OpenTalks.AI - Антон Колонин, От узкого и слабого к сильному и общему

Ad8ae7af280edaecb09bd73a551b5e5f?s=47 OpenTalks.AI
February 21, 2020

OpenTalks.AI - Антон Колонин, От узкого и слабого к сильному и общему



February 21, 2020


  1. Copyright © 2020 Anton Kolonin, Aigents® 1 From weak and

    narrow to strong and general Anton Kolonin, Ph.D. AIGENTS
  2. Copyright © 2020 Anton Kolonin, Aigents® 2 194? 20?? Adaptive

    Programmable Autonomous Guided Strong Weak Human-level AI (HLAI) Big Data, Machine Learning, Experimental Statistics Super- human AI Artificial General Intelligence (AGI) AI – where are we now? 2019 Narrow Artificial Intelligence (AI)
  3. Copyright © 2020 Anton Kolonin, Aigents® 3 General Intelligence: Reaching

    complex goals in different complex environments, using limited resources and minimizing risks (Ben Goertzel, Pei Wang, et al.) Intelligent Being
  4. Copyright © 2020 Anton Kolonin, Aigents® 4 Biological Intelligence: Reaching

    food and parents for self- reproduction in natural environments using limited physical resources and minimizing existential risks + - + -
  5. Copyright © 2020 Anton Kolonin, Aigents® 5 509 Bandwidth Limit

    Exceeded 509 Bandwidth Limit Exceeded 429 Too many requests 429 Too many requests 404 Not found 404 Not found Personal Internet Assistant Aigents®: Reaching information in online environments using limited CPU and RAM resources, minimizing risk of being banned AIGENTS®
  6. Copyright © 2020 Anton Kolonin, Aigents® 6 Current AI/AGI frontiers

    • Neuro-Symbolic integration – progress in 2019 • Explainable AI – progress in 2019 • Transfer learning – progress in 2019 • One shot (few-shot) learning • Strong generalization • Generative models • Structured prediction and learning • Fighting catastrophic forgetting (and catastrophic remembering) • Incremental learning and life-time learning • New “Turing Test” (e.g., “Baby Turing Test”) • Solving the “consciousness” problem
  7. Copyright © 2018 Anton Kolonin, SingularityNET 7 Incremental development and

    Baby Turing Test Smile Smile me You smile me You smile at me You smile at me friendly You should smile at me friendly Learning curve Lexicon size Grammar complexity Corpus size
  8. Copyright © 2020 Anton Kolonin, Aigents® 8 Bridging the Symbolic-Subsymbolic

    gap for “explainable AI” and “transfer learning” Hooves Tail White Black Brown AND AND AND OR Red Σ Σ Σ Σ Σ Σ (Hooves AND Tail) AND ((White and Black) OR Brown) => Horse 0.5 0.5 1.0 1.0 1.0 0.5 0.5 0.5 0.5 Transfer Explain 1.0
  9. Copyright © 2020 Anton Kolonin, Aigents® 9 Bridging the Symbolic-Subsymbolic

    gap in NLP between distributed representations and formal grammars with ontologies I saw the rusty saw Sp*I A AND Mg sawing Σ Σ Σ Σ Σ Σ 0.5 1.0 0.3 0.3 0.5 1.0 0.5 0.5 0.5 Transfer Explain 0.3 +--------Os-------+ | +-----Ds-----+ +Sp*i+ | +---A--+---Mg--+ | | | | | | I.p saw.v the rusty.a saw.n sawing.v Ds Os AND OR I saw the rusty saw sawing Noun Noun 0.3 DNF Soft-DNF “Disjunct” “Link” Link Grammar:
  10. Copyright © 2019 Anton Kolonin, SingularityNET Foundation, Aigents® 10 Grammar

    Ontology from Parses Gutenberg Children corpus
  11. acetaminophen may significantly reduce feelings of existential anxiety acetaminophen tylenol

    aspirin acetylsalicylic acid reduce reduces OR OR treat treats OR OR existential anxiety frustration OR AND “positive effect” “drug” “disease” OR “procedure” “negative effect” OR NOUN VERB AND head cold Colored nodes may be thought as «activated» or «expressed» Aigents® “Deep Patterns” - Language Model
  12. <set> := <disjunctive-set> | <conjunctive-set> | <M-skip-N-gram> <disjunctive-set> := {

    <pattern> * } <conjunctive-set> := ( <pattern> * ) <N-gram> := [ <pattern> * ] <pattern> := <token> | <regexp> | <variable> | <set> Example: {[$description catheter] [$coating coating] [$inner-diameter {diameter inner-diameter}] [$tip tip] [$pattern pattern]} X “Convey Guiding Catheter. Unique hydrophilic coating. Small atraumatic soft tip. Ultra-thin 1 × 2 flat wire braid pattern” = { coating : "hydrophillic", description : "convey guiding", pattern : "ultra-thin 1 × 2 flat wire braid", tip : "soft" } Aigents® “Deep Patterns” - Text Mining Variables may have domain restrictions in ontology and/or refer to other patterns as subgraphs
  13. Applying theory of functional systems for purposeful activity learning (P.Anokhin)

    AFFERENT SYNTHESIS Set PR of rules P 1 , …,P k , A  G that forecast achievement of the goal G in situation P 1 ,…,P m if fulfil A and subgoals P 1 ,…, P n DECISION MAKING Selection of actions A and subgoals P 1 ,…, P n , providing maximum probability of the goal G achievement ACTIONS RESULTS ACCEPTOR Anticipation of result G RESULT EVALUATION G  R Reinforcement/ punishment RESULT R Backward afferentation Motivation- Request to achieve a goal G Afferentation about situation P 1 ,…,P m Subgoals P 1 , …, P n RESULTS R 1 ,…,R n Actions A Forecast for goal G achievement Forecast for result G RESULT R Forecast for subgoals P 1 ,…, P n     1 k 1 n 1 n Pr ob(G | P ,...,P ,P ,...,P , A ) Pr ob( rule ) Pr ob( P ) ... Pr ob( P ) Evgenii Vityaev, Alexander Demin: Adaptive Control of Modular Robots // Conference Paper in  Advances in Intelligent Systems and Computing, Conference: First International Early Research Career Enhancement School on Biologically Inspired Cognitive Architectures, Springer, August 2018 Evgenii E. Vityaev: Purposefulness as a Principle of Brain Activity // Anticipation: Learning from the Past, (ed.) M. Nadin. Cognitive Systems Monographs, V.25, Chapter No.: 13. Springer, 2015, pp. 231-254. Витяев Е.Е. Логика работы мозга. Подходы к моделированию мышления. (сборник под ред. д.ф.- м.н. В.Г. Редько). УРСС Эдиториал, Москва, 2014г., стр. 120-153.
  14. G The learning causal relations along the dendrites and the

    goal feature G may be presented by semantic probabilistic inference – it add to the premise of causal relation all new features that increase the conditional probability of the goal feature G excitation. Vityaev E.E. A formal model of neuron that provides consistent predictions // Biologically Inspired Cognitive Architectures 2012. Proceedings of the Third Annual Meeting of the BICA Society. 196, Springer: Heidelberg, New York, Dordrecht, London. 2013, pp. 339-344. E.E. Vityaev, L.I. Perlovsky, B.Y. Kovalerchuk, S.O. Speransky. Probabilistic dynamic logic of cognition // Biologically Inspired Cognitive Architectures. Special issue: Papers from the Fourth Annual Meeting of the BICA Society (BICA 2013), v.6, October, Elsevier, 2013, pp.159-168 Semantic probabilistic inference as the formal model of neuron (E.Vityaev)
  15. C. elegans nematode model learning A.V. Demin, E.E. Vityaev. Learning

    in a virtual model of the C. elegans nematode for locomotion and chemotaxis // Biologically Inspired Cognitive Architectures. (2014) v.7, pp.9–14.
  16. Copyright © 2020 Anton Kolonin, Aigents® 16 AIGENTS Thank

    you stay in touch! Anton Kolonin, Ph.D. Telegram: @AGIRussia Anton Kolonin, Ph.D. Anton Kolonin, Ph.D.