches with breaking waves, and snow-covered mountains could all be simulated to a level of realism that would make them, in practice, indistinguishable from the real world. Interaction with other digital minds would proceed straightforwardly, since those minds would inhabit this virtual world. Interactions with the actual physical world could also exist, although they would require additional devices. The state of the physical world, or of parts of it, could be fed into the virtual simulation by cameras and other sensors, and appropriate renderings of the humans and other agents in the physical world could be performed by the computer running the virtual-world simulation. Actual physical interaction with persons and objects in the outside physical world would remain the limiting factor; this limitation might be addressed by using robots or some sort of mechanism of synchronization between the two worlds.
The concept of a large number of digital minds living in a virtual world and interacting with one another may seem outlandish and far-fetched. However, it may be the most obvious and cost-effective way of creating conditions appropriate for digital minds. The future may, indeed, be something like the world depicted in Greg Egan’s novel Permutation City (2010a), in which the bodies and brains of 10 million people are emulated on a single computer chip.
It is a trivial fact, but still worthwhile noting, that in a virtual world the laws of physics as we know them need not necessarily apply. In a purely virtual environment, people can travel instantaneously from one place to another, anywhere in the universe, as long as the destination can be simulated. In such a virtual world, digital persons can fly over mountains, swim to the bottom of the ocean, or walk on the surface of the Moon with no special equipment. In fact, the possibilities are endless, and I cannot anticipate even a shadow of the virtual realities that will one day be created. Some of these strange possibilities are already familiar to players of virtual-reality games.
In such a virtual world, even going back in time becomes genuinely possible. From inside a computer-generated virtual reality, it must be possible to re-run the simulation from any particular point in time. Of course, some parameter in the simulation would have to be changed; otherwise the simulation would run again in exactly the same way. Perhaps some digital persons would like to run (virtual) reality again, from a specific point in the past, with some added knowledge they didn’t have the first time they went through that point. Since the whole simulated world would have to be re-run, this would correspond to a duplication of (virtual) reality, with a corresponding expense of computational resources and with the implied duplication of all digital entities in the simulated world. That would raise another set of questions that are simply too alien to discuss here.
I cannot anticipate whether digital persons would, in general, prefer to live in a realistic depiction of the outside physical world, or would instead opt for a simulation environment more flexible and free from the laws of physics that have kept humanity limited in so many aspects for millions of years. If such a virtual environment were to be created, we would truly be in the realm of science fiction. I don’t believe we are equipped, at the moment, to think about the issues that would have to be addressed.
So far, we have assumed that digital minds, created by uploading biological minds, would be created by means of a procedure that would ensure that their behavior would be essentially indistinguishable from that of the original minds. Synthetic minds are different, since they would have been designed in accordance with entirely different principles and since they probably would work in a way very different the way the human mind works. Neuromorphic intelligent systems would be somewhere in between. The most obvious way to guarantee that a digital copy of a biological mind behaves like the original is to make the copy as faithful as is possible. We know the technology that would be needed to create such a copy doesn’t yet exist and, indeed, may never exist. However, even if the technology required to copy a mind to a digital support comes into existence, it may not be sufficient to understand and modify the emulated mind.
In the case of uploaded minds, one may imagine that the copying is done by some automatic process that simply identifies brain structures and mechanisms and copies them to the memory of a computer, which will then be able to emulate the processes taking place inside the brain. Understanding how a brain works is a different challenge, perhaps much more complex. The activity patterns in the brain that lead to a particular feeling or memory are probably complex and hard to understand from first principles. Emotions and other brain functions probably are emergent properties of complex firing patterns in the brain that cannot be easily understood, much less changed or improved.
Our understanding of how the brain works may eventually reach a point at which minds can be modified, finely tuned, improved, and changed at the owner’s discretion. Manipulating the brain’s structures and patterns to erase memories selectively or to improve intelligence may be possible. However, such manipulation will require a level of understanding of brain behavior that doesn’t necessarily result automatically from an ability to create digital minds.
There has been significant progress in brain science in recent decades, and such progress is likely to continue at an ever-accelerating pace. But even if it is possible to gain detailed knowledge of the mechanisms that control the inner workings of the brain, we must keep in mind that every human brain is different from every other. Understanding the general mechanisms is not the same thing as understanding the behavior of a specific brain. General chemical imbalances and other large-scale processes that lead to illnesses could be addressed much as they are addressed by today’s medication-based techniques. To achieve that, it would suffice to change some parameters in the emulation corresponding to the environment variables that one wants to change—variables that, in real life, are controlled by drugs, by surgery, or by other macro-scale processes. However, detailed neuron-level surgery capable of wiping out a single memory or a single behavior, if such surgery were to become possible, would require very detailed knowledge of the behavior of every individual brain. This may be much more difficult than obtaining a general understanding of the general mechanisms that underlie brain behavior.
There is one factor that will make it much easier to achieve a detailed understanding of the behavior of specific minds. The behavior of a digital mind can be observed with any required level of detail, whereas with a biological mind instrumentation and physical limitations set hard limits on what can be observed. Even with the advances in instrumentation I have discussed in chapter 9, it is still not possible to obtain detailed information about the firing of individual neurons in a working brain. A digital mind, obtained by emulating a human brain in a computer, would make available for inspection every single firing pattern, every single connection, and every single phenomenon occurring in a working brain. Such information could be used by researchers not only to understand how existing minds work but also, presumably, to learn how to design more powerful minds. This raises the interesting possibility that digital minds might be able to improve themselves, since they would, in principle at least, be able to inspect and understand their own behavior with an unprecedented level of detail.