Date: Tue 15 Nov 1988 17:46-EST From: AIList Moderator Nick Papadakis Reply-To: AIList@AI.AI.MIT.EDU Us-Mail: MIT LCS, 545 Tech Square, Rm# NE43-504, Cambridge MA 02139 Phone: (617) 253-6524 Subject: AIList Digest V8 #128 To: AIList@AI.AI.MIT.EDU Status: R AIList Digest Wednesday, 16 Nov 1988 Volume 8 : Issue 128 Philosophy: Lightbulbs and Related Thoughts Attn set comments from a man without any Artificial Intelligence and Intelligence Notes on Neural Networks ---------------------------------------------------------------------- Date: 14 Nov 88 15:42:51 GMT From: rochester!uhura.cc.rochester.edu!sunybcs!lammens@cu-arpa.cs.corn ell.edu (Johan Lammens) Subject: Re: Lightbulbs and Related Thoughts In article <778@wsccs.UUCP> dharvey@wsccs.UUCP (David Harvey) writes: >Don't forget to include the iconic memory. This is the buffers >so to speak of our sensory processes. I am sure that you have >saw many aspects of this phenomenon by now. Examples are staring >at a flag of the United States for 30 seconds, then observing the >complementary colors of the flag if you then look at a blank wall >(usually works best if the wall is dark). [...] Perhaps this question is a witness to my ignorance, but isn't the phenomenon you describe a result of the way the retina processes images, and if so, do you mean to say that iconic memory is located in the retina? ------------------------------------------------------------------------------ Jo Lammens Internet: lammens@cs.Buffalo.EDU uucp : ..!{ames,boulder,decvax,rutgers}!sunybcs!lammens BITNET : lammens@sunybcs.BITNET ------------------------------ Date: Mon, 14 Nov 88 15:14:20 CST From: alk@ux.acss.UMN.EDU Subject: Attn set comments from a man without any The problem of constraint of the attention set by prior knowledge which was observed by Tony Stuart, i.e. that a known solution may inhibit the search for an alternative, even when the known solution does not have optimal characteristics, goes far beyond the range of David Harvey's statement that 'the only thing that can be said is that insconsistencies of data with the rule base must allow for retraction of the rule and assertion for [sic] new ones.' Stuart's observation, unless I miscontrue [please correct me] is not focused on the deduction of hypotheses, but extends also to realms of problem-solving wherein the suitability of a solution is (at the least) fuzzy-valued, if not outright qualitative. The correctness of a solution is not so much at issue in such a case as is the *suitability* of that solution. Of course this suggests the use of fuzzy-valued backward-chaining reasoning as a possible solution to the problem (the problem raised by Tony Stuart, not the "problem" faced by the AI entity), but I am unclear as to what semantic faculties are required to implement such a system. Perhaps the most sensible solution is to allow resolution of all paths to continue in parallel (subconscious work on the "problem") for some number of steps after a solution is already discovered. (David Harvey's discussion prompts me to think in Prolog terms here.) Why do I quote "problem"? Deconstruct! In this broader context, a problem may consist of a situation faced by the AI entity, without the benefit of a programmatic goal in the classical sense. What do I mean by this? I'm not sure, but its there, nagging me. Of course goal-formulation must be driven, but at some point the subgoal-goal path reaches an end. This is where attention set and sensation (subliminal suggestion? or perhaps those continuing resolution processes, reawakened by the satisfaction of current goals--the latter being more practically useful to the human audience of the AI entity) become of paramount importance. Here I face the dilemma: Are we building a practical, useful, problem solving system, or are we pursuing the more elevated (???) goal of writing a program that's proud of us? Very different things! Enough rambling. Any comments? --alk@ux.acss.umn.edu, BITNET: alk@UMNACUX. U of Mn ACSS "Quidquid cognoscitur, cognoscitur per modum cognoscentis" ------------------------------ Date: 15 Nov 88 02:29:12 GMT From: quintus!ok@unix.sri.com (Richard A. O'Keefe) Subject: Re: Artificial Intelligence and Intelligence In article <484@soleil.UUCP> peru@soleil.UUCP (Dave Peru) writes: >Definition of Intelligence: > >1. Know how to solve problems. >2. Know which problems are unsolvable. >3. Know #1, #2, and #3 defines intelligence. > >This is the correct definition of intelligence. If anyone disagrees, please >state so and why. > (Gilbert Cockton is going to love me for this, I can tell...) Intelligence is a social construct, an ascription of value to certain characteristics and behaviours deemed to be mental. One child who has memorized the periodic table of the elements will be deemed intelligent, another child who has memorized baseball scores for the last N years will be deemed sports-mad, even though they may have acquired comparable bodies of information _by_comparable_means_. If we have three people in a room: Subject, Experimenter, and Informant, if Subject does something, and Informant says "that was intelligent", Experimenter is left wondering "is that a fact about Subject's behaviour, or about Informant's culture?" The answer, of course, is "yes it is". Dijkstra's favourite dictionary entry is "Intelligent, adj. ... able to perform the functions of a computer ..." (Dijkstra doesn't think much of AI...) In at least some respects, computers are already culturally defined as intelligent. >Human beings are not machines. I agree with this. >Human beings are capable of knowing which problems are unsolvable, while >machines are not. But I can't agree with this! There are infinitely many unsolvable problems, and determining whether a particular problem is unsolvable is itself unsolvable. This does _not_ mean that a machine cannot determine that a particular problem _is_ solvable, only that there cannot be a general procedure for classifying _all_ problems which is guaranteed to terminate in finite time. Human beings are also capable of giving up, and of making mistakes. Most of the unsolvable problems I know about I was _told_; machines can be told! Human beings are not machines, but they aren't transfinite gods either. ------------------------------ Date: Mon, 14 Nov 88 22:35:47 CST From: David Kanecki Subject: Notes on Neural Networks Notes on Neural Networks: During the month of September while trying various experiements on neural networks I noted two observations: 1. Based on how the data for the A and B matrix are setup the learning equation of: T w(n)=w(n-1)+nn(t(n)-o(n)*i (n) may take more presentations for the system to learn then A and B output. 2. Neural Networks are self correcting in that if a incorrect W matrix is given by using the presentation/ update process the W matrix will give the correct answers, but the value of the individual elements will differ when compared to a correct W matrix. Case 1: Different A and B matrix setup For example, in applying neural networks to the XOR problem I used the following A and B matrix: A H | H B ------- |------ 0 0 0 | 0 0 0 1 0 | 0 1 1 0 0 | 0 1 0 1 1 | 1 1 My neural network learning system took 12 presentations to arrive at the correct B matrix when presented with the corresponding A matrix. The W matrix was: W(12) = | -0.5 0.75 | | -0.5 0.75 | | 3.5 -1.25 | For the second test I set the A and B matrix as follows: A H | B ------------ 0 0 0 | 0 0 1 0 | 1 1 0 0 | 1 1 1 1 | 0 This setup took 8 presentations for my neural network learning system to arrive at a correct B matrix when presented with the corresponding A matrix. The final W matrix was: W(8) = | -0.5 -0.5 2.0 | Conclusion: These experiements indicate to me that a systems learning rate can be increased by presenting the least amount of extraneous data. -------------- Case 2: Self Correction of Neural Networks In this second experiment I found that neural networks exhibit great flexibility. This experiment turned out to be a happy accident. Before I had developed my neural network learning system I was doing neural network experiments by speadsheet and hand transcription. During the transciption three elements in 6 X 5 W matrix had the wrong sign. For example, the resulting W matrix was: | 0.0 2.0 2.0 2.0 2.0 | |-2.0 0.0 4.0 0.0 0.0 | W(0)= | 0.0 2.0 -2.0 2.0 -2.0 | | 0.0 2.0 0.0 -2.0 2.0 | |-2.0 4.0 1.0 0.0 0.0 | | 2.0 -4.0 2.0 0.0 0.0 | W(24) = | 0.0 2.0 2.0 2.0 2.0 | |-1.53 1.18 1.18 -0.25 -0.15 | | 0.64 0.12 -0.69 1.16 -0.50 | | 0.27 -0.26 -0.06 -0.53 0.80 | |-1.09 1.62 0.79 -0.43 -0.25 | | 1.53 -1.18 -0.68 0.25 0.15 | By applying the learning algorithm it took 24 presentations the W matrix to give correct B matrix when presented with corresponding A matrix. But, when the experiment was run on my neural network learning system I had a W(0) matrix of: W(0) = | 0.0 2.0 2.0 2.0 2.0 | |-2.0 0.0 4.0 0.0 0.0 | | 0.0 2.0 -2.0 2.0 -2.0 | | 0.0 2.0 -2.0 -2.0 2.0 | |-2.0 4.0 0.0 0.0 0.0 | | 2.0 -4.0 0.0 0.0 0.0 | After 5 presentations the W(5) matrix came out to be: W(5) = | 0.0 2.0 2.0 2.0 2.0 | |-2.0 0.0 4.0 0.0 0.0 | | 0.0 2.0 -2.0 2.0 -2.0 | | 0.0 2.0 -2.0 -2.0 2.0 | | 2.0 -4.0 0.0 0.0 0.0 | Conclusion: Neural networks are self correcting but the final W matrix way have different values. Also, if a W matrix does not have to go through the test/update procedure the W matrix could be used both ways in that a A matrix generates the B matrix and a B matrix generates the A matrix as in the second example. ---------------- I am interested in communicating and discussing various aspects of neural networks. I can be contacted at: kanecki@vacs.uwp.wisc.edu or at: David Kanecki P.O. Box 93 Kenosha, WI 53140 ------------------------------ End of AIList Digest ********************