Mike Tintner wrote on Thu, 6 Dec 2007: > > ATM: >> http://mentifex.virtualentity.com/mind4th.html -- an AGI prototype -- >> has just gone through a major bug-solving update, and is now much >> better at maintaining chains of continuous thought -- after the >> user has entered sufficient knowledge for the AI to think about. >> > It doesn't have - you didn't try to give it - > independent curiosity (like an infant)? No, sorry, but the Forthmind does have an Ask module at http://mentifex.virtualenty.com/ask.html for asking questions -- which, come to think of it, may be a form of innate curiosity.
Meanwhile a year and a half after receiving a bug report, the current bug-solving update has been posted at http://tech.groups.yahoo.com/group/win32forth/message/13048 as follows FYI: > OK, the audRecog subroutine is not totally bugfree > when it comes to distinguishing certain sequences > of ASCII characters. It may be necessary to not use > MACHINES or SERVE if these words confuse the AI. > In past years I have spent dozens of painful > hours fiddling with the audRecog subroutine, > and usually the slightest change breaks it worse > than it was before. It works properly probably > eighty percent of the time, if not more. > Even though the audRecog module became suspect > to me over time, I pressed on for True AI. On 14 June 2006 I responded above to a post by FJR. Yesterday -- a year and a half later -- I finally tracked down and eliminated the bug in question. http://mind.sourceforge.net/audrecog.html -- the auditory recognition "audRecog" module -- was sometimes malfunctioning by misrecognizing one word of input as the word of a different concept, usually if both words ended the same. The solution was to base the selection of an auditory recognition upon finding the candidate word-match with the highest incremental activation, rather than merely taking the most recent match. By what is known as serendipity or sheer luck, the present solution to the old audRecog problem opens up a major new possibility for a far more advanced version of the audRecog module -- one that can recognize the concept of, say, "book" as input of either the word "book" or "books." Since audRecog now recognizes a word by using incremental activation, it should not be too hard to switch the previous pattern-recognition algorithm into one that no longer insists upon dealing only with entire words, but can instead recognize less than an entire word because so much incremental activation has built up. The above message may not be very crystal clear, and so it is posted here mainly as a show of hope and as a forecasting of what may yet come. http://mind.sourceforge.net/mind4th.html is the original Mind.Forth with the new audRecog. http://AIMind-I.com is FJR's AI Mind in Forth. (Sorry I can't help in the matter of timers.) ATM -- http://mentifex.virtualentity.com/mind4th.html ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=73193379-092711
