Kunstig intelligens! Del I: Klassisk KI Mikkel Willum Johansen! mwj@ind.ku.dk! Adjunkt ved Institut for Naturfagenes Didaktik, Københavns Universitet
Plan 09.00-09.30: Oplæg 1: Klassisk KI! 09.30-10.10: Gruppearbejde 1! 10.10-10.20: Pause! 10.20-10.50: Oplæg 2: Ny KI! 10.50-11.00: Pause! 11.00-11.45: Gruppearbejde 2! 11.45-12.00: Opsamling 2
Klassisk kunstig intelligens GOFAI - motivation Euclid The Elements, Book V, Proposition 18! If magnitudes be proportional separando, they will also be proportional componendo (p. 427 in Heath, 2006). Proof! Let AE, EB, CF, FD be magnitudes proportional A E B separando, so that, as AE is to EB, so is CF to FD; I say that they will also be proportional componendo, C F D that is, as AB is to BE, so is CD to FD.!! For, if CD be not to DF as AB to BE, then, as AB is to BE, se will CD be either to some magnitude less than DF or to a greater. First, let it be in that ratio to a less magnitude DG. Then, since, as AB is to BE, so is CD to DG, they are magnitudes proportional componendo, so that they will also be proportional separando. Therefore, as AE is to EB, so is CF to FD. But also, by hypothesis, as AE is to EB, so is CF to FD. Therefore also, as CG is to GD, so is CF is to FD. But the first CG is greater than the third CF; therefore the second GD is also greater than the fourth FD. But it is also less: which is impossible. Therefore as AB is to BE so is not CD to a less magnitude than FD. Similarly we can prove that neither is it in that ratio to a greater: it is therefore in that ratio to FD itself. Therefore, etc. (Heath, 2006, pp- 427)!
Klassisk kunstig intelligens GOFAI - motivation Euclid The Elements, Book V, Proposition 18! If magnitudes be proportional separando, they will also be proportional componendo (p. 427 in Heath, 2006). Proof! Let AE, EB, CF, FD be magnitudes proportional separando, so that, as AE is to EB, so is CF to FD; I say that they will also be proportional componendo, that is, as AB is to BE, so is CD to FD.!! Skift til abstrakte symboler!! Sætning A C Bevis: E F B D Francois Viète! (1540-1603)
Klassisk kunstig intelligens GOFAI - motivation Tænkning som symbolmanipulation Characteristica universalis: Formelt sprog, der kan gøre generel tænkning til en formel kalkyle, som den man ser i matematik Gottfried Leibniz! (1646-1716) Automatisk symbolmanipulation... Pascaline (ca. 1642)! + og - Staffelwaltze (ca. 1671)! +,-, * og / Babbages Analytical engine! Første programmerbare computer.! Udtænkt ca 1837.
Klassisk kunstig intelligens GOFAI Within ten years a digital computer will discover and prove an important mathematical theorem.! (p. 7 in Simon & Newell (1958): Heuristic problem solving: The next advance in operations research. Operations Research 6(1), 1 10.! (Newell & Simon (1963): GPS, a program that simulates human thought, in Feigenbaum (ed.): Computers and Thought.
Tænkning som heuristisk søgning General Problem Solver! (Newell, Shaw & Simon 1959) objekt objekt objekt objekt (Newell & Simon: GPS, a program that simulates human thought pp. 279-293 in Feigenbaum & Feldman: Computers and Thought, AAAI Press, 1995. Side 281)
Tænkning som heuristisk søgning General Problem Solver! (Newell, Shaw & Simon 1959)??? (Ibid. side 287)
Tænkning som heuristisk søgning General Problem Solver! (Newell, Shaw & Simon 1959) Protokol fra testperson! Well, looking at the left hand side of the equation, first we want to eliminate one of the sides by using rule 8. It appears too complicated to work with first.! Now - no, - no, I can t do that because I will be eliminating either the Q or the P in that total expression. I won t do that at first. Now I m looking for a way to get rid of the horseshoe inside the two brackets that appear on the left and right sides of the equation. And I don t see it. Yeh, if you apply rule 6 to both sides of the equation, from there I m going to see if I can apply rule 7.! (ibid. side 282) (Ibid. side 287)
Tænkning som heuristisk søgning A B B 01001001001010010000???? A B A B?? A Formelt system B 0100101 A A B A B
Klassisk kunstig intelligens GOFAI To hovedteser i GOFAI 1) Formelle systemer er tilstrækkelige til at skabe intelligens! 2) Alle former for intelligens er i bund og grund formelle systemer (Newell & Simon: Computer Science as Empirical Inquiry: Symbols and Search, 1976)
Klassisk kunstig intelligens GOFAI Første intelligente robot: Shakey (1970)
Forudsigelser og håb Within ten years a digital computer will discover and prove an important mathematical theorem.! (p. 7 in Simon & Newell (1958): Heuristic problem solving: The next advance in operations research. Operations Research 6(1), 1 10.! "It is not my aim to surprise or shock you - but the simplest way I can summarise is to say that there are now in the world machines that can think, that can learn and that can create. Moreover, their ability to do these things is going to increase rapidly until - in a visible future - the range of problems they can handle will be coextensive with the range to which the human mind has been applied."! Herbert Simon (1957). Citeret i Crevier: AI: The tumultuous history of the search for artificial intelligence, (Basic Books, 1993), side 1.! Machines will be capable, within twenty years, of doing any work that man can do! Herbert Simon (1965): The Shape of Automation for Men and Management, (New York: Harper & Row), side 96.! Within a generation the problem of creating 'artificial intelligence' will substantially be solved! Marvin Minsky (1967), citeret i Dreyfus (1985): From Socrates to Expert Systems: The Limits and Dangers of Calculative Rationality.
Hvordan er det gået? Område: Matematik
Hvordan er det gået? Hyperblen nærmer sig mere og mere x-aksen, jo længere vi går væk fra origo. (Currant and Robins citeret i R. Núñez: Do Real Numbers Really Move? Language, Thought, and Gesture: The Embodied Cognitive Foundations of Mathematics )
Hvordan er det gået? Konceptuel blending Domæne 1 Domæne 2-2 -1 0 1 2 Rum! (linje) Tal Tal-! linjen Blended domæne
Hvordan er det gået? Venn-diagrammer (Jacobson: Basic algebra, (New York: W.H. Freeman and Company, 1985), p. 4.)
Hvordan er det gået? Venn-diagrammer
Hvordan er det gået? En sætnings liv 1: Heuristic behandling 2: Kontrolbehandling 3: Artikel (Misfeldt: Mathematical writing,(copenhagen: Danish University of Education, Learning Lab Denmark. (Online publication avalible at http://www.dpb.dpu.dk/dokumentarkiv/publications/ 20051219135403/ CurrentVersion/Mathematicalwriting.pdf) p. 23, 24 and 25)
Turingtesten Computing Machinery and Intelligence. 1950 Alan Turing! (1912-1954)
Problemer - Turingtesten CAPTCHA! (Completely Automated Public Turing test to tell Computers and Humans Apart)
Problemer - Turingtesten Alicebot (www.alicebot.org) judge: How do you smell?! ALICE: It is a ritual I do the same way every time.! judge: I smell with my nose.! ALICE: What does it smell like? Dommer: Can we talk personally?! AliceBot: Do you mean you and me? I speak only English right now, but I d like to learn some other languages. Can you teach med a few phrases in personally?
Plan 13.00-09.30: Oplæg 1: Klassisk KI! 13.30-14.10: Gruppearbejde 1! 14.10-14.20: Pause! 14.20-14.50: Oplæg 2: Ny KI! 14.50-15.00: Pause! 15.00-15.45: Gruppearbejde 2! 15.45-16.00: Opsamling Gruppearbejde 1! (Opgave 1: GPS og logikken)! Opgave 2: Turingtesten! Opgave 3: Det kinesiske værelse!! Vælg og arbejd med mindst to af spørgsmålene i grupper på 2-3 personer
Plan 09.00-09.30: Oplæg 1: Klassisk KI! 09.30-10.10: Gruppearbejde 1! 10.10-10.20: Pause! 10.20-10.50: Oplæg 2: Ny KI! 10.50-11.00: Pause! 11.00-11.45: Gruppearbejde 2! 11.45-12.00: Opsamling
Oplæg 2: Fast, cheap and out of control!
Fast, cheap and out of control! 1986: To nye paradigmer Agentbaseret ny-ki Kunstige neurale netværk
Neurale netværk Input Kunstig neuron! (hjernecelle) Tænke, tænke...! Fyre eller! ikke fyre? Output Vægte
Neurale netværk Minesøger Problemer Kompleksitet. Hjernen: 100 milliarder nerveceller! 100.000 milliarder forbindelser Realistisk model for læring?
Ny kunstig intelligens Klassisk KI:! Oppefra og ned! Ny KI:! Nedefra og op... (ca. 3,5 milliarder år siden) (ca. 500 millioner år siden) (250 millioner år siden) (2,5 millioner år siden)
Ny kunstig intelligens William Grey Walters væsner
Ny kunstig intelligens Klassisk paradigme Nyt paradigme Ydre Indre Sansning Handling
Ny kunstig intelligens
Ny kunstig intelligens Allen Målrettet undersøgelse Opdagelse Undvige
Ny kunstig intelligens Herbert
Ny kunstig intelligens Problemer! Kan man virkelig alt uden indre modeller? Boss
Ny kunstig intelligens Problemer! Kan man virkelig alt uden indre modeller? Boss Chris Urmson et al. (2009) Autonomous Driving in Traffic:! Boss and the Urban Challenge, Association for the Advancement of Artificial Intelligence, the AI Magazine, side 20
Ny kunstig intelligens Problemer! Kan man virkelig alt uden indre modeller?! Kompleksiteten vokser meget hurtigt!! Det er svært at overføre færdige moduler fra en robot til en anden Allen
Ny kunstig intelligens Problemer! Kan man virkelig alt uden indre modeller?! Kompleksiteten vokser meget hurtigt!! Det er svært at overføre færdige moduler fra en robot til en anden Den naturlige evolutions drivkraft:! Tilfældig variation! Naturlig selektion! (survival of the fittest) Ny KI s drivkraft?! Nøje planlægning! Debugging Ny KI:! Nedefra og op... (ca. 500 millioner år siden) (250 millioner år siden) (2,5 millioner år siden)
Evolutionær programmering Program 1: Program 2:... Første generation Anden generation Tredje generation 10000 10% bedste 10000 10000 10%... bedste (rigtig, rigtig mange gange) 0 0 0
Evolutionær programmering John Muir-stien: Resultat af typisk eksperiment
Evolutionær programmering Evolutionært udviklet Tronspiller! Prøv selv på:! www.demo.cs.bran deis.edu/tron/
Evolutionær programmering Problem! Hvem bestemmer hvilke programmer der er bedst og hvordan?
Selvstrukturerende adfærd Swarmbots! http://www.swarm-bots.org/ Fra: Shervin Nouyan og Marco Dorigo: Chain Based Path Formation! in Swarms of Robots, side 3
Robotterne kommer! Næppe Hvad den kunstige intelligens mangler! Forståelse af kognitive, kropslige og omgivelses-resurser! Indlevelsesevne! Analogi (brug af eksisterende viden fra andre områder)! Fantasi (kreativ brug af eksisterende viden)! Planlægning! Følelser! Sprog! Alt det, vi ikke engang ved, at vi ikke ved
Gruppearbejde 2! Opgave 1: Ny kunstig intelligens! Opgave 2: Evolutionære algoritmer! Opgave 3: Sværmintelligens (engelsk)! Opgave 4: Fremtiden!! Vælg og arbejd med mindst to af spørgsmålene i grupper på 2-3 personer Plan 13.00-09.30: Oplæg 1: Klassisk KI! 13.30-14.10: Gruppearbejde 1! 14.10-14.20: Pause! 14.20-14.50: Oplæg 2: Ny KI! 14.50-15.00: Pause! 15.00-15.45: Gruppearbejde 2! 15.45-16.00: Opsamling