Tuesday, February 12, 2013

Versatile Self Improvement - VSI - VI - VLSI - VLI | ВСИ - Всестранен Самоусъвършенстващ ВсИчко | Terminology | Universal Self-Improver - USI

I'm adding another term to my AGI dictionary, following the:

Self Improving General Ingelligence - SIGI which was used as a title of the mini-conferene last year.

Versatile Self-Improver - VSI - Universal Self Improver (USI)

A thinking machine/artificial mind that has all capabilities and characteristics of a universal genius (all kinds of cognitive activities)

Other Variants: (Update, 13/2/2013)

Versatile Self-Improvement - The process of development of the VSI entity.

Versatile Limitless Self-Improvement (VLSI) - More powerful VSI.

Versatile Convergent [Limitless] Self-Improvement (VCLSI) - Emphasizes another aspect of the system.

It of course includes/implies also the externalized part of the improvement, and a shorter variant is just:

Versatile Improver and Versatile Improvement, respectively

Versatile Limitless Improver and Versatile Limitless Improvement.

Български: Всестранно Неограничено Самоусъвършенстване, Всестранно Усъвършенстване, Всестранно Развитие, Всестранно развиване на всичко ... Универсален изкуствен разум


In brief - although AGI - "[artificial] general intelligence" is better than the whore-term "AI"*, it still needs clarification, as mentioned in a recent article. "General" is also ambiguous, meaninig both "universal" and "non-specific".

Thinking machines as I envision them are both universal and specific, as an ingenious human should be capable to deal both with the tiniest and most specific technical detail, and with the most general matter, compared to average humans, and the optimal genius should do in all possible domains, not only in "general" in the sense of "vague".

The new term avoids also "intelligence" for several reasons, such as the ridiculous nonsense that I've mentioned again lately, in which many AI-ers and AGI-ers believe: "yet nobody knows what intelligence is" and "there's no consensus about what intelligence is".

Not that the latter is not true, some scholars or laypeople mean only "reasoning", "decision making" or boring shallow boolean logic - because that's what they can understand, and "art is something different, about emotions, blah-blah!"; others include creativity and art, and wonder how could the others put creativity in a different domain, since it's obviously essentially all the same.

Those things are "general" for an observer who has universal/"general" enough capabilities so that he can understand the common among all apparently different domains. However, in a situation of domain-blind* evaluators "versatile" is somewhat more appropriate word, because these activities which in a versatile mind are internally following the same laws and are just relabeled, look to domain-blind observers as radically different, "radically novel" to each other, or "magical" - Magical Self Imrpover. :)

The meaning of versatile is also more clear, unlike the polysemy of "general", and "versatile" means both general-purpose and specific, when a domain is specified.

As I suggested in a recent post (I will unveil in a later work), almost none of humans possesses versatile intelligence/truly general intelligence, humans have only a subset of it, limited by severe thresholds.

Improver/Improvement underlines the developmental aspect and that the potential for improvement matters**, not the actual state of knowledge or skills at the moment of evaluation.


* "Artificial" is a redundant word
** Correlated with the differential of the progress
*** Universal AI, as defined by M. Hutter, or the Schmidhuber's Goedel Machine are related concepts, but the optimality and rewards are too vague concepts if forced to appliy to versatile environments - see also a recent article: Universal Artificial Intelligence - Matches of Todor Arnaudov's teenage works from 2001-2004 to later (2005-2006-2007) related more popular works by Marcus Hutter and Shane Legg (...) a new Challenge/Extension of the reasons for Herbert Simon's "Bounded Rationality" (Satisficing)

A VLSI doesn't have to be optimal compared to all other possible agents in specific environments, in order to be qualified as VLSI, it should be better if applied to versatile environment and should be strictly or virtually limitless/unbounded (or just to self-improve) in its development and given particular environmental resources for development.

Particular data/environment/time range have developmental "juice" - when the juice is over, there's no more stuff to learn or clues to get.


Думи: Изкуствен интелект, термини, мислещи машини, разумни машини, интелигентни, изкуствен разум, самоусъвършенстване, развитие, психологоия на развитието, детска психология, машината на Гьодел, AIXI, Маркус Хутер, Юрген Шмидхубер, оптимален, решавач, задачи, проблеми, съвършенство, прогрес, неограничен, безграничен,

No comments :