Tuesday, March 3, 2009

The Mechanization of the Mind

I read a good book over the holidays: The Mechanization of the Mind, by the French philosopher Jean-Pierre Dupuy. (Coming out in paperback at the end of May, if you're wondering what to get me for my birthday. ;-)

It's a history of Cybernetics, the early 20th century intellectual movement that gave rise to contemporary cognitive science (as well as various other fields like information theory, artificial intelligence, and analytic philosophy of mind). Thus it offers a kind of recent pre-history of today's prevalent assumptions about the nature of the mind and the brain.

The most interesting thing I got out of the book was the fact that the computational theory of mind did not, as I had always assumed, arise as a consequence of the invention of the digital computer. That is, it's not the case that we first invented computers, and then started using them as a metaphor or model for the way our own minds work. As it turns out, the modern digital computer and the computational view of mind share a common origin in mathematical logic.

Here's how it all went down, according to Dupuy.

1. Frege et al. invent modern mathematical logic, which attempts to give a purely syntactic (i.e. symbolic, algorithmic, formal, "mechanical") account of logical inference.

2. The power of this new logic, along with the view that logic prescribes the "laws of thought", lead to the claim that thinking just is this sort of formal symbol-manipulation. Logicism is born.

2. Alan Turing, in his paper "On Computable Numbers, With an Application to the Entscheidungsproblem", tries to give a purely syntactic definition of logical inference by describing an imaginary machine that could write and erase symbols on an infinitely long tape, and "remember" the symbols it has recently scanned in a finite "memory". This machine scans up and down the tape, modifying it in response to the symbols it finds.

While the formal inference rules of symbolic logic had been described as "mechanical" before, in the sense of proceeding without understanding or insight, this was the first time anyone had proposed the idea that these formal inferences could be carried out by an actual machine.

In fact, Turing's machine could only be imaginary, since its tape was infinitely long. But it's not hard to see how a similar machine could be constructed with a finite "tape" - and indeed, it wasn't long before John Von Neumann proposed such a machine. And so the computer as we know it was born: a machine whose instructions are stored in its own memory, and so can be modified by the machine's own activity; a machine in which hardware and software can be distinguished.

4. Since Logicism had already identified thinking with the symbol-manipulation of formal logic, two conclusions seemed inevitable:
a) Machines are capable of thought;
b) The human brain is itself -- or at least, can be adequately modelled by -- a Turing/Von Neumann machine, a computer with a very complex program.

Thus, the computational theory of mind was not the consequence, but rather the antecedent of the invention of the modern computer. The idea that thought consisted in the purely formal manipulation of symbols gave rise more or less simultaneously to the idea that a machine might be able to think, and that the human brain must be such a machine.


john said...

Did they really believe 4b?

It seems like the halting problem that Turing outlines in that very paper defines a precise boundary between computational theory and human intelligence.

For those of you interested in subjectivity, I think the Turing Test is pretty interesting. Basically, a machine is intelligent if, in its interactions with another human, it can convince that person into believing that it is intelligent. Which isn't actually all that hard, ahem, given the right context. Often the job of computer science folks (the non theoretical, software types anyway) is to figure out how to limit context or frame a problem so that a computer could in fact behave with a high degree of "intelligence".

Anyway, can you tell I'm Turing fan? :)

Noah said...

I didn't mean to imply that Turing's "On Computable Numbers" paper argues for either 4a or 4b. Turing was just trying to give a theoretically rigorous definition of provability in formal systems.

However, since the view that thinking is a kind of computation was already floating around, the idea that computation could be carried out by a machine had a big impact. This is what I found interesting about Dupuy's history: the view of thinking as computation preceded the invention of the modern computer; the very idea of computation, which we now identify so closely with computing machines, actually preceded and led to their invention.

Dupuy has some interesting things to say about the Turing test too, but unfortunately I had to return the book to the library and don't remember all the details. One thing I do remember, though, is that the test Turing actually proposed in his paper, Computing Machinery and Intelligence [http://www.loebner.net/Prizef/TuringArticle.html], is rather more strange than is usually reported.

Turing asks us to imagine a test in which two subjects, a man and a woman, communicate with an interrogator via text. The man tries to persuade the interrogator that he is the woman, while the woman tries to help the interrogator discover the truth.

Now, Turing says, imagine replacing the man in this scenario with a computer. If the machine can fool the interrogator as often as the man can, then we have grounds for saying that the machine can think.

Isn't that odd? I wish I could remember what Dupuy had to say about it...

Mlle. Le Renard said...

obviously it would be more complicated to replace the woman with a computer ...

Noah said...


Mlle. Le Renard said...

I'm just kidding. Because she's a Woman. Thus her more complex cognition could not be represented with tape ... sorry I just taught de Beauvoir and Lloyd today, so I'm feeling a bit uppity

Noah said...

de Beauvoir was a die-hard computationalist.

john said...

Can you think of another discipline in which practice far outstripped theory?

I don't think there are too many. And I guess I'm not that surprised with that particular conclusion.

I can however think of a number of disciplines that ran off without ethics...

Mlle. Le Renard said...

Practice farther than theory: visual art, poetry, musical composition, ethics... physics is at least split, who knows who's ahead.

Well, mostly, when does theory really go ahead?

john said...

Well, that's kind of my point.

You've mentioned mostly the arts where practice came way ahead (pre-moderns!) Is that also the case with the sciences?

I'm tempted to say no.

Which is why I'm surprised when computer science is looked differently then say, physics or math. Theoretical math, physics, and computer science aren't going to be experimental...ie based on a series of lab experiments or tied to a physical machine like a computer. My intro algorithms class was handwritten proofs, no programming.

Also, maybe Turing didn't notice the woman in his experiment because he was gay. Maybe.