Mon.13.SEP.2010 -- Inhibition Revisited We need to go into the MindForth free AI source code and start removing inhibition code from verbs of being, because in clarifying our rules of activation, we have concluded that predicate nominatives should endure inhibition, but not the be-verbs that link to them.
Tues.14.SEP.2010 -- Code, Test, Debug In our coding yesterday we switched many times back and forth between debug and test, debug and test. We were aided by our initial search for, and culling out, of current instances of neural inhibition implemented in the NounPhrase, BeVerb, VerbPhrase and EnCog mind-modules. Having formulated yesterday two rules -- that inhibition should occur in pairs, and that reasons for query-response variance should drive inhibition -- we are able now to focus more carefully now on where and why neural inhibition should occur in our open source code for artificial intelligence (AI). In the light of the new rules, we comment out the inhibition code in the EnCog thinking module, because we fear that it may interfere with our troubleshooting. Then we fire up the AI and we put the same question to the AI seven times, with the following results. Human: what are you Robot: I AM ANDRU Human: what are you Robot: I AM ME Human: what are you Robot: I AM ME Human: what are you Robot: I AM ME Human: what are you Robot: I AM ME Human: what are you Robot: I AM ME Human: what are you Robot: I AM ANDRU We interpret the above exchange as showing that the response-idea "I AM ANDRU" was initially inhibited as a pair of two identical thoughts, one in the innate knowledge of the EnBoot English bootstrap, and one in the response made by the AI when asked, "What are you?" The inhibiition on the predicate nominative "ANDRU" lasts so long, that the "I AM ANDRU" KB-response is suppressed during five subsequent queries, until finally the inhibition of "ANDRU" has worn off and the AI can once again answer, "I AM ANDRU". It is up to the mind-designer and/or AI coder to decide how severely to inhibit ideas in the AI, under considerations such as just how much mind-control do we wish to engage in; what is the proper trade-off between time-duration of inhibited concepts and exhaustive KB-search; and so forth. Right now we are eager to have inhibition strong enough to aid our code-test-debug cycle, and we especially want to verify that each instance of inhibition is happening on the horns of an idea pair (old and new), so that no stray inhibitions are ruining a chain of AI thought. Uh-oh. We've got some good results. Let's upload the code. http://www.scn.org/~mentifex/mindforth.txt Meanwhile, here is what happened. In our hovering, looming quest for self-referential thought, we were testing the AI not only by asking "what are you" but by first telling the AI "you are a program". Then we were getting "garbage out" in response to our "what are you" queries. But instead of dealing with the central issue (properly implementing inhibition), we tried removing the annoyance of seeing "I ARE ME" and "PROGRAM ARES ME" as some actual responses that the AI has issued today. So we went into the VerbPhrase module and inserted the following code. motjuste @ 58 = IF BeVerb EXIT THEN \ BE; 14sep2010 motjuste @ 57 = IF BeVerb EXIT THEN \ AM; 14sep2010 motjuste @ 66 = IF BeVerb EXIT THEN \ IS; 14sep2010 motjuste @ 67 = IF BeVerb EXIT THEN \ ARE; 14sep2010 The above code forces the AI to treat "AM IS ARE" not as normal, regular verb-forms, but instead to let the BeVerb module enforce strict rules for the usage of forms of the "BE" verb. Immediately we obtained the following dialog which still has its problems but is on track. Transcript of AI Mind interview at 6 31 38 o'clock on 14 September 2010. Human: you are a program Robot: PROGRAM IS PROGRAM Human: what are you Robot: PROGRAM Human: what are you Robot: I AM PROGRAM Further inspection of the activation-levels indicates that inhibition does not kick in until the AI states the idea, "I AM PROGRAM", leaving out the article "A" because the EnArticle module is not being called. It is an eerie feeling to see the above results, as if the AI Mind is awakening before our very eyes. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/8660244-d750797a Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
