Subject: TM as 1st AI Prog? From: "William J. Rapaport" Date: Wed, 24 Feb 2010 11:57:48 -0500 (EST) In class today, I suggested that the Turing machine could be considered as the first AI program. Brian Borncamp suggested that, if that were to be considered an AI program, why not also consider earlier calculators as AI devices. And Gene Wang observed that one difference was that calculating devices were intended to help humans calculate, whereas AI programs are intended to simulate/replicate human cognitive processes. (I'm liberally paraphrasing their ideas; if I'm seriously misrepresenting them, please let me know.) I think one reason that the TM can be considered an AI program whereas calculators are not is this: Turing first did a very careful analysis of how humans actually compute/calculate and then showed how that can be modeled by a TM. Previous calculating devices were merely mechanical (even analog), and were not intended to model how humans did it. (That may be part of Gene's comment.) So, what Turing showed is that human computation is mathematically computable; i.e., he showed that a certain kind of human cognitive process was computable--and that's one of the definitions of AI ======================================================================== Subject: Re: TM as 1st AI Prog?: 201001_049223_CL From: "William J. Rapaport" Date: Wed, 24 Feb 2010 20:09:58 -0500 (EST) A student writes: "If a "true" Turing machine requires infinite tape to do any sort of calculation, does that mean that a computer in order to be a Turing machine would require infinite memory? Turing in his paper says that people have finite memories though? Or am I reading that wrong? So if a Turing machine is modeling people, then the machine does not need infinite memory, I thought. I'm just trying to conceptualize what's going on here. Sounds to me like two conflicting arguments. And if this is to be considered an AI program, then AI would require infinite memory, as well, which would imply that we humans have infinite memory, too." Reply: Interestingly, as far as I can tell, Turing is silent in his 1936 paper about how "long" the tape is. The informal way TMs are usually introduced does talk about an "infinite" tape. But the more mathematically precise way to describe it is as an "arbitrarily long" tape. That is, the tape is as long as you need it to be. For most computations (the ones that really do halt with a correct answer), the tape will be finite. Since no real machine can print out an infinitely long decimal, no real machine will require an infinite tape, either. In real life, you can only print out a finite initial segment of the decimal part of a real number; i.e., it will always be an approximation, but you can make the approximation as close as you want by just printing out a few more numbers. So I don't think there's any problem about modeling people with infinite memory. People don't have infinite memory, and neither do TMs or, certainly, real computers. The major difference between TMs, on the one hand, and people and real computers, on the other hand, is that TMs can have a tape/memory that is as large as you need, while people and real computers are limited. ======================================================================== Subject: Re: TM as 1st AI Prog? From: "William J. Rapaport" Date: Wed, 24 Feb 2010 20:15:12 -0500 (EST) Another student writes: "I'm not convinced that the purpose of the creator ought to be considered when taking into account the classification of whether or not something is an AI program. In our day to day experience, humans classify objects without taking into account some teleological nature of the object. I see no reason why there ought to be a requisite that an AI program have a particular teleology associated with it if other object classifications not need it. I would additionally argue that humans add, multiply, subtract, and divide on a regular basis. I would also argue that these mathematical abilities are cognitive processes. If an AI program is something which replicates a human cognitive process, anything which replicates the said mathematical behavior is in fact an AI program. In effect calculators are then AI programs." Reply: Fair enough. In fact, both Shapiro and I have made a distinction between what we call "computational philosophy" and "computational psychology". The latter is a research project that tries to develop computational theories of human cognition using algorithms that are as close as possible to the ways that humans actually cognize. The former is a research project that tries to develop computational theories of human cognition using any algorithm that has the same input-output behavior. For more on these ideas, link to "3 Goals of AI Research": http://www.cse.buffalo.edu/~rapaport/563/aigoals.html