Can robotics shed light on the human mind? On evolution? Daniel Dennett -- whose work unites neuroscience, computer science, and evolutionary biology -- has some provocative answers. Is he on to something, or just chasing the zeitgeist?

Back when I was a student of philosophy, in the late sixties, it was customary to divide the discipline into two schools: "analytic" and "continental." Continental philosophers typically built up large edifices of meaning. Analytic philosophers broke large systems down, scrutinizing every brick. Continental philosophers at times overreached, acting as though thought alone were capable of storming the gates of heaven and hell. Analytic philosophers were at times intellectually stingy. If continental philosophy could sound like poetry or music, analytic philosophy seemed anxious to sound like science. Georg Wilhelm Friedrich Hegel, who saw every aspect of history, politics, religion, and art as a moment in the unfolding of the World Spirit, might be nominated to be the standard bearer for continental philosophy. Ludwig Wittgenstein might be picked, on behalf of analytic philosophy, to deflate Hegel's overflowing theses and antitheses. Wittgenstein, after all, is well known for writing, in Tractatus Logico-Philosophicus (1921), "What we cannot speak about we must pass over in silence."

The work of the philosopher Daniel Dennett, who heads the Center for Cognitive Studies at Tufts University, points not only to an array of contemporary issues but also to the continental-analytic fault line that runs through philosophy itself. When I started reading Dennett recently, I thought at first that I was dealing with an analytic philosopher who used artificial intelligence and computer science to pare philosophical problems down to manageable size. That's true as far as it goes -- it's tempting to say that Dennett has never met a robot he didn't like, and that what he likes most about them is that they are philosophical experiments. Instead of arguing interminably about how a mind works, Dennett believes it makes sense to build one, however rudimentary, and set it loose to see what it can do. If you're staging a Turing Test, in order to see if a computer can fool humans into thinking it is truly intelligent, Dennett would be exactly the philosopher to sit on the board of judges -- as he has several times. If you're designing a state-of-the-art robot in order to see how it negotiates the real world (or some subset thereof), Dennett would be the man for your team -- and not surprisingly he does have ties to the artificial intelligence (AI) community, and is invited to go where other philosophers are not encouraged or have no wish to go. He works closely, for example, with Rod Brooks, the head of MIT's AI Lab, on the design of Cog the "cognitive robot."

As posed by Alan Turing, the question of machine intelligence has become a central theme of our time -- and here, as elsewhere, Dennett brings analytic rigor to bear. To the question of whether machines can attain high-order intelligence, Dennett makes this provocative answer: "The best reason for believing that robots might some day become conscious is that we human beings are conscious, and we are a sort of robot ourselves."

This is part of Dennett's campaign to overcome the mind-body split bequeathed to us by Descartes, who identified his existence with his self-consciousness (his Cogito) and believed that the thinking portion of the self was attached almost accidentally to the body. Like many in cognitive science, Dennett wants to show that mind and matter are not necessarily opposed. Mind is not made of different stuff than body -- not if body is understood to be an enormously complex information-processing system of which the brain is a part. And if that is so, it's not so obvious that man-made information-processing machines are incapable of breaking into self-awareness. If you do not acknowledge a divine spark of some sort (whether called the soul, an élan vital, or, to use Dennett's term, a "sky-hook"), how can you be sure that life, cooked up over eons in the laboratory of nature, is different -- fundamentally, rather than by degree of refinement -- from the models produced in an AI lab?

Dennett's refusal to allow a basic distinction between human and machine intelligence has earned him his enemies. He and John Searle, of the University of California, have rumbled up and down the philosophical block for years over the question of whether computers can truly experience themselves (or truly experience anything at all, for that matter, which Searle denies they can do). Dennett has also cited unexpected precursors to his position: at a time when everyone else seems to be bashing Freud, Dennett has nice things to say about him. For Dennett (as for the AI pioneer Marvin Minsky, in The Society of Mind), Freud was ahead of his time in showing how the ego stole its precious self-awareness from the activities of innumerable processes that are anything but self-aware -- in other words, from the unconscious. Freud's unconscious becomes a placeholder for neural networking, massively distributed parallel processing, or some other trick of wiring that will one day allow Cog or one of its kin to be launched mindfully into the world.

Dennett is a skillful writer who has probed mind-body-machine connections in his books Brainstorms: Philosophical Essays on Mind and Psychology (1978), Consciousness Explained (1991), and Brainchildren: Essays on Designing Minds (1998). When I met with him recently in his office at Tufts, he was quick to acknowledge how difficult it is to talk about the mind these days without using metaphors drawn from computer science. In his view, this is just fine. "Taking on new concepts," he said, "new ways of thinking about things, so that you suddenly open up new model spaces to explore -- that's great, but you are also tying your hands when you do that." He continued, "Now, it's very important that you tie your hands. Working under constraint is a necessary condition for really important inventions. All the great art of the Renaissance was done under the constraint that it had to be in the service of Christian iconography. Can you make great art under those circumstances? You sure can. Would they have made greater art if they had been free bohemians instead of coddled slaves of bishops and dukes? No, I don't think so."

"Does that make us coddled slaves of the computer?" I asked.

"Sure," he replied.

If Dennett stopped right there, at the relationship between human (coddled and enslaved or not) and machine, he would be among a select group of thinkers who keep the traffic flowing from brain science to computer science and back again. But Dennett takes the argument a few steps further, arriving at a synthesis whose sheer scope makes me mutter "Hegel" under my breath. Dennett may have begun as an analytic philosopher using AI to clarify problems of philosophy, but when he puts mind and brain in the context of evolution, it seems obvious that he has matured into something else again.

In a footnote in his book Darwin's Dangerous Idea: Evolution and the Meanings of Life (1995), Dennett points out that Charles Babbage (the mathematician and early computer pioneer) and Charles Darwin attended the same London parties, probably chewed the same mutton, and quite possibly discussed some of the notions that later became so hugely influential in evolutionary theory and computer science. The meeting of Darwin and Babbage brought a central idea of Darwin's Dangerous Idea alive for me -- namely, that evolution and computers are driven by similar processes that are familiar, at least in part, to any software engineer. You write small pieces of dumb code that work with other simple pieces of code in order to produce systems of greater complexity, which in turn interact with other complex systems in order to give higher degrees of functionality, and so on, until you wind up with a program that is smart -- or, at least, smart enough to do something that needs doing. Finally, you get operating systems, you get an Internet -- or, depending on your raw materials and the time allotted, you get DNA, mammals, and self-awareness.

Of course, in the case of evolution there is no software engineer; there are only the bits of genetic code worked by natural selection over millennia into myriad forms of life. Darwin's Dangerous Idea argues that evolution is not a feat of pure thought or of magic or of brilliant planning but simply of engineering over time -- except that it's a case of engineering minus any particular engineer, design minus a designer. In a sense, you have a synthesis as broad as any Hegel attempted, but instead of the World Spirit you have dumb processes of evolution that all on their own suffice to bring about animation, self-replication, intelligence, and the rest. Does evolution obey the dictates of some presiding genius? Does it require a bit of divine guidance? No. Is it nevertheless sensible to think of it as possessing an intelligence that can be usefully compared to that of thermostats and minds? Yes.

When I suggested to Dennett that his fusion of Cog, the Cogito, and Darwin involved three of the dominant motifs of our day -- computers, the human brain, and natural selection -- I wanted to use the image of tectonic plates coming together. Dennett put it somewhat differently. He knows he's done something right, because he entertains, as he says, "a vivid sense of the alternative": "I've seen it when people have a theory and it starts to go sour -- they keep having to do ad hoc fixes, plugging one leak after another. I know what that feels like. And I don't feel it. On the contrary, things keep falling into place. Very few leaks to repair."

Some would disagree about the size and significance of the leaks, or about the integrity of the whole enterprise. John Searle would argue that there is a profound difference between computer algorithms, however sophisticated, and human thought -- a difference of kind that Dennett has seen fit to ignore. Stephen Jay Gould, one of Dennett's harsher critics, would say that any comparison between the logic of computers and the contingencies of natural selection is bound to be forced, and any system that relies on such a comparison is doomed to come apart under stress. Dennett, who still calls himself an analytic philosopher, comes under the kind of fire continental philosophers often had to take from their analytic peers -- hold on, slow down, that center does not hold, things are different, more disparate, than they appear.

Disparateness, difference, fragmentation, discontinuity -- these are familiar postmodern verities. In some ways Dennett reinforces them. The self as he pictures it is much more a collection, an ensemble, a neuro-environment- cum-information-system than a unified entity. His writings are full of devices to get under the hood of self-consciousness and examine possible ways a Cogito might be engineered. Thus his interest in Multiple Personality Disorder -- and in the way the brain can generate not just a self but selves. But similarities between Dennett's thinking and postmodernism are of secondary importance. It's the differences that will count. While so much talk has been devoted to postmodernism, the fields of neuroscience, computer science, and evolutionary biology have been gaining explanatory power, increasing their hold on the imagination, and imposing their constraints on our thought. In linking these disciplines and smoothing away rough spots at the joints, Dennett may well be proposing a sort of overrarching system that is suited to the next century. It's a neuro-cyber-Darwinian synthesis that may just mean it's time to break out the guitars and sing, "Roll over Georg Hegel and tell Friedrich Nietzsche the news."

A Conversation with Daniel Dennett

Harvey Blume: It seems computers help realize one of the goals of analytic philosophy -- namely, to produce unambiguous statements. After all, you can't get rid of ambiguity in natural language but you can -- and you have to -- in the algorithms of computer code.

Daniel Dennett: Computers keep you honest in a way that philosophers have been hankering after for a long time. There's no place for hand waving. There's no place for impressionism in creating a computer model or an algorithm. At the same time, there's plenty of room for hand waving when it comes to interpreting that model. Computers force you to get clear about things that it's important to get clear about. AI is really a new and better way of doing certain sorts of philosophy.

HB: When did you begin to see cybernetic processes as similar to the processes of evolution?

DD: Right at the outset. Right at the center of my dissertation I argued that learning had to be evolution in the brain, and there had to be learning algorithms in the brain.

What happened was that over the years I kept underestimating the amount of hostility there was to evolution. I wanted to say: This is obvious. This is like gravity, folks -- what's going on? Reading Richard Dawkins greatly deepened my understanding of evolution, and excited me. It was when I read The Selfish Gene that I became more than casually interested in what was going on in evolutionary biology.

HB: Has Cog the robot lived up to your expectations?

DD: Cog hasn't come through in the last four years the way I'd hoped. It's also true that Cog has not been the primary work of more than three graduate students. Rod [Brook] hasn't been able to devote even a quarter of his time. I haven't been able to devote even 10 percent of my time. Other things have intervened.

Herb Simon once predicted that a computer would beat the human chess champion in a decade. [Marvin] Minsky did too -- a lot of them put their money on it. They were wrong. They were wrong not by an order of magnitude but by a factor of three or four. It didn't take ten years; it took more like thirty years. But that's not bad. The fact is that research went in other directions. Things bogged down, and it took a while. Similarly, Cog is not on its schedule. We're still in the first wave of features. But we're making steady progress on those. I don't think any abyss has opened up in front of us.

HB: Analytic philosophy has always been influenced by science. That seems to be true of your thought as well.

DD: Analytic philosophy certainly aspires to the sorts of objectivity and opportunities for confirmation and refutation that science does. One of the things analytic philosophy always held against various continental schools was that they seemed to be doing something more like verbal ballet.

My view of science is very much an enlightenment view. Aside from minor disagreements, it's pretty close to [E. O.] Wilson's view in Consilience. That's not an accident. We've spent a lot of time talking about it. Much of what is said about science as an objective, progressive, best-ever technology for getting at the truth I simply think is right, and I believe people who think otherwise are deeply mistaken.
I also think the way progress is made in science is greatly abetted by the temporary hampering of the imagination. Schools of scientific thought constrain themselves in ways that, in retrospect, turn out to have been wrong. But that very constraint helped inspire and provoke the imagination.

HB: Can you expand on the role of the algorithm in your thought? It seems to be a unifying idea.

DD: It's one of them. Let me get at it this way: David Hume wrote about complex ideas and impressions. What he really wanted to do was explain what he called the association of ideas -- how one idea brings the next idea along in its train. He wanted to explain the marching order of ideas without having to postulate a director to direct the show. I was trying to explain this and a student said, "Hume's got to get the ideas to think for themselves" -- to which I said, "to think themselves." You've got to get the thinker out of there. If you've still got the thinker in there, you haven't begun working on the mind.

How do you break that regress? Hume tried. Locke tried. Skinner tried. Turing succeeded. It was Turing who figured out how you could get ideas to think themselves. You write down a recipe for how to do some thinking, and give it to a mathematician. He follows the recipe, does the thinking. Turing says, Yeah, but you can leave the mathematician out of it. You can just give the recipe to the machine and eliminate the middleman.

Eliminate the middleman. And the thinking just happens.

Turing shows that if a computer can add, subtract, multiply, and divide, and if it can tell the difference between zero and one, it can do anything. You can take that set of mindless abilities and build them up into structures of indefinite discriminative power, indefinite discerning power, indefinite reflective power. You can make a whole mind; you can solve Hume's problem; you can get ideas to think for themselves on this slender base. That's the idea of an algorithm.

And what would Darwin say? What does it mean to have an evolutionary algorithm? We look out and see all this beauty, all this fabulous design, all this R&D. Darwin showed how all that research and development can be performed by an ultimately mindless, motiveless, mechanical, if not necessarily malignant, process.

HB: I infer from your work that the job of philosophy is to show how various disciplines are similar to one another in a way that people working within those disciplines might not be able to see as clearly. Is that the role of philosophy?

DD: That's one of the roles. Life is short, and life is complicated. People can't do everything they'd like to do. And one of the things people can't do is keep track of how their particular bailiwick, their somewhat blinkered specialization, fits into the larger picture. There are always problems at the interface: How does this fit with that? One of the goals for philosophers is to do this better than other people. It's not the only role, but it's one I take very seriously.

HB: Can you name another role?

DD: In the beginning, it was all philosophy. Aristotle, whether he was doing astronomy, physiology, psychology, physics, chemistry, or mathematics -- it was all the same. It was philosophy. Over the centuries there's been a refinement process: in area after area questions that were initially murky and problematic became clearer. And as soon as that happens, those questions drop out of philosophy and become science. Mathematics, astronomy, physics, chemistry -- they all started out in philosophy, and when they got clear they were kicked out of the nest.

Philosophy is the mother. These are the offspring. We don't have to go back a long way to see traces of this. The eighteenth century is quite early enough to find the distinction between philosophy and physics not being taken very seriously. Psychology is one of the more recent births from philosophy, and we only have to go back to the late nineteenth century to see that.

My sense is that the trajectory of philosophy is to work on very fundamental questions that haven't yet been turned into scientific questions. Once you get really clear about what the questions are, and what would count as an answer, that's science. Philosophy no longer has a role to play. That's why it looks like there's just no progress. The progress leaves the field. If you want to ask if there has been progress in philosophy, I'd say, look around you. We have departments of biology and physics. That's where the progress is. We should be very proud that our discipline has spawned all these others.

HB: Your starting point is analytic philosophy, but you end up with a grand synthesis, don't you?

DD: It's a grand synthesis, but one that has a sort of minimalist starting point. The minimalist starting point is recognizing that you don't have to start with Descartes' division, you don't have to start with a Cogito. You don't have to start with a dichotomy between subjective and objective, between the first-person point of view and the third-person point of view.

Descartes says there's the rule of mind and the rule of matter. This polarity, this division, has many manifestations. In the continental tradition they talk about Geisteswissenschaft and Naturwissenschaft, sciences of the mind and the rest of the sciences. Everywhere you look you see people who like this division. They say there's nothing that biology or physics or chemistry can teach us about literature, ethics, religion, or even psychology. And then there are those who say you can get all of this under one rubric. There have always been people who wanted to be good materialists.

The only thing that's novel about my way of doing it is that I'm showing how the very things the other side holds dear -- minds, selves, intentions -- have an unproblematic but not reduced place in the material world. If you can begin to see what, to take a deliberately extreme example, your thermostat and your mind have in common, and that there's a perspective from which they seem to be instances of an intentional system, then you can see that the whole process of natural selection is also an intentional system.

It turns out to be no accident that biologists find it so appealing to talk about what Mother Nature has in mind. Everybody in AI, everybody in software, talks that way. "The trouble with this operating system is it doesn't realize this, or it thinks it has an extra disk drive." That way of talking is ubiquitous, unselfconscious -- and useful. If the thought police came along and tried to force computer scientists and biologists not to use that language, because it was too fanciful, they would run into fierce resistance.

What I do is just say, Well, let's take that way of talking seriously. Then what happens is that instead of having a Cartesian position that puts minds up there with the spirits and gods, you bring the mind right back into the world. It's a way of looking at certain material things. It has a great unifying effect.

harvey blume, books & authors, 9 december 1998
www.theatlantic.com/unbound/digicult/dc981209.htm