Reality+ : Virtual Worlds and the Problems of Philosophy (9780393635812), page 30
So far, the virtual worlds we’ve created don’t contain digital creatures with anything like the complexity of a human being. In most virtual worlds, the biological human players are by far the most sophisticated creatures, and the digital nonplayer characters seem mindless—few people would consider them conscious. But eventually, there will be simulated worlds with much more sophisticated nonplayer characters who have virtual brains as complex as ours. Once there are worlds like this, the question of digital consciousness cannot be avoided.
The question matters even more when we think about mind uploading: the attempt to transfer our minds from a biological brain to a digital computer. Many people see this as our best hope for achieving a sort of immortality. In the Black Mirror episode “San Junipero,” people approaching the end of their biological life can choose to upload their brains into a digital replica, which is then plugged into a virtual world. The virtual world functions as a sort of heaven where people can live forever.
Mind uploading raises many scientific issues. Some concern the behavior of an uploaded system. Can such a system generate the same sort of intelligent behavior or display the same memories and the same personality traits as the original biological system? Can we really measure a brain well enough to simulate it? Can we simulate networks of neurons perfectly in a digital system?
If these scientific issues can be solved, deeper philosophical issues await. One of the deepest concerns consciousness. For uploading to work as a path to immortality, it’s crucial that the uploaded system be conscious. If the uploaded system is an unconscious zombie system, then uploading won’t be a form of survival; it will amount to destruction, at least where the conscious mind is concerned. Most people would view this sort of zombification as little better than death.
Another deep issue about uploading concerns identity. If I upload to a computer, will the uploaded system be me? Or will it be a wholly new person who behaves just like me, akin to a newly created twin? If I create an uploaded version of myself while leaving the biological original intact, most of us would consider the biological version to be me and the digital copy to be someone new. Why should it be any different in the case in which the biological version is dead and only the digital copy persists?
In effect, uploading raises versions of Picard’s three questions. First is a question about intelligent behavior: Will the upload behave like me? Second is a question about consciousness: Will the upload be conscious? Third is a question about the self: Will the upload be me? We’d have to answer yes to all of these questions for uploading to be a feasible path to survival.
I’ll say something about all of these questions in what follows, but I’ll focus mainly on the question concerning consciousness. This question is especially relevant to assessing the simulation hypothesis: If simulations cannot be conscious, our consciousness rules out the pure simulation hypothesis from the start.
The problem of consciousness
What is consciousness? Consciousness is subjective experience. My consciousness is a sort of multitrack inner movie capturing how my life seems from the first-person point of view.
Consciousness has many components. I have visual experience of colors and shapes. I have auditory experience of music and voices. I have bodily experience of pain and hunger. I have emotional experience of happiness and anger. In my waking hours, I experience a stream of conscious thought—thinking, reasoning, talking to myself. I decide; I act. All this is somehow unified into an encompassing state of consciousness, making up the conscious experience of being me.
Why is there consciousness in the universe? How do physical processes give rise to consciousness? How can there be subjective experience in an objective world? Right now, no one knows the answers to these questions.
I became a philosopher in order to think about the problem of consciousness. My academic background was in mathematics and physics: I did a first degree in mathematics in Australia in the 1980s and got partway through a doctorate at Oxford. I loved those fields because they seemed to address truly fundamental questions. But it gradually came to seem to me that most of the really hard questions had been answered, and the fundamentals were pretty well understood (or at least so I supposed at the time). Perhaps that was wrong, but that was how things seemed to me then.
At the same time, there seemed to be a truly fundamental problem as yet wide open: the problem of consciousness. The human mind seemed to pose many of the harder questions left for science, and of these, consciousness seemed the hardest of all. Consciousness was the most familiar thing in the world—but also the thing we understood the least. How did it fit into the physical world? How could there be subjective experience in an objective world? No one knew.
I became obsessed by these questions, to the point where I decided to leave mathematics and work on the problem of consciousness directly. I moved in 1989 from Oxford to Indiana University to work in the cognitive science group headed by Douglas Hofstadter—the author of Gödel, Escher, Bach and other books I loved. I learned a lot of cognitive science and did a lot of work in artificial intelligence, but consciousness remained my driving passion. It seemed to me that the best way to tackle the biggest problems about consciousness directly was through philosophy. So I became a philosopher and ended up writing a PhD thesis on consciousness, which turned into my first book, The Conscious Mind.
Around that time, in April 1994, I gave a talk at one of the first international conferences on consciousness, in Tucson, Arizona. In the talk, I called the problem of explaining consciousness the hard problem. This name caught on more quickly than anything else I’ve ever said. People have written books on “the hard problem.” The playwright Tom Stoppard wrote a play about consciousness called The Hard Problem. None of this was because the idea was radical or original—in fact, quite the opposite. The name caught on so fast because everyone knew what the hard problem was all along. The label just captured the problem and made it more difficult to avoid.
To explain the hard problem, and the contrast with other easier problems, it’s useful to start by examining the relationship between two of Picard’s questions: consciousness and intelligence.
What is intelligence? To a first approximation, intelligence is sophisticated and flexible goal-directed behavior. If a system is good at only one sort of goal—winning a game of chess, say—then at best it’s a narrow intelligence. If a system can make reasonable attempts at achieving a wide range of goals, we call it a general intelligence.
So far, many existing digital systems exhibit a narrow intelligence. The AlphaZero program developed by DeepMind is good at winning games of chess and Go. Self-driving cars are good at navigation. No existing digital system is yet close to general intelligence, though. The only generally intelligent beings we know of are humans and perhaps some other animals.
As I understand intelligence, it’s an objective feature of a system that mostly comes down to behavior. Intelligence isn’t a matter of how a system feels. What matters are the objective processes in the system and the behavior they produce.
As a result, we have a much better grip on intelligence than on consciousness. There are standard methods for explaining the behaviors of a cognitive system. To explain a behavior, you need to identify a mechanism and show how it produces the behavior. The mechanism might be a system in the brain, or some sort of algorithm that we think the brain is using. That’s why I call the problems of explaining intelligence, and explaining behavior in general, the easy problems. They include: How do we navigate? How do we communicate? How do we discriminate objects in our environment? How do we control our behavior to achieve our goals? The easy problems aren’t really easy; some may take a century or more to solve. But at least we have an idea of how to go about solving them.
While intelligence is a matter of objective behavior, consciousness is a matter of subjective experience. The hard problem is the problem of explaining subjective experience. All conscious experience seems to be tied to a conscious subject who is having the experience. It’s partly this subjectivity that makes the problem of consciousness so hard.
My New York University colleague Thomas Nagel famously defined consciousness as what it is like to be a system. There’s something it is like to be me, or to be you. If so, you and I are conscious. Most people think there’s nothing it’s like to be a rock; a rock has no subjective experience. If they’re right, a rock is not conscious. If there’s something it’s like to be a bat, as Nagel suggested, a bat is conscious. If there’s nothing it’s like to be a worm, a worm is not conscious.
Many people think of consciousness as something very complicated, at the top of the hierarchy of intelligence. Some think it requires a complex form of self-consciousness. For example, on the TV series Westworld, consciousness is depicted as an awareness of one’s inner voice as belonging to oneself. According to that view, only humans or other reflective beings are conscious. I think this is the wrong way to think about consciousness. Consciousness is manifested even in a simple state, like seeing red or feeling pain. To invoke Nagel again, there’s something it is like to see red, and something it is like to feel pain, so these are conscious states. These states don’t require an inner voice or reflective awareness of oneself. Certainly, an inner voice is an aspect of consciousness in those who have one, and so is reflective self-awareness. But these shouldn’t be confused with consciousness in general.
Even simple conscious states such as seeing red or feeling pain raise the hard problem of consciousness. When my visual system processes a stimulus in a way that leads me to identify it as red, why do I have a conscious experience of redness? Why is there something it is like to see red? The objective methods that work so well on the easy problems don’t work so well on subjective experience. Identifying a brain mechanism that leads us to classify the stimulus as red doesn’t tell us why we have a conscious experience of redness. More generally, explaining behavior doesn’t explain why the behavior is accompanied by consciousness. In any description of brain processes, there seems to be a gap between that story and consciousness. Why should brain processes give rise to conscious experience? Why don’t they just go on “in the dark” without any subjective experience at all? No one knows.
Figure 37 Does Mary the color scientist (here modeled on Mary Whiton Calkins, a leading philosopher and psychologist of the early 20th century) know what it’s like to see red?
The standard methods of neuroscience and cognitive science are directed at explaining behaviors, so they don’t give us much of a grip where the hard problem of consciousness is concerned. At best, they give us correlations between brain processes and consciousness. Neuroscientists are gradually making progress toward what they call the “neural correlates of consciousness.” But correlation is not explanation. So far, we have no explanation of why and how these processes give rise to consciousness itself.
We can bring out the distinctive problem of consciousness with a thought experiment constructed by the Australian philosopher Frank Jackson. Mary is a neuroscientist who knows everything there is to know about physical processes in the brain and how they respond to colors. However, Mary has spent her whole life in a black-and-white room, studying the world through books, black-and-white screens, and other equipment. She has never experienced color herself. Mary knows the full objective story about how red or blue or green things produce certain wavelengths of light, how these affect the eye and the brain, how they produce associations in people, and how they lead to reports like “That barn is red.” However, there’s one central thing about colors that Mary doesn’t know: She doesn’t know what it’s like to experience red, blue, and other colors.
Physical knowledge of the brain tells Mary all sorts of things about color, but it doesn’t tell her about the conscious experience of color. So knowledge of conscious experiences seems to go beyond knowledge of brain processes. This doesn’t yet tell us what consciousness is, but it brings out why there’s a problem.
I once met someone who was just like Mary. Knut Nordby was a Norwegian neuroscientist with achromatopsia, or total color-blindness. The cone cells in his retina, which process color, didn’t work. Despite this, Nordby specialized in psychophysics (the study of sensory processes) and had published many articles on colors. He knew all about the brain systems involved in color processing. When I met him in 1998, he was having his brain scanned and stimulated by the Stanford cognitive neuroscientist Brian Wandell to see if he might be able to experience colors that way. Alas, the experiment was unsuccessful. “The world of colors will forever remain a mystery to me,” Nordby told me.
In The Conscious Mind, I argued that no explanation of consciousness in purely physical terms is possible. The basic idea was that physical explanations are great at explaining behavior, but that ultimately they only explain behavior. More precisely, physical explanations are always a matter of objective structure and dynamics, and all they explain is further objective structure and dynamics. This is perfect for solving the easy problems, but it cannot solve the hard problem. To solve the hard problem, you need something more.
I further argued that if you can’t explain consciousness in terms of existing fundamental properties (space, time, mass, and so on) and existing fundamental physical laws, then you need new fundamental properties in nature. Perhaps consciousness is itself fundamental. You also need to acknowledge further fundamental laws—perhaps laws connecting physical processes to consciousness. The search for a science of consciousness will, in effect, be the search for these fundamental laws.
Since then, there has been an explosion of proposals to solve the hard problem. Some involve new fundamental laws connecting physical processes and consciousness. Some theories connect consciousness to information processing; others connect it to quantum mechanics. An especially popular idea in recent years has been panpsychism, the idea that there’s some element of consciousness in every physical system throughout nature. Other views are more reductionist, trying to deflate the hard problem so it can be solved in physical terms. Perhaps the most extreme version of this strategy is illusionism, the idea that consciousness itself is an illusion, and that for some reason our evolutionary history makes us believe we have the special properties of consciousness when in fact we do not. If that’s right, there’s no consciousness and no hard problem of explaining it.
There’s much more to say about the hard problem of explaining consciousness, but let’s set it aside for now and move to a somewhat narrower question: Can machines be conscious?
The problem of other minds
It’s hard to know for sure whether silicon machines can be conscious. One reason is that it’s hard to know for sure whether any entity other than oneself is conscious. Through my own subjective experience, I’m confident that I am conscious. Descartes’s cogito for consciousness says that I am conscious, therefore I am. But this tells me only about one case of consciousness. It doesn’t tell me about anyone else.
This is what philosophers call the problem of other minds. How can we know that anyone else has a mind? And how can we know what their minds are like? This is a skeptical challenge on a par with the skeptical challenge about the external world. As with the external world, almost all of us believe that other people do have minds and we think we sometimes know what they are thinking and feeling. But how can we know this for sure?
Figure 38 The problem of other minds: Zhuangzi, Huizi, and the happy fish.
A simple version of the problem of other minds arises for non-human animals. In a famous parable, Zhuangzi observes some jumping fish and says they are happy. His companion Huizi says “You’re not a fish. How do you know what makes fish happy?” Zhuangzi replies “You’re not me. How do you know what I know?” Huizi replies that we can’t know in either case, while Zhuangzi is more optimistic. The parable has been used to make many points, but at a basic level it wonderfully illustrates the problem of other minds: How can we know what is going on in the minds of other animals and of other people? Or as Thomas Nagel puts it, how can we know what it’s like to be a bat or to be another person? (Figure 38 interprets the parable along these lines.)
The core of the problem of other minds is the problem of other consciousnesses. Perhaps I can know that other people perceive, remember, and act—if I understand these capacities as independent of consciousness. But consciousness seems to be private and subjective, which makes it very hard to observe in others. Your behavior may suggest to me that you’re conscious, and you might even tell me you’re conscious—but how strong is this evidence? Couldn’t an unconscious robot do the same?
We can illustrate the problem of other minds by asking: How do we know that other people aren’t zombies? As we’ve seen, a philosophical zombie isn’t the Hollywood version that’s risen from the dead. Instead, it’s a being that looks and behaves just like an ordinary human being but is not conscious at all. For a philosophical zombie, everything is dark inside. The extreme case of a zombie is a complete physical duplicate of a conscious human being, with the same brain structures but no subjective experience.
Few people think that zombies actually exist. Almost all of us believe that other people are conscious. But the very idea of zombies is enough to raise the problem of other minds. I can at least imagine that someone else is a zombie—behaving normally with a normal brain, but not in any way conscious. As far as I can tell, there’s no contradiction in the idea that there could be a physical structure that’s atom-for-atom identical to Donald Trump and is not conscious. Again, most of us find the zombie hypothesis fanciful and implausible. But as with skeptical challenges in general, the challenge here is, “How can we know for sure?”
