The Computational Universe

            Searle has a point, and it’s this: it’s entirely possible to perform a set of tasks in what appears to be the usual way without having the mental state that people performing such tasks are usually taken to have. But the implications that this is supposed to have for AI, or indeed for the mind, are not as clear or conclusive as he takes them to be. I don’t propose to say in this short space what the proper response to his thought experiment would look like. Rather, I’m going to pose two questions that are meant to focus our reaction in the way that I’d like it to be focused.

            First, what is it that’s important and special about human cognition? If you think that the sorts of visible, external behavioral outputs that are a consequence of human cognition constitute just what’s important about it, then Searle’s thought experiment lands you in hot water. If, on the other hand, you think that the internal experience of mental phenomena is what’s important, then you’ve got to think of how a computer might ever come to have this kind of thing in a qualitatively similar way.

            Second, and this follows very closely, how do we *know* that other people have the internal experience of mental phenomena that’s similar to our own experience of the same thing? This is a variation on the problem of other minds, which asks how we know that there are such things as other minds *at all*, never mind whether they have internal experience that are qualitatively similar to our own. Once you’ve got your standard of evidence set straight on this question, then you’re ready to ask it of AI.

It was clear that Sir Timothy Berners-Lee would play the part of the mad Oxonian physicist when he began a discussion of modeling social specialization with an anecdote about how a physicist would model all of the air in room: ‘Imagine the room is filled with billiard balls!’ Sir Timothy possesses a rapier-quick, if occasionally overheated intellect. He’s very clever, there’s no doubt. But listen to his words rush out in a big tumble, falling all over each other, or hear the Oxford ‘r’ or the slightly sibilant ‘s’, or observe the endearing typographical errors all over his presentation (‘acedemic’) – and you’ll know just the sort of character I’m describing.
The substance of Sir Timothy’s talk had much to do with the degree to which the utility of the internet is determined by its ability to various rules for effective and mutually beneficial social behavior. Social rules and technical rules, the joint theme seemed to go. Sir Timothy began by discussing the ‘philosophical engineering’ – cf. ‘experimental philosophy’, as those who read physics at Oxford in Sir Timothy’s undergraduate days used to call their subject – of the internet. The essence of the web server is peer review plus, a more democratic, but ideally equally reliable, means of information transfer. The social ends of the internet are clear to Sir Timothy: to ‘serve useful stuff’, ‘make useful links’, and respect the law (intellectual property rights, libel, fraud). Sir Timothy’s hope for the future? The increased popularity of the semantic web, intended to serve much the same social purpose, but more effective in its structuring of information than the tree-shaped internet. The world awaits.

Turing machines are, according to the Church-Turing thesis, supposed to capture just what computation is. The Turing model itself consists in an infinite “scratch pad,” the symbols “0” and 1,” and instructions for scanning the scratch pad and giving “0”s or “1”s according to a series of simple rules. Now, this probably does not sound very impressive. Indeed it would be fair to ask in what sense exactly a simple set of instructions for producing certain patterns of “0”s and “1”s is this vast and powerful thing called computation.
The sense in which the Turing model is supposed to encapsulate computation turns on the capacity of Turing-Post machines to simulate other programs, i.e. on the universality of such machines. (A machine is a universal program just in case for any program ‘x’ the universal program will simulate the behaviour of ‘x’ in relation to a given set of data). It follows from the claim that the Turing-Post machine is universal that the Turing-Post machine should be able to simulate pseudocode programs, which in fact it can. Let’s not imagine, though, that there’s something magical about Turing machines: they have very definite limitations, like an in principle inability to solve the halting program.
There is a school in the philosophy of mind which holds that mental states just are particular groups of sensory inputs, behavioural outputs, and causal relations to other mental states. Philosophers who believe this are “functionalists.” Turing-machine functionalists go further, and say that the proper analogy for the functionalist should be between mental states and machine states. I do not happen to share precisely this view, but the belief of a number of eminent philosophers that our mental states are in fact machine states should make us listen with interest the next time we hear about (prima facie, trivial) rules for writing “0”s and “1”s.

There are at least a thousand computer languages, familiar ones being Java, C++, Basic, and Python. Happily, all of these languages share a certain basic grammar. In addition to sharing the basic features of human languages, they have in common certain formal features; witness their ability to represent variables, respond to conditional and loop statements, or to do arithmetic. Realize in considering this confluence of characteristics that for computers, all objects are sequences of numbers.
Scribbler pseudocode, insofar as it shares most of these formal features, could correctly be described as an answer to the computational Tower of Babel. The Scribbler responds to two kinds of instructions: simple and compound. Simple instructions include “move forward for x seconds,” “reverse for y seconds,” etc. Compound instructions consist in simple instructions joined by conditionals (“If this, then that”) or loops (“Do for ___”) specifying the conditions under which an action should be repeated.
Why should we be interested in artificial languages? The obvious answer is that, when cleverly manipulated, they yield machines of enormous power. Perhaps more controversially, I would also claim that formal languages like logic – to which pseudocode bears a kind of resemblance, as for example when it sets rules for correct inference – tell us something important about the proper constraints for coherent thought. By way of comparison, they also shew us interesting things about the structure of the natural languages.
I think we should bear a kind of anti-psychologism in mind when we scrutinize the contour of artificial languages like pseudocode, regardless of what psychological processes they are supposed to resemble. Scribbler commands obtain just in case the Scribbler behaves in a certain way; they have nothing to do with internal states, of which robots have none.

I am a freshman in Forbes who intends to major in analytic philosophy. I’m particularly keen on metaphysics, philosophy of math., and moral philosophy. Several years ago I thought I might become a classicist instead, and wanted to go into Greek textual criticism or palaeography. My friends tell me that at the time I also held some very dogmatic views about Greek phonology, particularly where the pronunciation of voiceless aspirate stops (which in Byzantine Greek – but not before! – become spirants) was concerned. But now I’ve more or less permanently traded in that set of abstract concerns for another, more philosophical set of (I suppose, equally abstract) concerns.

I’m from New York, and went to Hunter (71 East 94th Street). I listen to and play lots of Renaissance and Baroque music; I also enjoy some eighteenth and nineteenth century opera, especially when it’s put on at the Met. Just at the moment I’m boycotting City Opera, which recently introduced amplified singing. When I’m at home I like to cook, throw dinner parties, read, write, take tea, walk around the Great Lawn, and nap.

I have a lot of respect for the natural sciences, and am particularly interested to learn two things from COS 116: a) what practical use formal logic has, and b) what the scientific potential for developing artificial intelligence is like. As far as I’m concerned, the central objection to AI expressed by John Searle in a famous thought experiment involving a “Chinese Room” has not been satisfactorily answered. This is not to say that I would mind if it were answered in the future.

Update 4:15 PM on 12 February: I use a 15″ Powerbook G4.