RE: [agi] Psychometric AI
I like the gist of it ... though just did quick skim of the paper. In particular I like the idea of pushing/orienting AGI systems toward NLU and human standards to promote usability (or more properly: our ability to mutually relate). As AGI testing and validation goes, some might recall in my IVI Architecture posted here about a year ago, I specified testing to proceed from Mental Status Tests (basic orientation, attention, memory, etc. tests like a human neurologist would administer) - Personality Tests (to detect any severe psychoses, in interest of FAI :-)) - IQ Tests (here's where WAIS, and others would come into play). The latter, I agree, is largely the crux of what is meant by Intelligence. But there is a lot of cognitive framework that needs to be in place first. After standard IQ tests, one would start testing in particular narrower domains of interest to the AGI's application at hand, e.g., AP Chemistry, Astrophysics, Auto Mechanics, Symphonic Composition, or whatever. J. W. Johnston -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Ben Goertzel Sent: Friday, September 17, 2004 10:49 AM To: [EMAIL PROTECTED] Subject: RE: [agi] Psychometric AI Hi, I don't think that trying to overfit one's AGI system to some specific set of tests is a really useful approach. Also, I don't think that intelligence tests, as currently formulated for psychometric testing purposes, form a very natural set of developmental milestones for an AGI system. I think it would be possible to create a narrow AI system that passed a lot of IQ tests but still lacked general intelligence -- just as one can create narrow AI systems to play chess, checkers, and so forth. Psychometric tests are only moderately meaningful in the human-intelligence context for which they were devised; applying them beyond the human domain weakens their meaning even further... I don't think it's a boundlessly dumb approach or anything; but it's not an approach I would particularly recommend... -- Ben -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Shane Sent: Friday, September 17, 2004 10:37 AM To: [EMAIL PROTECTED] Subject: Re: [agi] Psychometric AI Hi Ben, You think it's a silly approach because...? I'm just about to read their paper and thus I haven't formed an opinion on their approach yet myself. Thanks Shane --- Ben Goertzel [EMAIL PROTECTED] wrote: This may be of interest to someone... Psychometric AI: http://www.cogsci.rpi.edu/peri/main.html A slightly silly approach, IMO, but it would certainly be a tractable research program to apply NM to these tasks I'm more interested in the AGI-SIM approach, however... -- Ben --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] Find local movie times and trailers on Yahoo! Movies. http://au.movies.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Psychometric AI
I noticed that too. Seemed like this list doesn't archive attachments (or has particularly good SPAM filter :-). I don't have the paper posted on any site. Will send you a PDF (748 KB). If others want a copy, let me know via email. Thanks! J. W. -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Peter Voss Sent: Friday, September 17, 2004 12:39 PM To: [EMAIL PROTECTED] Subject: RE: [agi] Psychometric AI I can't find it in the archives. Can you give me a link? Thanks, Peter -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of J. W. Johnston ...As AGI testing and validation goes, some might recall in my IVI Architecture posted here about a year ago, I specified testing to proceed from Mental Status Tests (basic orientation, attention, memory, etc. tests like a human neurologist would administer) ... --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.764 / Virus Database: 511 - Release Date: 9/15/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] AGI's and emotions
Title: Message Folks interested in this thread should check out the draft of Marvin Minsky's upcoming book "The Emotion Machine". Been available at his web site for quite some time: http://web.media.mit.edu/~minsky/ The current draft doesn't seem to have an executive summary that lays outthemain thesis, but in a 12/13/99 posting (http://www.generation5.org/content/1999/minsky.asp), Minsky says: The central idea is that emotion is not different from thinking. Instead, each emotion is a type or arrangement of thinking. There is no such thing as unemotional thinking, because there always must be a selection of goals, and a selection of resources for achieving them. From my notesafter skimming some of the book about a year ago, it seemed thatMinsky sees emotions as kinds of "presets" (his term - "Selectors") that determine what mind resources and goals are active at a given time to solve a particular "problem". [I seem to recall Antonio Damasio also had a similar conception... and he called the emotional "set points" PATTERNS!] The following isfrom the draft of Chapter 1 Section 6: Each of our major emotional states results from switching the set of resources in useby turning certain ones on and other ones off. Any such change will affect how we think, by changing our brains activities. In other words, our emotional states are not separate and distinct from thoughts; instead, each one is a different way to think. For example, when an emotion like Anger takes over, you abandon some of your ways to make plans. You turn off some safety-defenses. You replace some of your slower-acting resources with ones that tend to more quickly reactand to do with more speed and strength. You trade empathy for hostility, change cautiousness into aggressiveness, and give less thought to the consequences. And then it may seem (to both you and your friends) that youve switched to a new personality. Good stuff! (IMHO) J. W. Johnston -Original Message-From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Ben GoertzelSent: Wednesday, February 25, 2004 11:25 AMTo: [EMAIL PROTECTED]Subject: RE: [agi] AGI's and emotions Agreed --- we tend to project even abstract experiences back down to our physical layer, and then react to them physically ... a kind of analogy that AGI's are unlikely to pursue so avidly unless specifically designed to do so ben g -Original Message-From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]On Behalf Of Philip SuttonSent: Wednesday, February 25, 2004 12:00 PMTo: [EMAIL PROTECTED]Subject: RE: [agi] AGI's and emotions Emotions ARE thoughts but they differ from most thoughts in the extent to which they involve the "primordial" brain AND the non-neural physiology of the body as well. I guess we call emotions 'feelings' because we feel them - ie. we can feel the effect they trigger in our whole body, detected via our internal monitoring of physical body condition. Given this, unless AGIs are also programmed for thoughts or goal satisfactions to trigger 'physical' and/or other forms of systemic reaction, I suppose their emotions will have a lot less 'feeling' depth to them than humans and other biological species experience. Cheers, Philip To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] One AGI or many?
Finally got around to skimming the referenced paper. Per lots of Ben's stuff, found it quite readable and interesting. (Especially a good reminder of key Novamente concepts for those of us who incompletely waded through the more comprehensive documentation in the past :-) Two quick comments... 1. Haven't purged the bogus [comment] label yet :-( 2. Have you considered and/or does Novamente support non-atomic memory structures? In particular-- episodic type memories. In my IVI architecture posted here a while back, I suggest a Memory Subsystem consisting (mainly) of a Knowledge Base (compiled concepts/semantic net-type system) AND Memory Files (raw video, sound, and other sense data). Seems to me, when you start talking about distributed AGIs talking Psynese, might be worthwhile to support the transfer of Raw/Episodic Memories as well. For instance-- has your favorite cluster seen Lord of the Rings III yet? :-) Or maybe more to your point, be able to send videos from a ShapeWorld UI session between distributed AGIs. The best Memory Files would contain raw, but synchronized, vision (electromagnetic), sound (vibration), chemical (scents, tastes, motions), and other packaged data files. Might be useful for distributed AGIs to digest THESE sources for fairly unambiguous compilation into their own Atoms/Maps/KBs. J. W. Johnston -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Ben Goertzel Sent: Thursday, January 15, 2004 3:02 PM To: [EMAIL PROTECTED] Subject: RE: [agi] One AGI or many? This theme of partial mind-melds between future AI's, leading to a kind of hybrid between a society and an individual, was discussed in a paper I wrote last year, which was (I think) briefly discussed on this list. I called this kind of hybrid being a mindplex; see.. http://www.goertzel.org/dynapsyc/2003/mindplex.htm This paper will be published in the proceedings of the 2001 Global Brain conference, one day... -- Ben Maybe the most successful approach will be a community of individual AGIs that can specialise but that also engage in exchanges of data/knowledge and can also do partial mind-melds when super- mentation is required. Cheers, Philip I think that this last choice is the correct one. In fact it will be forced, basically due to the speed of light limitation in the transmission of signals. But note that there will be broadband connections between the separate members of the community. Thus it will in a sense be one individual in the sense that a corporation is one individual. (Well, actually a bit more so. They will be able to engage in genuine thought transmission, where here a thought would be an entire mental model of a situation, complete with desired goal states and physical sensoria. Probably bz2 is pushing the limit on compression, but if they share high level mental constructs, what would need to be transmitted would be analogous to the source code of a program, and it's data, but NOT the libraries. This would enable a real compression in the needed bandwidth. [Note that speech among humans automatically assumes some of these properties, so a verbal description is much more compact than a AV recording.]) This, of course, assumes that they will have identical primitives (i.e., the same version of the library). --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]