Technische Universitat Wien


Grundlagen wissenschaftlichen Arbeitens, 2.0 PS

LVA 185.170












Artificial Intelligence: "Much ado about not very much ?!"





Research paper







Matej Marcak


Matrikelnummer: 0425503
Studienkennzahl: 534

1  Introduction

No doubt, the topic of the Artificial Intelligence has long been attracting people's attention ever since it was raised the very first time by Kurt Goedel and Jacques Herbrand in the early 20-th Century. Having proposed their theories at that time as the modern conception of computability ("general recursiveness"), A. Turing later reformulated this notion in terms of connection with digital computers (not yet invented!) suggesting also abstract computers as a model for a mind ([1] 268). Among people it came to mean not only narrowing the distance between reality and prospects that people had been idealizing for years before but it also created the vast unexplored horizons especially for the research and scientists. As the technology and computers have been developed over previous years and still are being developed forth today, we may find the AI-based systems implemented in various aspects of society and science ranging from automated programming, planning, organizing, e-store, e-banking and most of all, or rather said most significantly in the field of entertainment – especially computers (computer games), digital-media etc. These in fact being just one part of the impact, the negative opinions about AI were raised afterwards as well. First of all, the question "What's AI all about for???" is a good point to start, when trying to consider negatives or let's say not such passionate opinions about it. In fact, the major target of the whole theme named Artificial Intelligence has become far more ambitious - that is creating a pure artificial human mind possibly. Is it a good or bad idea one may ask??? The most appropriate answer might likely be that one of golden middle way. Yes, it's a good idea for sure would be an answer without any hesitation of all the supporters of the AI - humans have always longed and wanted to reach for something what might be called utmost end of their knowledge indeed - but then comes a very important "but". Hilary Putnam characterizes that "but" and sets skepticism about the ability of AI to bring valuable insights into the nature of human understanding pretty well when posing the crucial question: "Has Artificial Intelligence taught us anything of importance about the mind?" ([4] 1). Later in his article "Artificial Life: Much Ado About Not Very Much" he continues saying Ï am inclined to think that the answer is no." and follows on ÄI has so far spun off a good deal that is of real interest to computer science in general, but nothing that sheds any real light on mind." ([1] 270). At last, he closes up his arguments with "What's all the fuss about now? Why don't we wait until AI achieves something and then have an issue?" ([1] 271).

2  Putnam's Issues

Among possible guesses what AI has actually achieved nowadays for sure no doubt, an overwhelming reference could be mentioned namely pointing to "expert systems". These systems often being related to as "woow, computers are thinking" are merely high-speed data-base searchers, not even worthy to be considered a mental capacity ([1] 271). James McNelis' comment takes it all: "the dramatic title 'Expert Systems - Computers That Think Like People.' It sounds like quite a breakthrough: 'An expert system ...is a pipe-puffing savant with deep experience in a field of human knowledge. It can size up a situation and draw on a lifetime of experience to make reasoned judgments …handles ideas in the same way that humans do ...it contains human wisdom.' But one's expectations take a nosedive when one realizes that this alleged genius is nothing more than an electronic symptom catalogue. You have a runny nose? This computer will tell you that perhaps you have a cold. Fingers fall off? Leprosy is the answer. It may be a useful contraption, but it doesn't think any more than does your phone book. Neither does any other computer." ([5]). On the other hand, people are interested in and fascinated by that magical idea of modeling the artificial mind or brain as a digital computer. However, succeeding in doing so and being able to do it correctly, thus achieving a real AI, is completely different matter than just the belief itself upon building a brain or a mind or an intelligence as such computers with an appropriate software ([1] 271). It isn't that simple indeed whereas it is not possible to consider the evolution and the mind to be just a piece of software working according to a given order. The idea of tinker and millions of bits of "tinkering" that describes evolution – as Putnam states it by further calling AI "one damned thing after another" ([1] 272) – might mean the amount of such "damned things" would be enormous, thus the model of mind AI will have eventually been developed equally large, making AI impractical. At this point D.C. Dennett in his work "When Philosophers Encounter Artificial Intelligence" sets a contrary and opposes such views while relating to Minsky's tertium quid where the mind as crystal and the mind as chaos both include also the mind as a gadget – not to be maintained by "deep" mathematical laws whereas it is a designed object analyzable in functional terms: ends and means, cost and benefits, elegant solutions as well as shortcuts, jury rigs, and cheap ad hoc fixes ([2] 286). This view, though having several parallels with the view of scientists is resisted by philosophers according to him, because their traditional aprioristic approaches to the investigations of mind actually empower them only a little to explore phenomena of odd hacks ([2] 287). A "Master Program" is what Putnam sees as a solution for AI to be able to simulate human intelligence ([1] 273) if ever by some circumstances should the AI be able to do something like that. The reason for that is arisen upon relating the AI to math and its general rules and insisting upon various ways proofs if there is anything to prove, or indeed anything to be claimed to be existing but not proven yet. This is, in fact, how the science of math keeps its clarity. Often used example might be the mathematical induction and its proofs. When we look at the AI, nothing like general rules or proofs had ever existed before and therefore the AI may never reach its major goal of modeling artificial brain or mind successfully. When considering other particular problems that AI is facing to the significant extent - and Putnam mentions that too - is its dealing with issues concerning induction and cognition as well as the background knowledge. Let's take an example of uranium atoms and the future members being in the future farther than we can personally imagine, then there's a tendency to believe that the future members will resemble on the average ([1] 274). However, we do have a less tendency to talk about people the same way because in that moment we think of much more aspects and situations concerning people, in such range that we simply are not able to predict whatever might happen. Thus, from our point of view, utilizing of the same induction on two different cases seems senseless, moreover in case of machines even impossible. Or would anyone ever be able to make a machine being able to use appropriate induction to all the cases it will have ever encountered? The cognition, thus the ability of recognizing similarities among things with no input from sense organs or any physical stimulation cannot be ignored too. The humans have no particular difficulties in finding similarity among knives not that they are alike, but in terms of all having been made to cut or stab ([1] 275). Could machines ever be able to do such attributing of purposes to agents? Another example of cognition covers even harder aspect - similarities expressed by adjectives that have a "family resemblance" ([1] 275). Just take an adjective "white" - it expresses a color primarily, but the people usually call also other people as "white" i.e. not because they are really white as snowmen or albinos but they express their race that way. The same occurs when saying "open a door" - which does not only mean that particular action being performed in the reality, but it does express opening a trade or borders as well ([1] 275). What would be a solution to the machines in terms of problems with cognition? One may propose an "ideal language" solution - a language with words not changing their extensions in context-dependent way ([1] 276). Of course, it might be - however one ßmall" problem persists - previously mentioned "family-resemblance" adjectives. The serious problems are formed by so-called "conflicting" inductions. Hilary Putnam cites an interesting example given by Nelson Goodman. Let's suppose there’s a statement that no one who has ever entered Emerson Hall at Harvard University has been able to speak Inuit language. Statement itself through the induction suggests that if any person enters Emerson Hall, then he for sure does not speak Inuit. Shall we predict if Eskimo from Alaska enters that Hall he will no longer be able to speak his mother tongue? ([1] 276). The induction itself is good at the very first sight but on the other hand it is bad too. Nobody indeed forgets his mother tongue upon entering the Hall. Goodman further characterizes this apparent non-sense as an inductively supported law of "better entrenchment" that people do not loose their ability to speak upon entering a new place ([1] 276). And here is the point, if not hardest till now - how to implement this matter into machines, how to give them a clue to think in such extensive consequences? Even proposing the likely solutions seems hopeless when dealing with such conditions. The first one could be creating huge software from all the information a sophisticated, educated human being has or may have ([1] 277). Dennett says this has already been applied in CYC project conducted by Douglas Lenat ([2] 291). A successful project would be of course rather anything else but miracle because that might likely take generations, thus the whole project seems being almost impossible to complete to a sufficient extent and as Dennett later ultimately agrees being a matter of finances as well and those indeed are not infinite. The second one could be a sort of "learning device" on the basis of a small child learning things and trying to recognize things around him. ([1] 278). Again, considering various aspects this intention arises a "Natural language" problem further related to the simulation of direct understanding a human language. The core of problem appears when accepting a view that such understanding in case of human beings means in fact a "maturation of an innate ability in a particular environment" ([1] 278). Would that maturation ever be possible to simulate on the machines? Surely, it would not be far from truth to say and claim it was never and will have also never been possible by any time.

3  Dennett strikes back

It is D.C. Dennett himself who sums up and tries to ease the whole matter by claiming Putnam's extraordinary and exaggerated setting "a worst-case possibility" as the only alternative to "Master Program" ([3]) since most AI projects are just "explorations of ways things might be done" ([2] 289), rather than their real implementation into reality. Furthermore, by all means AI deserves a particular respect by having set new problems and issues as well as allowing "trade-off" ([2] 294) of various approaches and providing raw materials ([2] 293), hence inventing the actual solutions that so many are calling for, seems really unimportant or at least not that crucially important.

4  Conclusion

Finally, it is also worth to mention some words concerning the future the Artificial Intelligence has provided people with a hope definitely. A hope to make all their dreams they had ever been dreaming of, come true. One must however never be forgotten whatever the final outcome of success might be - a sense for carefulness or awareness - since no matter what a human being comes up with to this world it will be going to be to some extent bipolarized - with positives and negatives both included. Most people do see just that one positive part as the only one but it is worth to always remember that nothing's so perfect as it might appear at the very first sight. Having been depicted in such movies like "I Robot" and "The Minority Report" already, even formerly thought Asimov's perfect laws of robotics have been proven not ultimately true and perfect. May also the possible negative aspect of AI be a momentum for us forevermore.

References

[1]
Putnam, H. (1998): Much Ado About Not Very Much. In: Daedalus, Vol.117, Nr.1, pages 269-271.
[2]
Dennett, D.C. (1998): When Philosophers Encounter Artificial Intelligence. In: Daedalus, Vol.117, Nr.1, pages 283-295.
[3]
EXTROPY Journal. "Well, Oh Yeah"
Online Available http://www.extropy.com/ideas/journal/previous/1992/09-03.html, November 17, 2004.
[4]
Lockhorst, Gert-Jan C. "Has AI taught us anything of importance about the mind?"
Online Available http://www2.eur.nl/fw/staff/lokhorst/putnam.html, November 17, 2004.
[5]
McNelis III, James I. ÄI: Much ado about not very much"
Online Available http://members.aol.com/mcnelis/AI.html, November 17, 2004.


File translated from TEX by TTH, version 3.64.
On 23 Nov 2004, 19:30.