Some of the most profound questions in science are also the least tangible. What does it mean to be sentient? What is the self? When the discussion turns to these imponderables, many minds defer rather than get mired in such muddy issues. Neuroscientist Gerald Edelman dives right in. A physician and cell biologist who won a Nobel Prize for his work on the structure of antibodies, Edelman is now obsessed with the enigma of human consciousness—except he doesn’t see it as a mystery. In Edelman’s grand theory of the mind, consciousness is a biological phenomenon. The developing brain undergoes its own process, similar to natural selection: Neurons proliferate and form connections in infancy; experience weeds out the useless from the useful, molding the adult brain in sync with its environment. Edelman first put this model on paper in the Zurich airport in 1977 as he was killing time waiting for a flight; since then, he’s written eight books on the subject, including most recently Second Nature: Brain Science and Human Knowledge. Edelman is also chair of neurobiology at Scripps Research Institute in San Diego, California, and founder and director of the Neurosciences Institute, a research center dedicated to unconventional “high risk, high payoff” science. In their conversations, he and DISCOVER editor Susan Kruglinski range far into this untamed territory, exploring synthetic consciousness, how to build a mechanical brain, the best way to teach robots—and why we all have rhythm.

You’ve coined the phrase “neural Darwinism.” What is it?

Many cognitive psychologists see the brain as a computer. But every single brain is absolutely individual, both in its development and in the way it encounters the world. Your brain develops depending on your individual history. So, for example, identical twins will not have identical brains because each brain is exposed to different circumstances. It’s very likely that your brain is unique in the history of the universe. Neural Darwinism looks at this enormous variation in the brain at every level, from the biochemistry to anatomy to behavior. [When coming up with the term] I personally borrowed from Darwin, who brought forth a great principle called population thinking, in which you could get species by selection in a population. That means if you had a vast population of animals and each differed, under competition certain variants that were on average fitter than the others would be selected, and their genes would go into the population. A similar principle could be applied to the development and the coordination of [human] brains. You have variant microcircuits in the brain.

Can consciousness be artificially created?

I believe that someday scientists will make a conscious artifact. And of course, there are certain requirements it would have to fulfill before scientists would be satisfied. For example, they might give a conscious artifact an ability to report through some kind of language, and then test it in various ways. They would not tell it what they are testing, and they would continually change the test. If the report corresponds to every changed test, then the scientists involved could be pretty secure in the notion that it is conscious. And of course, when they do, it will not be like us. In my opinion, it wouldn’t be alive. That might horrify people. How can you possibly have consciousness in something that isn’t alive? There are people who are dualists, who think that to be conscious is to have some kind of special immaterial agency that is outside of science. The soul, floating free—all of that. There might be people who say, “If you make it conscious, you just increase the amount of suffering in this world.” They think that consciousness is what differentiates you or allows you to have a specific set of beliefs and values, et cetera. You have to remind yourself that the body and brain of this artifact will not be a human being. It will have a unique body and brain, and it will be quite different from us. If they ever achieve it, it won’t be living.

Would a conscious artifact have the value of a living thing?

Well, I would hope it would be treated that way. Even if it isn’t a living thing, it’s conscious. If I actually had a conscious artifact, even though it was not living, I’d feel badly about unplugging it. But that’s a personal response.

By proposing the existence of artificial consciousness, aren’t you comparing the human brain to a computer?

No. For example, if you come into this room 10 times, you are not getting an identical set of signals each time, even though the room is relatively stable. Your brain has to be creative about how it integrates the signals coming into it. Computers don’t do that. Our brain is capable of symbolic reference, not just syntax. There’s a neurologist at the University of Milan in Italy named Edoardo Bisiach who’s an expert on a neuropsychological disorder known as anosognosia [a brain injury in which a patient is not aware of the impairment]. A patient with anosognosia often has a stroke in the right side, in the parietal cortex. That patient will have “hemineglect”: He or she cannot pay attention to the left side of the world. Shaves on one side. Draws half a house, not the whole house, et cetera. Bisiach had one patient who had this. He was intelligent. He was verbal. And Bisiach said to him, “Here are two cubes. I’ll put one in your left hand and one in my left hand. You do what I do.” And he went through a motion. And the patient said, “OK, doc. I did it.” Bisiach said, “No, you didn’t.” He said, “Sure, I did.” So Bisiach brought the patient’s hand into his right visual field and said, “Whose hand is this?” And the patient said, “Yours.” Bisiach said, “I can’t have three hands.” And the patient very calmly said, “Doc, it stands to reason if you’ve got three arms, you have to have three hands.” That case is evidence that the brain is not a machine for logic but in fact a construction that does pattern recognition. And it does it by filling in, in ambiguous situations.

Could we make something that has the equivalent consciousness of a mouse?

I would not try to emulate a living species because—here’s the paradox—the thing will actually be nonliving. Isn’t that something? There still might be an ethical problem but, you know...

Unlike an AI device, the Darwin X improves by learning
(with the help from humans such as
researcher Donald B. Hutson).

What does “living” mean to you?

Well, we know what living means. Living is—how shall I say?—the process of copying DNA, self-replication, under natural selection. Anything that’s self-replicating under natural selection is a living system. But this thing isn’t working that way. If they ever achieved it, it doesn’t necessarily have to be living in that sense.

Although if you combine this science with work being done in, for example, synthetic biology...

Oh, yeah. And in fact, who knows? It seems reasonably feasible that, in the future, once neuroscientists learn much more about consciousness and its mechanism, why not imitate it? It would be a transition in the intellectual history of the human race.

What is your definition of consciousness?

William James, the great psychologist and philosopher, said consciousness has the following properties: It is a process, and it involves awareness. It’s what you lose when you fall into a deep, dreamless slumber and what you regain when you wake up. Second of all, it is continuous and changing. Finally, consciousness is modulated or modified by attention, so it’s not exhaustive. People argue about something called qualia, which is a term referring to the qualitative feel of consciousness. What is it like to be a bat? Or what is it like to be you or me? That’s the problem that people have argued about endlessly, because they say, “How can it be that you can get that process—the feeling of being yourself experiencing the world—from a set of squishy neurons?”

Do insects have primary consciousness?

We’ve underestimated insects. Bees, for example, are simply astonishing in their capabilities of discrimination and perceptual categorization. But of course their nervous systems are radically different from ours. They have neurons, but the anatomy is very different. Here at the institute we have shown that insects sleep. And we could even show that certain genes turn off when you go to sleep and turn on when you wake up, both in us and in insects. We can study the genetics of sleep in fruit flies. But if someone asked us to study consciousness in fruit flies, I would have a hard time. The nervous system is radically different. We can’t really discuss what it’s like to be an insect per se. I’d be very surprised if anyone could ever show that they had what we call awareness of the kind that vertebrate mammals have. But you don’t want to underestimate them either.

I’m always surprised when neuroscientists question whether a dog is conscious.

I believe that there’s every indirect indication that a dog is conscious—its anatomy and its nervous system organization are very similar to ours. It sleeps and its eyelids flutter during REM sleep. It acts as if it’s conscious, right? But there are two states of consciousness. One is what I call primary consciousness. The second one is called higher-order consciousness. Humans have both. I believe that about 250 million years ago, a neuronal structure evolved in some animals that allowed for an interaction between perceptual categorization—vision, touch, hearing, et cetera—and memory. At that point an animal could construct a set of discriminations—qualia. They could create a scene in their own mind and make connections with past scenes. At that point primary consciousness sets in. But that animal has no ability to narrate. It cannot construct a tale using long-term memory, even though it has long-term memory affecting its behavior. Then another event occurred much later in hominid evolution that connected conceptual systems, resulting in higher-order consciousness, developing semantics and true language. Chimps have it a little bit. We have it fully because we have true language. Now we can become conscious of being conscious.

The robot can learn to pick up and “taste” blocks . . . and to stay away from the bad-tasting blocks

It’s interesting to think of what you can remove from, say, a human being, and still call it conscious.

That’s cool. That’s very good. I’ll tell you exactly—primitively, but exactly. First, if I remove parts of your cortex, like the visual cortex, you are blind, but you’re still conscious. When I take out auditory cortex, you’re deaf, right? So the cortex is responsible for a good degree of the contents of consciousness. Now, if I took out an awful lot of cortex, there gets to be a point where it’s debatable as to whether you’re conscious or not. For example, there are some people who claim that babies who are born without much cortex, what’s called hydranencephaly, are still conscious because they have their midbrain. It doesn’t seem very likely. If you touch a hot stove, you pull your finger away, and then you become conscious of pain, right? So the problem is this: No one is saying that consciousness is what caused you to instantly pull your finger away. That’s a set of reflexes. But consciousness sure gives you a lesson, doesn’t it? You’re not gonna go near a stove again.

So it’s internalizing stimuli and then remembering them.

That’s it.

But then we have human beings who lose short-term memory, like the famous case of H.M., who could live only in the present and could not remember what happened a minute ago.

Yeah. But he’s not unconscious.

So each thing you can think of removing—memory, emotion, sensory perception—does not itself equal a lack of consciousness. Clearly consciousness is a combination of several things.

Exactly. It’s the combination of all these interactions.

How useful is it to have a complete understanding of how consciousness works?

Well, first of all it gives you a much wider view of how knowledge is acquired. It’s pretty clear that when you create intellectually, you don’t start with an absolutely precise vision of anything. You start with a rather vague metaphorical idea. Right? It’s like when you are a child and you see any animal and you call it a cat until you find out that dogs actually have another name. Through your consciousness of consciousness, you proceed to refine it however you can. But what has gone on in your own brain and its consciousness over your lifetime is extremely history bound. And so that complicates our understanding of the acquisition of knowledge. I think understanding consciousness could make for a more tolerant and lenient picture of how knowledge is acquired, and I think that’s important.

What do you do at the institute in the pursuit of understanding consciousness?

We construct what we call brain-based devices, or BBDs, which I think will be increasingly useful in understanding how the brain works and modeling the brain. But it also may be the beginning of the design of truly intelligent machines.

What exactly is a BBD?

It looks like maybe a robot—R2D2 almost. But it isn’t a robot, because it’s not run by an artificial intelligence [AI] program of logic. It’s run by an artificial brain modeled on the vertebrate or mammalian brain. Where it differs from a real brain aside from being simulated or emulated in a computer is in the number of neurons. Most complex brain-based devices presently have almost a million neurons and maybe up to 10 million or so synapses. [There are at least 100 billion neurons in the human brain.] But what is interesting about BBDs is that they are embedded. They’re in the world and sample the real world that we have. Our BBD called Darwin VII can actually undergo conditioning. It can learn to pick up and “taste” blocks, which have patterns that can be identified as good tasting or bad tasting. It will stay away from the bad-tasting blocks with images of blobs instead of stripes—not pick them up and taste them. It learned that all on its own. Now, why is this important? It’s important because if you try to replicate that input of the environment by simulation, as they would in AI, you really run into a terrible problem because you can’t trace a complete picture of the environment. At the invitation of the Defense Advanced Research Projects Agency, we incorporated a brain of the kind that we were just talking about into a Segway transporter. And we played a match against Carnegie Mellon University, who are certainly experts in the field of artificial intelligence. Their team had a human working with an AI-based Segway, and ours had a human colleague working with a BBD-based Segway. We won five games out of five. And that’s because our device learned. The devices have to learn to pick up a ball, kick it back to the human colleague, learn the colors of its teammates. So our BBD won because of its ability to learn, not just because of executed algorithms.

What has gone on in your own brain and its consciousness over your lifetime is not repeatable, ever. Not even with identical twins. Not even with conjoined twins

Technically, what is the difference between an AI-driven device and one of your BBDs?

In AI it’s algorithmic. You write a series of instructions that are based on conditionals, and you anticipate what the problems might be. AI players make mistakes because you can’t possibly anticipate every possible scenario on a field. Instead of writing algorithms, we have our BBDs play sample games and learn just the way you train your dog to do tricks. And because we don’t have to put in millions of electrodes, as in animal experiments, we can trace all of the neuronal connections and all of the anatomy during an act. It doesn’t look at all like a hi-fi amplifier or a computer. It’s vast populations of neurons, some of which are more causally efficacious than others. We’re busy analyzing the causal change in thousands of neurons, using what you call a back trace. Every 200 milliseconds after the behavior we ask, what was firing? What was connected? How strong was this synapse? We use mathematical techniques to analyze this—time-series analysis—and we can actually see the whole thing converge to an output. Of course we are not working with a real brain, but it’s a hint as to what we might need to do to understand real brains.

It’s hard to comprehend what you’re working with. What is a neuron in your BBD?

The ones I’ve been talking about so far are like neuronal groups, and they usually are the order of 100 equivalent neurons. We look at what are called mean firing-rate models. You average the firing rate of a number of neurons, which is a reflection of synaptic change. We also work with spiking neurons [neurons that are communicating by electrically activating one another through action potentials]. We have a mathematician here, Eugene Izhikevich, who’s famous for having devised algorithms that give you behavior identical to the real thing. If you recorded from the neurons of a living animal and compared that to Eugene’s spikes, you couldn’t tell which one came from the animal and which one came from his algorithm. The responses are exactly like those of neurons.

You just said you use algorithms to simulate the brain, but earlier you said that you do not use an algorithmic approach.

What we’re actually doing is writing a program that simulates the brain. The brain isn’t a computer, but we’re simulating it in a computer. The number of ways you can carve up the world is probably infinite, right? So a Turing machine depends on precise operations based on symbols or code, like 0, 1, 1, 0, 0, et cetera. It’s instructive, it’s unambiguous. If you make an error, you have to correct the code. It can’t tolerate error. It will crash, whereas BBDs learn by making errors.

What are you working on now?

We have a new BBD called Darwin 12. It has legs and wheels and a very complex structure, and it’s able to navigate in unknown circumstances. Each one of its leg wheels has almost 100 different sensors of different kinds. This is the first time we’re fooling around with the body. It can lift its legs up. We will eventually have it climbing stairs. What we want to see is what happens when you have this kind of variation in the phenotype—the actual body shape—and how that affects how the brain works.

What else is going on at the institute?

Ani Patel and John Iversen have been studying rhythm and melody as a kind of royal road to speech. If you ask an individual to tap according to a metronome, and then you stop the metronome, we already know that the person will tap pretty much in sync, the right rhythm. But if you do the same thing with a blinking light—a person tapping his finger in time along with a light flashing in front of him—the person cannot keep the synchrony after the light stops flashing. Ani Patel and Evan Balaban studied what happens when you play a melody—not a song the person is familiar with but a diatonic melody—and record with MEG [magnetoencephalography, the recording of magnetic fields coming from electrical activity in the brain]. The brain converts the pattern of the pitch into a pattern in time. So you can actually see the shape of the melody on a computer screen using MEG. You see the contour of the melody. Humans, by the way, are the only species that can imitate and synchronize to any arbitrary pattern of sound. As far as we know, even birds can’t in the same way.

What does the understanding of auditory rhythm in the brain tell us about humans?

One of the consequences of that study is that music and language are much more closely related than people had previously thought based on clinical studies. One of the mysteries of evolutionary theory is: What is it that developed that gave us rhythm synchronization? We don’t really know. Some schools say it is dance and gesture. Other people, like Steven Mithen, who wrote The Singing Neanderthals, think that singing was a direct communicative line before speech. While I personally doubt that hypothesis, we don’t really know. What we do know is that there’s a big differentiation amongst the modalities like vision and hearing and touch but a deep connection between the hearing system and the motor system, which certainly relates to dance.

Describe your ultimate vision for the institute.

The thing that’s characteristic of the institute is the freedom. The scientists are family, and they do whatever they please. We think that it complements what goes on in more conventional university departments. So while we don’t work directly on disease—although we certainly look at the diseases inside mice and things like that—everything we determine about the normal brain is going to be reflected in the practice of neurology, psychiatry, and psychology. Our ultimate goal is to understand how the human brain works.