Algemeen Jacques Ellul Günther Anders Lewis Mumford Herbert Marcuse Andrew Feenberg Wiebe Bijker Algemeen Luddieten Wapentuig als voorbeeld Transport Radio en televisie Internet Artificiële intelligentie Algemeen Hypermedia en hypertext Informatieovervloed Privacy Spionage, inlichtingendiensten Cyberoorlog Hackers Algemeen Het schrift Boeken en tijdschriftenMcCorduck besloot een geschiedenis van de AI te schrijven toen de meeste van de eerste generatie AI-pioniers nog leefden. De eerste editie verscheen in 1979.
De insteek is op verschillende manieren persoonlijk. Bijvoorbeeld in die zin dat McCorduck ook veel aandacht schenkt aan de mensen achter het AI-werk. Daar is niets misd mee. Maar ik vind haar ook persoonlijk in de zin van 'persoonlijke vooroordelen' hebben die ze niet los kan laten. Ze is niet in staat tot fundamentele kritiek op de AI. Ze wijst soms kritisch op beperkingen van de AI, maar blijft desondanks enthousiast en naïef op het spoor zitten van dat AI zo belangrijk is voor ons mensen en dat het allemaal wel goed komt in de toekomst.
Vijfentwintig jaar later, in 2004, verscheen deze tweede editie. De auteur constateert dat er veel veranderd is in de wereld van de AI. AI is nu een internationaal fenomeen geworden met veel onderzoekers in andere landen dan de VS. Maar ze houdt het in haar bespreking bij de VS. Ook dat is nogal typisch voor McCorduck, vind ik.
Pamela McCORDUCKMcCorduck bedacht dat het zinnig zou zijn om prominenten in de AI te interviewen om te komen tot een soort van geschiedenis van de AI. Maar ze was niet opgeleid in 'history of science' en kon moelijk officiële ondersteuning krijgen. Ze was er verbaasd over dat mensen uit die discipline nog helemaal niet bezig waren met zo'n inventarisatie. Uiteindelijk kwam het boek er toch, in 1979. Deze editie met een nieuw nawoord verscheen in 2004, 25 jaar later.
"This book is a history of artificial intelligence, that audacious effort to duplicate in an artifact what we humans consider to be our most important, our identifying, property — our intelligence." [mijn nadruk] (xvii)
[Dat lijkt me helemaal niet zo'n handige aanpak. Waarom zou je een mens kopiëren in plaats van een machine te maken die op zijn of haar eigen manier dingen uitvoert. Dat lijkt me tot oponthoud leiden, omdat je steeds de mens als maatstaf hanteert in plaats van de machine. Het leidt ook al gauw tot antropomorf gepraat over machines. De auteur praat verderop al over "intelligent artifacts" alsof dat een probleemloze uitdrukking is.]
"Another example: there’s a distinct American bias in my history, which may have been appropriate in earlier days, but is no longer. Vigorous research groups now exist in Europe, the Soviet Union, and Japan, and the British effort echoes throughout the history of artificial intelligence, sometimes merrily, sometimes sadly, but always a presence." [mijn nadruk] (xvii)
[Het toegevoegde deel Afterword is wel internationaal.]
De aanpak is persoonlijk qua keuzes en ook persoonlijk door zijn aandacht voor de betrokken mensen. Ze bedankt ze uitvoerig.
[Ik vraag me af wat ze daarmee bedoelt: het denken mechaniseren? Het gaat over logica, over machines / programma's die kunnen rekenen en via die weg allerlei taken kunnen uitvoeren.]
Hier wordt een historisch verhaal neergezet - in overeenstemming met de zojuist gegeven tijdlijn - over allerlei manieren waarop mensen en menselijke functies (als rekenen en zo) afgebeeld / nagemaakt / geïmiteerd werden.
[Ik ga dat niet weergeven en beperk me hier tot een paar kwesties.]
"“Can a machine think?” This question is in a class with those snappy vaudeville comebacks: does a chicken have lips? And like them, it ought to end the discussion at once by its self-evident nonsense. After all, we agree, our one essential, identifying property is thinking."(3)
[Precies. En bleef het daar maar bij, want de vraag alleen al deugt niet. Maar helaas ... Dat mensen zichzelf op allerlei manieren afbeelden en 'reproduceren' vindt McCorduck fascinerend. Dat mag. En het is leuk dat mensen vanaf de vroegste tijd automata maakten. Maar zegt dat nu iets over AI?]
"From a philosophical point of view, the Arab heirs to the Hellenes may have been the first to state formally that a distinction existed between natural and artificial substances. Such a distinction did not mean that the natural was superior to the artificial, only different. But not very different: one excellent means of knowing the natural was declared to be by studying the artificial (Labat, 1963). We’ll hear the same assertion from modern workers in artificial intelligence." [mijn nadruk] (10)
[Ik denk dat dat een fundamentele denkfout is.]
"Vaucanson had been interviewed for an encyclopedia of the sciences in 1777, and was deliberately vague about the secrets of the duck’s insides. He did not intend, he said, a perfect imitation of the digestive processes, with nourishment and blood manufacture, and so forth; rather, he hoped to imitate the larger aspects, which would include intake, maceration, and an obvious chemical change prior to excretion (Chapuis and Droz, 1949). Vaucanson’s aim serves as a good example of simulation, and his duck should be kept in mind when we come to the simulation of human thought processes." [mijn nadruk] (17)
[Verstandige man, die Vaucanson. Hij heeft niet de pretentie een perfectie kopie te maken van het beest. Dat zou ook moeten gelden voor de simulatie van menselijke intelligentie, denkprocessen, en zo verder. Ik denk dat veel mensen in de AI - in ieder geval vroeger, maar ik vraag me af of het niet nog steeds zo is - wel die pretentie hadden dat een machine een perfekte kopie van mensen kon zijn. Die insteek leidt tot zinloze discussies die alleen maar tijd en geld en energie kosten.]
"One statement of Lady Lovelace’s has often been quoted: “The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform.” And this statement has been adduced as evidence that machines cannot, in any way, be said to think. It’s a true statement, but a misleading one, and will bear some looking into later on." [mijn nadruk] (33)
[Hoe kan iets wat waar is misleidend zijn?]
"True to its speculative origins, artificial intelligence poses a set of grave moral questions while, true to its claims to be a science, it promises answers to puzzles about the nature of intelligence."(35)
Dit hoofstuk bespreekt denkers over de lichaam-geest-kwestie in relatie tot de genoemde 'mechanisering van het denken'. Bijvoorbeeld Descartes dualisme.
"A history of psychology is nothing less than a history of the effort to explain what many people still regard as ultimately inexplicable, the workings of the human mind. As philosophers, and later psychologists, illuminated bits and pieces of the mechanism, led here by speculative argument and prompted there by empirical evidence, the notion of mind slowly gave way to the notion of brain function. This latter idea suggests that the brain is an organ, fully as explainable as the pancreas or the kidneys, though surely more complex than even those awesomely complicated structures." [mijn nadruk] (37)
[De hersenen scheiden gedachten en gevoelens en bewustzijn en intelligentie af zoals de pancreas insuline? Wat was die uitspraak ook weer? Maar wat weten we tegenwoordig nu eigenlijk van die hersenen en van dat proces?]
" ... efforts were made to find a universal grammar, common to humans and computers alike."(41)
[Ook die illusie - gebaseerd op het idee 'computer = kopie van mens - heeft tot eindeloos veel tijdverlies geleid. Het is nog steeds Newtoniaans: 'reasoning from mechanical principles', dus een gedateerde manier van kijken naar de menselijke geest.]
"The notion of human as a machine is a heuristic — a rule of thumb or point of view, from the same Greek root as eureka — to aid in an understanding of how human beings work. To call us machines does not define us, and it will not “encompass the human essence.”"(44)
[Nee, maar die heuristiek kan er wel helemaal naast zitten en leiden tot een manier van kijken die mensen / onderzoekers op verkeerde ideeën brengt. Zoals ook metaforen - bijvoorbeeld allerlei technische metaforen voor de menselijke geest of andersom allerlei antropomorfe metaforen voor machines - veel meer invloed hebben op onze kennisverwerving dan we weten.]
"While Binet was confident that he knew what intelligence was, and moreover that he could measure it, endless debate ensued over how humans accomplished what everyone “knew” was intelligent behavior."(47)
"We’ll say more about Shannon’s role further along. Now it is only necessary to say that it wasn’t farfetched to ask if, since the laws of thought could express the behavior of electronic circuits, electronic circuits could express thought."(50)
[Logica en wiskunde als de wetten van het denken. Maar het denken verloopt niet alleen maar via logische wetmatigheden, dat is al een reductie van jewelste. En het is typisch dat die ideeën komen van logici en wiskundigen, allemaal mannen indertijd. Dat het gedrag van elektronische circuits beschreven kan worden met logische symbolen is prima, maar zo gauw vandaaruit weer gezegd wordt dat die elektronische circuits ons iets leren over het denken is er een versterking van die reductie. Dit zijn echt enorme redeneerfouten.]
"Or is human thinking more various, encompassing the rigor of logic among several — maybe many — other modes of thought? Is it basically irrational to exclude the irrational as a component of thinking? Are there several kinds of irrational, that is, nondeductive, nonlogical ways of thinking which form an essential part of human cognition?"(50)
[Ja, dat is zo. Maar dat wordt nu niet uitgewerkt.]
"The cybernetics theory of thought, with feedback as a central notion, seemed to promise the possibilities of imitating human cognition by modeling systems of neurons. This modeling was not based on detailed biological knowledge of the natural cell, which we didn’t have (nor for such purposes do we yet have). Instead, it seemed certain that the correspondence between the on-off behavior of the neuron and the on-off behavior of the electronic switch would be sufficient to allow significant modeling of neural systems, and then intelligent behavior. Its basic assumption was that brain cells were on the whole general-purpose, organized for specific functions because of external stimuli. McCulloch was convinced of it, and his forceful personality inspired a flock of young researchers around him at MIT, where he eventually moved, to work at developing such systems. Sadly for him and for them, the approach was to prove sterile."(56)
[Je weet allerlei dingen niet over de hersenen, je neemt aan dat neuronen werkelijk aan-uit-schakelen - wie weet is dat helemaal niet wat er gebeurt al lijkt dat voor de oppervlakkige waarnemer zo -, en dan blijkt de aanpak niet te werken. Tja.]
"The honor for the first general-purpose program-controlled digital computer up and running (and we’re talking about mat- ters of months here) doesn’t seem to belong to the Americans, who have long claimed it, or even to the British, who have recently been declassifying evidence that they were ahead of the Americans. Instead it goes to Konrad Zuse, who was then a young German engineer. Nearly all of his machines were destroyed dur- ing the war, but patent applications seem to support Zuse’s claim." [mijn nadruk] (61)
[Hè hè, dit begint zelfs in de VS door te dringen. ]
Iets over Zuse. Meer over Turing.
"In 1942, Turing made the risky trip across the Atlantic to the United States, where he remained from November to the following March. Here he seems to have talked to John von Neumann, whom we’ll consider later in this chapter, but there’s no direct evidence that he learned of other work close to his own, such as the McCulloch-Pitts paper, or the Rosenblueth-Wiener-Bigelow paper. But with von Neumann, there was evidently a fruitful exchange of ideas."(66)
"He asked permission to take a year’s sabbatical at King’s, and there, in September 1947, he wrote a paper which has only recently been published, called “Intelligent Machinery” (Turing, 1969). In the spring of that year, Norbert Wiener visited him at Teddington where, according to Wiener, they talked over the fundamental ideas of cybernetics. Talk with Wiener surely stimulated Turing, but the thoughtfulness of the 1947 paper seems to show that Turing had already been speculating about the possibilities of intelligent machinery before their meeting."(67)
"Turing soon comes to the central issue, man as machine. He points out various machines that already imitate parts of man — the television camera as eye, the microphone as extended ear, and servomechanistic robots that imitate actions of the limbs, and so forth."(68)
[Ja, maar zeggen we: de camera kan zien? de microfoon kan luisteren? Nee. Waarom dan wel: de computer kan denken? Dat machines dingen van mensen imiteren is prima, soms kunnen ze iets wat mensen plegen te doen zelfs beter dan mensen (zoals schaken). Maar dat betekent niet automatisch dat ze alles van mensen kunnen imiteren en evenmin dat machines alles beter kunnen dan mensen.]
"he suggests that a “brain” be built that might apply itself to any of the following tasks: games such as chess, tic-tac-toe, bridge, and poker; the learning of languages; the translation of languages; cryptography; and mathematics. This list turns out to be the main part of the program that occupied artificial-intelligence researchers for the next two decades."(68)
[ Alles wat logisch is werkt dan redelijk. Maar bijvoorbeeld talen leren of vertalen helemaal niet zo, omdat taal niet alleen maar logisch is. ]
"A paper of Turing’s that was published just after it was written, and that received wide attention, came out in October 1950. It was called “Computing Machinery and Intelligence,” and it addressed the question of whether machines could think. Here Turing proposed what came to be known as Turing’s Test ..."(70)
"Officially, Turing was working on the design of Madam [Manchester Automatic Digital Machine - GdG] at Manchester, but at the same time he’d become fascinated by biology. He was invited to join the Ratio Club (named by Albert M. Uttley, who consulted an etymological dictionary and liked the way it combined reasoning, relations, and number), an informal group of physiologists, physicists, mathematicians, and engineers who met periodically in London to discuss problems of mutual interest. He submitted two papers to the group, evidence of his belief that a mathematical inquiry into some biological problems might yield interesting results." [mijn nadruk] (72)
"Nevertheless, Turing refused to be misled by the obvious differences he could see between the digital computer and human neural physiology. He understood that at another level, these days called the information-processing level, the brain and the computer had much in common, and that insight into the organization of one would surely give insight into the organization of the other." [mijn nadruk] (74-75)
["Much in common"? Wat dan? Want het antwoord op die vraag is nogal belangrijk. Als dat zoiets is als: 'er kan zowel met hersenen als met een computer logisch geredeneerd worden' dan ja, als het gaat om andere zaken die we met de hersenen doen is dat nog maar de vraag. En inzicht in de 'organisatie van de hersenen' - wat dat ook is - kan volgens mij geen inzicht geven in de organisatie van de computer, en andersom al helemaal niet.]
Vervolgens gaat het over John von Neumann
"Another intellectual giant associated with computers did not share Turing's optimism about whether they would eventually think. He was John von Neumann, the extraordinary mathematician who was born in Budapest, and had come to the United States in 1930 ..."(76)
Voetnoot 4
"So far as I can discover, von Neumann was the first to do computer anthropomorphizing on such a scale (an oddity, considering his lifelong skepticism about the connection between human brains and computing). Edsger Dijkstra, the well-known Dutch computer scientist, considers this habit pernicious, and continuing indulgence of it a big obstacle to the maturation of the entire field of computer science. “I should have guessed it was Saint Johnny,” he grouched to me once. Others do not share his view." [mijn nadruk] (77)
[Ik ben het eens met Dijkstra. ]
"By now [zeg 1950 - GdG] the term “giant brains” had captured the public imagination. That computers could do low-level sorts of intellectual tasks, such as arithmetic and collation, seemed to suggest to the overly enthusiastic that high-level intellectual behavior was just around the corner."(79)
[Tja, als de media er zich mee gaan bemoeien gaat het fout. Dan krijg je verkeerde metaforen en antropomorf taalgebruik en zo. ]
"Von Neumann was obviously fascinated with the idea, but he simply saw no ultimate way that the connection between human thinking and machine performance could be made. In a paper written in 1951, he took pains to point out the differences between the human nervous system and a computer." [mijn nadruk] (79)
"Perhaps it isn’t entirely surprising that neither von Neumann nor his friends were persuaded to view the two systems they were confronted with, namely human beings and computers, as two instances of information processors, and to examine a grosser level of functioning than cells and diodes. But a younger group of scientists, in many cases the students of these same men, did perceive those functional similarities, and at least in the case of von Neumann and Wiener, confronted their mentors with such a view." [mijn nadruk] (82)
"The information-processing model, with which artificial-intelligence research concerns itself, explains thinking in terms — unsurprisingly — of the processing of information in the human brain, representing it in symbolic form, storing it, the means of recalling it, controlling it, and so forth. This model often casts the act of thinking into terms of a problem to be solved, and examines the techniques by which thinkers, whether man-made or begotten, solve them." [mijn nadruk] (86)
[Dat tweede punt lijkt me al eenzijdig en alleen gelden voor bepaalde vormen van informatie. En hoe wordt 'informatie' gedefinieerd? ]
"Everything from symphonies to simultaneous equations to situation ethics is finally produced by those electro-chemical processes. This view can be considered mechanistic."(86)
[Zijn er in de hersenen alleen maar elektrochemische processen te vinden? En wat betekent 'mechanistisch' in dit verband?]
Volgt nu de bespreking van Warren McCulloch.
"one of the most influential persons in the germination of modern artificial intelligence was the neurophysiologist Warren McCulloch."(88)
"What they [McCulloch & Pitts - GdG] did was to provide a new definition of computing machine, which allowed the brain to be construed as a machine in a more precise way than before."(90)
[Welke richting gaat dat nu uit? Is het - van hersenen naar machine - om betere informatieverwerkende machines te kunnen bouwen? Of is het - van machine naar hersenen - om de hersenen beter te begrijpen door die als informatieverwerkende machine te zien? Dat laatste lijkt me heel aanvechtbaar.]
"Even in the late 1950s, when it began to become clearer — based on the evidence acquired by the microelectrode, which could be implanted in a single nerve cell — that McCulloch’s model of the neuron was misleadingly simple, he had a band of loyal followers who were busy through the mid-1960s building nets based on his early work (and even simplifications of it), nets which failed to yield any but the most trivial behavior." [mijn nadruk] (91)
"Far more important than McCulloch’s ideas was his certainty that mind could be known and described in scientific terms. He was an enormous influence on Minsky, much greater than, say, Turing, whose work Minsky knew somewhat, but regarded as somehow off to one side. It was McCulloch’s certainty that convinced Minsky and many another that machine intelligence was possible — and to get going on the problem." [mijn nadruk] (93)
[Wat McCullen zegt gaat over de 'mind' en dat we die wetenschappelijk kunnen kennen. Dat laatste punt - dat intelligente machines mogelijk zijn - is heel wat anders. Wat heeft het een met het ander te maken? Daarop heb ik nog geen bevredigend antwoord gezien.]
Bepaalde bijeenkomsten als die van de Teleological Society, de Josia Macy meetings, en wat later de Ratio Club werden belangrijk voor het uitwisselen van kennis en inzichten tussen mensen van verschillende disciplines met als studieobject de hersenen. Wat betreft die laatste:
"Except for Albert M. Uttley and Alan Turing, none of them had much connection with digital computers, and so the computer didn’t necessarily suggest itself as a useful model for the brain. Indeed, for many of them who were physiologists, the all-or-none logic that had so enchanted some researchers seemed, in MacKay’s words, perverse, because it didn’t provide for the kind of graded signals that the brain’s chemicals seemed to carry."(95)
"He tried to imagine a machine with the virtues of digital machines, which have great accuracy but, he felt, no way of representing gradations, and analog machines, which can represent gradations but are limited as to accuracy:
I was accustomed to getting a bit of flak from the computer-based chaps on the ground that it was all very well to talk in these terms but you couldn’t formalize it. And there was this great feeling that unless youcould formalize something mathematically, you weren’t really understanding it. I’ve resisted this all my life because I think it’s absolutely not true. There is so much of understanding which is essentially qualitative — the sort of aha! you get by seeing that somebody’s pulled the plug out of the bath and that’s why the water’s going down, and all sorts of analogous situations where formalization is just a waste of time." [mijn nadruk] (95-96)
[Kijk, dat is nu eens interessant. De overschatting van wiskunde, van formele talen, van formalisering in het algemeen, wordt zelden besproken. Heel belangrijk, vooral ook omdat die samenhangen met de dominantie van mannen in de ICT en AI. Het is een reductionistische manier van kijken, die misschien aardig werkt bij apparaten, maar grote nadelen heeft bij de menselijke hersenen en zijn functies.]
Het vervolg gaat over Marvin Minsky, die op een gegeven moment gaat samenwerken met Seymour Papert uit de UK.
"For Minsky it was all a little premature. His hope of reproducing mindlike behavior by simulating the physiology of the brain came to naught, except to earn him a Ph.D." [mijn nadruk] (103)
[Dat is dus richting apparatuur.]
"Case-Western’s Leon Harmon, who worked on the von Neumann machine at the Institute for Advanced Study at Princeton, and who describes himself as perhaps the first computer operator, still seethes about walking into the Smithsonian and discovering that beside the von Neumann machine, which well deserved to be there, stood a Perceptron, sharing floor space as if it were equally important. Harmon doubts that we’ll ever learn much about brain operation from studying electronic hardware, and believes that the really interesting and potent things the computer in our head does are inscrutable." [mijn nadruk] (105)
[Dat is dus richting hersenen. Harmon vindt dat je door het bestuderen van apparatuur weinig leert over de hersenen. Ik ben dat geheel met hem eens. ]
"And so much for the hope of making a machine think by trying as literally as possible to imitate the brain, the meat machine, at the cellular level. It didn’t work, as John von Neumann had said it wouldn’t. We didn’t then — nor do we now — know enough about how the brain works at the cell level to make such a model, and there’s presently speculation that in any event, a serial machine simply won’t be able to imitate what is very likely a series of parallel processes." [mijn nadruk] (107)
[Precies.]
"But researchers in artificial intelligence had ideas that there were other equally useful and effective ways of modeling or simulating thought, and it was these alternative models that were to preoccupy them in the future."(107)
[Ja, maar de vraag blijft hoe je naar iets kunt modelleren wat je niet kent. ]
"On the leafy campus of Dartmouth College in the summer of 1956, a handful of scientists met to talk about the work they were doing toward making machines behave intelligently. Although they came from different backgrounds — there were men trained as mathematicians, as psychologists, as electrical engineers — and although some worked for industry and others were at universities, they had in common a belief (more like a faith at that point) that what we call thinking could indeed take place outside the human cranium, that it could be understood in a formal and scientific way, and that the best nonhuman instrument for doing it was the digital computer.
Four of them had collaborated on a proposal to the Rockefeller Foundation which said, “We propose that a two-month, ten-man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”" [mijn nadruk] (111)
[Was er ook maar één vrouw bij? Het gaat trouwens inderdaad om geloof, een referentiekader dat zegt 'denken' = 'formeel en logisch', een benadering die 'denken' reduceert tot iets dat een computer kan uitvoeren. Het voorstel van John McCarthy, Marvin Minsky, en Claude Shannon - drie wiskundigen - is enorm pretentieus.]
"Antecedents were plentiful and varied, as this history has already shown, but they all pointed in the same direction, toward the idea that there was a rigorous and objective way of explaining the human intellect." [mijn nadruk] (114)
[En meteen gaan we weer in de andere richting, nu gaat het ineens niet om een machine die een bepaald formeel idee van 'denken' simuleert, maar om een strenge = formele manier om het 'menselijk denken' te verklaren.]
"McCarthy’s disappointment reflects a persistent problem with artificial intelligence — that making machines think, designing computer programs to behave intelligently, was far harder than anyone in 1956 thought it would be. Over two decades have imbued the field with more modesty than it had in its infancy, but the fact remains that the problems continue to be harder than anyone expected."(118)
Noot 2, p. 119
"Artificial-intelligence workers continually use machine when they mean what an outsider would call a program."(119)
[Au, dat is lekker verwarrend.]
"The hope that if Boolean algebra could be used to express the behavior of electrical switches, then conversely human thought might be expressible in the behavior of electrical switches occurred not only to Shannon but independently to others."(120)
[De bekende denkfout. Waarom zou je dat laatste zelfs maar willen? ]
"He [Shannon - GdG] too addressed the problem of whether such a machine could be said to think, and concluded, like Torres, that the definition of thinking was much too fluid to say for certain."(121)
Newell and Simon kwamen als met een opvallend programma: Logic Theorist, "a computer program that exhibited intelligent behavior"(125) Maar het kreeg niet de aandacht die het verdiende.
"Their work was certainly received with interest. But the evidence is that nobody save Newell and Simon themselves sensed the long-range significance of what they were doing."(124)
"Logic Theorist and the General Problem Solver were good models of how humans behaved. Minsky wasn’t sure if that was true and, furthermore, even if these were good models of human intelligence, he wasn’t necessarily convinced that human and artificial intelligence needed to resemble each other. This last was a very strong theme in AI research in the early years." [mijn nadruk] (126)
"So perhaps the most influential result of the Dartmouth Conference itself was the social patterns it set. Though Arthur Samuel was a participant in 1956, he’s one of many who have raised the question of professional nepotism. Accusations of clannishness have persisted since 1956, and they aren’t without foundation."(130)
ARPA financierde op AI-vlak voornamelijk
"Carnegie Mellon University, where Newell and Simon work; MIT, where Minsky works; Stanford, where McCarthy works; and Stanford Research Institute, which is heavily populated with former students of these Dartmouth Conference participants."(131)
"In any case, each of the four main centers was to evolve its own distinct style, set by the personalities of the men who dominated it."(133)
Carnegie Tech (later: CMU), de invloed van RAND en het vrije werk voor de Air Force, en alle briljante geesten die daarvoor aangetrokken werden. Allen Newell. Herbert A. Simon.
"It is important to remember that for Simon the mind as logic machine had preceded the computer as artifact:"(151)
"We've already seen that the ancient yearning to invent a double of the human brain had revived - lustier than ever - with the invention of the digital computer. Lurid predictions about giant electronic brains filled pulp science fiction and the Sunday supplements, guaranteeing that the reaction would be just as extreme. Norbert Wiener and John von Neumann felt obliged to deliver lectures warning against too facile an analogy between the digital computer and the human brain, and they were joined by many other distinguished scientists and engineers who assured the public and each other that computers were only high-speed morons, incapable of intuition, originality, or any other variety of intelligence." [mijn nadruk] (151-152)
"But in 1952 sober scientists ran a severe risk to their credibility when they moved beyond the realm of speculation and into actual work intended to simulate any sort of human thinking. A fair number of scientists today — not to mention members of the general public — continue to hold the view that human mental processes cannot be simulated by any machine, or at least not by the means presently proposed by workers in artificial intelligence. (...) I will explore the reasons for this opposition further on. Now the important point is that such feelings were at least as strong in the early 1950s. Those who thought otherwise were few, isolated from the moral support they might have offered each other and without any concrete results to give them courage to persevere." [mijn nadruk] (152)
"Aside from the fact that it worked — yhat it was proof positive a machine could perform tasks heretofore considered intelligent, creative, and uniquely human — it also exhibited some extremely clever solutions to programming problems that had plagued computer programmers and seriously hampered all kinds of computer applications." [mijn nadruk] (167)
[Tja, maar wat bedoel je daar dan mee? De machine zal die taken hoe dan ook anders uitvoeren dan de hersenen van een mens. Of ik zou kunnen zeggen: de intelligentie zit niet in de machine maar in het programmeerwerk door mensen vóór de machine.]
"Games are models of situations in life, just as physical models imitate, simplify, and express the essence of physical phenomena. In their collection of early papers on artificial intelligence, Edward Feigenbaum and Julian Feldman (1963) introduce the section on the machines that play games ..." [mijn nadruk] (172)
Arthur Samuel en een damprogramma.
"Samuel’s approach to the checkers-playing program was somewhat different from other learning programs. He did not believe that human learning processes ought necessarily to be imitated, because he believed that the differences between the human brain and the computer were simply too great." [mijn nadruk] (178-179)
[Verstandige man.]
Alex Bernstein en een schaakprogramma.
[Geklets over The Other, gepsychologiseer om het verzet tegen het idee 'denkende machines' te verklaren:]
"And I believe that this potent fear of The Other is one of the things that informs the violent reaction we have to the idea of thinking machines."(197)
[Ze lijkt niet te snappen dat het een kwestie van verkeerd taalgebruik is. Er bestaan geen 'denkende machines' en die zullen ook nooit bestaan. Dat is antropomorf taalgebruik dat vermeden moet worden. Machines 'denken' niet, maar voeren programma's uit, o.a. voor zaken die lijken op wat mensen doen. Waarom zo willen benadrukken dat het om 'denkende machines' gaat? Laten we daar eens over psychologiseren ... Dat taalgebruik leidt alleen maar tot misverstanden en zinloze discussies.]
"Defining intelligence and determining who might have it seem to be problems psychologists and semanticists ought to work out, but the notion that human beings might be just one instance of a larger class of intelligent beings is so provoking that everyone wants to get into the act with an opinion." [mijn nadruk] (198)
"It’s ironic that one scientist whom [Mortimer] Taube attacked with great gusto in his book [Computers and Common Sense - GdG], a Harvard professor named Anthony Oettinger, would himself write a book criticizing computer-aided instruction (an effort to accomplish some of the more pedestrian tasks of teaching with a computer terminal substituting for a human teacher). In a further disavowal of his youthful folly, and disillusioned by his participation in the failure of machine translation in the 1950s, Oettinger contributed the introduction to a book that was to be an even more detailed and sustained attack on artificial intelligence, a book called What Computers Can’t Do, by Hubert Dreyfus."(210)
[Samenvatting door deze auteur van Dreyfus' opvattingen. Zie hier mijn samenvatting van het boek, eerste en latere editie.]
"But none of this can be understood without looking at the circumstances of why Hubert Dreyfus has so persistently and passionately attacked artificial intelligence, and why the artificial intelligence community has responded with less than sweet reason at all times."(218)
[Er wordt enorm op de man gespeeld hier. Het is vooral ook een strijd van Dreyfus tegen alle pretentieuze voorspellingen, al het gebluf door de mannetjes in de wereld van de AI, een strijd tegen dat old-boys-netwerk waarover McCorduck het zelf eerder had. Ze konden niet goed tegen die kritiek. Het helpt dan niet erg als je steeds wijst op hoe ver "we" inmiddels al gekomen zijn met AI en hoeveel we computers al kunnen laten doen. De kwestie is principieel: wat ze ook doen, het is geen 'denken', geen 'intelligentie', etc. Je moet dat soort woorden gewoon niet gebruiken voor machines.]
"Thus, aside from some reviews of What Computers Can’t Do, which were necessarily too brief to address any but the grossest exceptions they took to Dreyfus, workers in artificial intelligence mainly take no public notice of him." [mijn nadruk] (235)
"With or without the advice of Dreyfus, AI programs are growing more complex, flexible, contingent, tolerant of ambiguities, and situation- and goal-oriented than they once were. My impression is that this progress has taken place piecemeal and in response to tough given problems, and owes nothing to Dreyfus. He may have been correct in some of his criticisms, but like that earlier critic Mortimer Taube, his derisiveness has been so provoking that he has estranged anyone he might have enlightened. And that’s a pity."(236)
[Typisch. Alsof al die besproken AI-mensen zo vriendelijk en voorkomend waren tegenover anderen of zelfs maar elkaar.]
"“AI is a symptom,” he said to me “and I’ve generalized it to all the human behavioral sciences. The idea that science and technology can be generalized to everything is something to really worry about and be concerned with — that’s my rational reason for what I do.” Which certainly puts Dreyfus at the antipodes of Simon’s position." [mijn nadruk] (237)
[Ik denk dat ik dat goed snap. McCorduck begrijpt er niets van en eindigt dit hoofdstuk met één groot argumentum ad hominem:]
"Dreyfus’s assertion that somehow the human body is key to intelligence, and that without it intelligence cannot exist, sounds strangely to me like the claims of nineteenth-century physicians, based roughly on the same kind of evidence and certainly with the same happy complacency, that women couldn’t think because they had female bodies, and that the male body was essential to real cognition." [mijn nadruk] (238)
[Ja, hoor. Werkelijk. Ze heeft blijkbaar geen idee van wat Dreyfus in lijn met bv. Merleau-Ponty bedoelt met dat waarneming / kennisverwerving / etc. altijd ook verbonden zijn met iemands lichamelijke bestaan en toestand. We kijken niet alleen met onze ogen, maar met heel ons lichaam - waardoor we bijvoorbeeld dingen over het hoofd zien als we moe en slaperig of somber of opgejaagd zijn. Wat heeft dat in hemelsnaam te maken met een stel seksistische medici die zeggen dat vrouwen niet het lichaam hebben om te kunnen denken? Het is te dom voor woorden. Ze is een beetje té enthousiast voor de heren in de AI die ze bespreekt.]
Gaat dus over robots en de behoefte om die te bouwen.
"In the context of the Mars robot, then, robots are one more tool, one more extension, of the human body and mind."(243)
[Precies, en dat zijn ze dus altijd, hoe 'intelligent' de taken ook zijn die ze uitvoeren. McCorduck begint steeds journalistieker te schrijven, lijkt het, met veel van dat over mensen generaliserende "we", en veel van de gegeven beweringen zijn nogal vaag of aanvechtbaar. We bouwen robots om onzelf als soort te leren kennen? Yeah right. We bouwen robots gewoon omdat we het leuk vinden om apparaten te bouwen die we van alles kunnen laten doen. Waarom zou dat onze visie op onszelf verhelderen of veranderen?]
Over de General Problem Solver van Newell, Shaw, en Simon.
"GPS did indeed codify a number of problem-solving techniques that humans have used without necessarily putting name to them. Among these techniques are what is called means-ends analysis, planning, and selective trial-and-error."(247)
[Dat is geweldig, niets mis mee, maar waar komt de behoefte vandaan om een GPS vervolgens 'intelligent' te noemen en om van 'kunstmatige intelligenties' te praten? Dat is helemaal niet nodig, voegt helemaal niets toe, leidt alleen maar tot verkooppraatjes.]
"A word about problem solving. It is not intended as a synonym for all thinking. In their book Human Problem Solving (1972), Newell and Simon call problem solving a subspecies of thinking, concerned explicitly with the performance of tasks. Under these circumstances, learning is viewed as a second-order effect, behavior that improves the performance of a system already performing in a given situation." [mijn nadruk] (248)
"Nevertheless, they could assert that GPS was a computer program capable of simulating, in first approximation, human behavior in a narrow but significant problem domain. It provided unequivocal demonstration, they went on to say, that a mechanism can solve problems by functional reasoning." [mijn nadruk] (249)
[Maar de simulatie zelf heeft geen 'gedrag' en 'redeneert' niet - dat zijn termen voor mensen - , de simulatie voert gewoon formele logische procedures uit die mensen er in gestopt hebben.]
Over John McCarthy's Advice Taker.
"McCarthy had been until then a pure mathematician, but a summer at IBM in 1955 gave him a better acquaintance with computers, and he marks that time as the point at which he took leave from mathematics and entered computer science and artificial intelligence, the term he coined." [mijn nadruk] (251)
[Ja en dat hij beter niet kunnen doen. Maar wat weet een wiskundige over de echte wereld?]
"LISP [door McCarthy ontwikkeld - GdG], with its offspring, is still the language of choice in most AI research."(252)
[Vandaag de dag niet meer, lijkt me.]
Ook het idee 'time-sharing' komt van McCarthy.
"I’d heard the term Advice Taker so often before I spoke to McCarthy that I told him I was surprised to learn that it was still a proposal. “No,” he said, “it doesn’t exist. Because in order to do it, you have to be able to express formally that information that is normally expressed in ordinary language. As far as I’m concerned, this is the key unsolved problem in AI. I uncovered the problem in 1958 and it’s still unsolved.” McCarthy has himself made several attempts to invent a formal language that would be able to express the events of everyday life: it’s the one scientific problem he’s stuck to, among the variety of others he’s taken up. But in his view, the general problem has simply not been attacked by enough good people to solve it, and he believes that until they do artificial intelligence will remain somewhat stuck."(254)
[Alle informatie die alledaagse taal normaalgesproken uitdrukt formaliseren / uitdrukken in een formele taal is principieel onmogelijk en niet iets van 'dat lossen we nog wel ooit op als ...'. Zoals de opkomst en de val van de Resolution Method vervolgens laat zien in die tijd.]
"Yet McCarthy still longs for a formal language that will express the facts of common knowledge, a rather lonely position in AI just now."(256)
Over naar Stanford, naar robotica.
"It embraces several aspects of intelligence, such as pattern recognition, problem solving, information representation, and natural-language processing, all of which have continuing research interest.(...) That a robot has to cope with the real world puts it in quite a different class from intelligent programs that operate in formal domains, such as mathematics or game playing" [mijn nadruk] (260)
"Thus three large robotics projects got underway in the mid- 1960s in the United States — at Stanford University, at Stanford Research Institute (which is nearby, but no longer officially connected with Stanford University, and is known officially now as SRI International), and at MIT. A fourth large project at Edinburgh University, sponsored mainly by the Science Research Council of Great Britain, soon joined in, and considerable exchange of ideas took place among the four centers. Though each project had its own flavor, the general aim was the same — to produce some sort of independent agent that would function in the real world, or at least a somewhat impoverished real world." [mijn nadruk] (261)
['Independent agent' is totaal vaag. En inderdaad: niks 'echte wereld', niet eens een 'somewhat impoverished' echte wereld. Dat is feitelijk een andere vorm van reductie zoals ook het alles willen formaliseren een reductie is. ]
"They began by pointing out that a computer program capable of acting intelligently in the real world must have some knowledge of that world, and to design such a program requires commitments about what knowledge is and how it’s obtained, central issues in philosophy since Greek times. Other points of philosophical debate must also be formalized: the nature of causality and ability, and the nature of intelligence."(261)
[Het gaat dus nog steeds om het formaliseren en dus reduceren van de werkelijkheid. ]
"But the solutions that were found began to suggest that many problems which seemed at first impossibly nonmechanical — exactly those problems that the most vehement critics of AI have declared will never be solved, such as “understanding” and “meaning” — slowly began to be brought into the domain of ordinary computational processes."(265)
"Joel Moses of MIT says,
The word you look for and you hardly ever see in the early AI literature is the word knowledge. They didn’t believe you have to know anything, you could always rework it all. And it’s a tremendously arrogant person who could believe that you could rework it all on the fly — start with this simple machine and just feed it a few things and all of a sudden you get Einstein’s theory of relativity. And nearly everybody bought that view. It took a long time for it to wither away."(266)
[Een realistisch standpunt, lijkt me.]
"In any event, robotics seemed to say very strongly that knowledge — lots of it in depth — was at least as essential as general principles of intelligence. And these results came to stand also as the most convincing denial yet of a difference between mind and body ..."(267)
"If there was a robot you could feel affection for, it had to be Stanford Research Institute’s Shakey."(268)
"The vision capabilities were essential. Nils Nilsson, who was project leader for Shakey sees scene analysis, as it’s called, as the one element common to the four robot projects. But each group took a different approach."(269)
"So I mean to treat vision research along with natural language in this chapter, but it will be somewhat subordinate, and I believe correctly."(279)
"If we could figure out how to make a computer “understand” our language, we’d finally know just what understanding and even language is all about."(280)
[Slechte redenering. Let weer op de richting. Het is natuurlijk asndersom: als wijzelf niets snappen van taal en begrip kunnen we dat een computer nooit aanleren.]
Over allerlei pogingen zoals The Conversation Machine, Baseball, Sad Sam, SIR, STUDENT, Project MAC, etc.
"But even in this very simple universe, some formidable problems arose having to do not only with denotation of words and concepts, but with connotation and implication as well ..."(384)
"But the researchers found that human intervention was essential to resolve ambiguities, both syntactic and semantic, once again, a real human who operated in the real world."(385)
[Mij lijkt dat je dat soort dingen tevoren kunt bedenken. En de gekozen onderwerpen zijn vaak van logische aard, waardoor het dan nog enigszins werkt en iets oplevert. Vincent Giuliano's commentaar is dan ook wat ik er zelf over denk:]
"Nonsense, scoffed Vincent Giuliano, in a comment appended to Simmons’s paper. Giuliano, then on the research staff at Bell Labs, wrote, “The paper might lead a casual reader to believe that considerable progress is being made — I tend instead to see evidence mainly of motion, with little real evidence of progress.” He pointed out what he called some brutal facts. The data the programs operated with were highly restricted, almost trivial. Beyond those areas, only the foggiest sort of understanding of semantics existed, and research in the area was likely to take many years to make real progress. There was still the problem of ambiguity, now resolved by human intervention, and nobody understood the relation between meaning and logical formalism. And, anyway, how do we decide what is relevant, which logical formal- ism cannot tell us anything about? Giuliano was skeptical as to whether any general principles had emerged from this work. The progression Simmons had cited of syntactic, semantic, and then logical analysis reminded Giuliano unpleasantly of the notion widespread among workers in machine translation that such translation should proceed through the stages of lexical, syntactic, and semantic processing, which hadn’t been achieved and didn’t seem likely to be.
In summary [wrote Giuliano], my reaction is that in a rush to demonstrate that question-answering can be done by computer, sight has too often been lost of the fact that much is yet to be learned about language, and that a demonstration can only be as good as the knowledge of language that goes into it. The existence of procedures of alchemy does not create a science — theories are needed which lead to testable hypotheses, and artifacts of computer usage are likely to be of utility only insofar as they are based on such theories or hypotheses. (Simmons, 1965)" [mijn nadruk] (289-290)
Drie jaar later zijn er betere programmeertalen en meer mogelijkheden. Weizenbaum's ELIZA wordt besproken.
"Taking advantage of these new languages and the interactive capabilities of a time-shared computer system, a restless young engineer named Joseph Weizenbaum produced a system he called ELIZA (for, like the famous Miss Doolittle, it could be taught to speak increasingly well)."(291)
"ELIZA was intended to simulate — or caricature, as Weizenbaum himself suggests — the conversation between a Rogerian psychoanalyst and a patient, with the machine in the role of analyst."(293)
[Er bestaan geen Rogeriaanse psychoanalytici, alleen Rogeriaanse psychotherapeuten.]
"Once the program was up and running, Weizenbaum was worried about how to present it to the world. Though it did indeed simulate the conversation between a psychotherapist and a patient, Weizenbaum was convinced that it might be misunderstood as giving some insight into therapy, into madness. As it turned out, he was right to be alarmed."(294)
"What additionally irked Weizenbaum, and helped accelerate the split between him and Colby, was the feeling that Colby had seized ELIZA and made it his own, under the name of DOCTOR, without giving due credit to Weizenbaum."(295-296)
"But the main disagreement, Weizenbaum strongly insists, is his fundamental belief that the program is of no therapeutic significance, whereas Colby maintained very strongly that it could be. This is not just a matter of two scientists disagreeing on a scientific issue. It speaks to a fundamental view of what machines are, what humans are, what psychotherapy is, and what intelligent machines might be and do. Weizenbaum’s view of these matters is the theme of his Computer Power and Human Reason, which would appear ten years after ELIZA." [mijn nadruk] (296)
"Thus, context, a shared world view based on mutually agreed upon facts, and a means of organizing this material for easy modification and access — all of these elements would come to replace the principles that had guided artificial intelligence research up until now. Instead of searching for a few general and uniform principles of intelligent behavior, AI researchers were beginning to suspect — reluctantly, for it violated the scientific canon of parsimony — that intelligence might very well be based on the ability to use large amounts of diverse knowledge in different ways." [mijn nadruk] (298-299)
Over Terry Winograd en het programma SHRDLU.
"Clearly, formal techniques of logic and mathematics are not easily applicable to such holistic models, as the critics of AI have been pointing out for years. But the fault was probably with logic and mathematics, Winograd declared. AI approaches to modeling cognitive processes can provide formalism without the limitations of mathematical or logical formalism. This involves a computer notion of procedure instead of proof, and the viability of this approach stands or falls on how well it provides a model of what we mean by understanding." [mijn nadruk]
Het programma HACKER - een verdere uitwerking van SHRDLU door Gerald Sussmann - kon zichzelf bijsturen.
"But the new version of SHRDLU, called HACKER, had the ability to debug — that is, to examine the procedure it had just undertaken and to identify the small flaw that had thrown it off the track. Thus it learned from its mistakes. It came to know that subgoals on the way toward a goal may sometimes conflict, and it attempted to reorder those subgoals so that the most pressing goal would be achieved first. HACKER exhibited expertise about debugging and repair, coupled with the ability to examine its own problem-solving goals and actions so that it was able to supply this debugging expertise to its own reasoning. In short, it had self-consciousness." [mijn nadruk] (304)
[Waarom moet dat laatste zinnetje nu toegevoegd worden? Er is geen enkele noodzaak om dat te doen. Het is een programma dat procedureel kijkt naar bepaalde elementen van zijn procedure zodat het de volgorde van taken kan veranderen. Dat is alles.]
"As the 1970s drew to a close, knowledge representation was perhaps the most hotly debated topic in artificial intelligence."(314)
"The two examples of applied AI to be described are useful instances of the way this field seems most likely to affect us in the immediate future. First is an intelligent assistant, working in a narrow but difficult task domain and helping a human expert to do some of the taxing but essential parts of a particular job. This new version of the sorcerer’s apprentice is called DENDRAL, and behaves as a chemist’s assistant in interpreting the data from mass spectrography, working at the intellectual level of a chemistry Ph.D. DENDRAL was the first such intelligent assistant to be designed and put into use, but many others have followed." [mijn nadruk] (318)
[Ik vraag me werkelijk af wat de toegevoegde waarde is van de uitdrukking "intelligent assistant". "Assistent" was goed genoeg geweest.]
"The second project I want to describe is a somewhat different application of artificial intelligence. In the effort to make machines think, some insights have been gained into how humans think, and therefore learn. Though these insights are very far from complete, they suggest some ways the process of educating schoolchildren might be altered to make it more effective and pleasant for everyone concerned — child, parent, and teacher." [mijn nadruk] (319)
Maar eerst DENDRAL. Edward Feigenbaum en Julian Feldman werkten samen en publiceerden Computers and Thought in 1963. Daarna gaat Feigenbaum naar Standford en McCarthy.
" ... how closely the task mimicked human methods began to be less interesting than how well the task was performed, regardless of method. In the end, those two concepts were less distant than had originally been believed, but I will come to that in a moment." [mijn nadruk] (327)
[Dat lijkt mij toch ook.]
"The project team was to discover some interesting things about expert chemist behavior, which they could soon generalize. First, it became clear that the acquisition of specialized knowledge was a bottleneck in the design and building of intelligent agents. The knowledge inside an expert’s head is largely heuristic knowledge, experiential and uncertain — mostly good guesses in lieu of facts and rigor. Much of it is private to the expert, not because he’s unwilling to share publicly how he performs, but because he’s unable. He knows more than he’s aware of knowing, something the early language-machine designers recognized." [mijn nadruk] (330)
Vervolgens het tweede project: LOGO van Seymour Papert.
"Instead, the LOGO project has been conceived as a means for applying what artificial-intelligence researchers have discovered about human thinking and problem-solving processes in the course of trying to get computers to think and solve problems. It’s concerned with teaching children to think — or, as Seymour Papert, the principal investigator on the project once put it, LOGO aims to teach children to be mathematicians instead of teaching them about mathematics. And not only mathematicians — musicians, physicists, engineers, story tellers, even teachers themselves."(335)
"If Papert didn’t see eye to eye with Piaget on every aspect of the issue, he was nevertheless profoundly influenced. Along with Piaget, and for that matter with Dewey and Montessori too, Papert believed that children learn by doing, and by thinking about what they do."(337)
"The turtle encourages the beginner to anthropomorphize, though in this instance for good reasons, not bad; the LOGO language puts the child in charge. This personal power is what Joseph Weizenbaum (see Chapter 13) sees as an evil thing in his description of hackers. But Papert and his group view it as a good thing, an intellectual power extended to those who have traditionally been the most powerless, namely, children.(...) Manipulating the turtle is an exercise in setting freedom free." [mijn nadruk] (347)
[Behoorlijk vage beweringen. Wat is 'personal power'? Wat is 'setting freedom free'?]
"LOGO and DENDRAL are two quite different but working examples of the application of artificial intelligence and its principles to real-life situations. I could have described other applied programs that work, but none that has been so thoroughly accepted by persons outside the field. And that points to a problem that few have addressed, the problem of transfer. If a program exists, say, for diagnosing bacterial infections in humans and prescribing cures, one that on the whole works with better success than human physicians, why aren’t patients and physicians using it? Our easy guess might be that any intelligent program that replaces professionals at what they do and get well-paid for, or even a program that is an intelligent aid, is going to meet mighty resistance. But the facts are that no resistance has been recorded because no one has had the resources to attempt a large-scale transfer from the laboratory into the field." [mijn nadruk] (350)
"Changes which are sudden, profound, and seem inevitable are often viewed as deeply threatening to us, and this is the theme of the next chapter."(350)
[Gepsychologiseer: we voelen weerstand, we voelen ons bedreigd ... bla bla. Alsof er geen goede argumenten kunnen bestaan om die transfer niet te maken. ]
"To this point we have been concerned with whether a machine can think. Now we turn to another question, suggested by the ambivalence — one might even say terror — evoked in us by the first: Should a machine think? To answer that, we have to make some guesses about how such thinking machines might behave, and in particular, how they would affect our lives. If such machines are to be our slaves, we can anticipate some splended benefits, but what about intelligent machines to whom we would be slaves? Is that a real possibility? Would they treat us benevolently? On the other hand, is it sen- sible to ask whether, if they are our slaves, we owe them benevolence? Most important, if at last we come to have machines who think, will we have to adjust our own view of ourselves radically, just as we did when Copernicus told us we weren’t at the center of the universe, and Freud told us we weren’t the altogether rational creatures we’d assumed?"(353)
[Een nietszeggende vraag hier die al uit gaat van het idee dat machines kunnen denken. Het is slecht taalgebruik. Machines kunnen niet denken, dus de vraag of ze wel zouden moeten denken is zinloos. En dat gebruik vabn "we" en "our" is weer eens ergerlijk.]
"In the spring of 1976, a book appeared called Computer Power and Human Reason, by Joseph Weizenbaum. It was a detailed attack on artificial intelligence, with some aspects echoing earlier attacks, and other aspects quite distinctly original. Among its distinctive features was the fact that it came from one of the major centers of AI, Massachusetts Institute of Technology, and that it was written by a man who had made an early major contribution to the field, the conversational program called ELIZA, described in Chapter 11. In his book, Weizenbaum is the first to raise the question of morality explicitly."(356)
Dat boek wordt uitvoerig besproken. Het wordt samengevat, het debat erover wordt ook weergegeven.
[Ja, maar McCorduck heeft zoals wel vaker de neiging op de man te spelen in plaats van met goede argumenten te komen. Ze lijkt Weizenbaums kritiek helemaal niet te begrijpen. Dat geldt overigens ook voor veel van de genoemde AI-mensen. Moralisme is blijkbaar een groot probleem voor mensen die hun eigen waarden en normen niet bewust hebben.]
"Can a machine think? The answer, as we’ve seen, depends very much on what we’re willing to admit as machine, and what we’re willing to admit as thinking. The definitions of either aren’t quite so simple as they looked when we embarked."(381)
[Precies en dat zou tot voorzichtigheid moeten manen, vooral ook tot een voorzichtig taalgebruik. Ik heb daar toe nu toe weinig van gemerkt bij de auteur.]
"Artificial intelligence sides with that waggish old Frenchman Julien Offray de la Mettrie, author of — and martyr to — L’Homme Machine, who was at pains to point out that one useful way of regarding human beings is as mechanisms — that is, all their functions are capable of being described in a logical, analytical way." [mijn nadruk] (482)
[McCorduck lijkt maar niet te begrijpen dat dat gewoonweg onmogelijk is en een reductionistisch perspectief vormt dat altijd tot problemen moet leiden. Ook al zie je de mens als alleen maar als een 'fysieke entiteit' - wat ttrouwens nog altijd wat anders is dan een mechanisme -, dan nog is een mens een fysieke entiteit met zelfbewustzijn, emoties, en zo verder. Dat betekent niet dat als een andere fysieke entiteit - een computer - 'hetzelfde gedrag' zou vertonen als een mens op een bepaald punt, dat dat 1/ ook maar iets zegt over de mens als fysieke entiteit en 2/ dat dat gedrag overeenkomst met hoe mensen zich gedragen. Het zijn denkfouten. Ik weet ook niet of ik AI een wetenschap wil noemen, de aanpak lijkt me meer die van ingenieurs die dingen uitproberen.]
Eerst volgt een schets van de grote lijnen in de ontwikkeling met daarna een uitwerking vanb concrete initiatieven, resultaten, en zo meer..
[Ze blijft te pas en te onpas het woord 'intelligent' gebruiken. Niet geweldig. Ik wil het verhaal ook niet samenvatten: er worden te veel details opgesomd die niet zo relevant zijn voor de normatieve kwestie waar het boek eigenlijk overgaat.]
"Throughout these two and a half decades, artificial intelligence research became genuinely international, with first-rate groups in North and South America, Asia, Europe, the Middle East, and Australia and New Zealand."(418)
[Maar McCorduck bespreekt desondanks alleen de Amerikaanse AI. Dat kan een keuze zijn, maar daar zegt ze niets over. ]
"A sign of the disordered state of AI research at the beginning of the 1980s was a survey taken by AI researchers Ron Brachman and Brian Smith, then at Bolt Beranek and Newman, who asked the scientists actually working in the field what they thought AI was, or should be. Not surprisingly, the Brachman and Smith survey yielded no substantial consensus from the scientists who responded." [mijn nadruk] (420)
"Surely one of the most striking aspects of AI during the late 1970s and 1980s was its move from obscure, almost fringe, science to hot public issue, and its subsequent popular (and perhaps scientific) swoon. I use the term hot public issue in several senses. First, the topic itself raises controversy, and more must be said about that further on. But second, even as the field was in scientific flux in the 1980s, its most public and aggressive face was on Wall Street. Every practitioner, it seemed, whether professor or graduate student, engineer or programmer, was being ardently wooed by venture capitalists who, in turn, helped their new celebrity clients to establish startups and find investors for these new enterprises — hot public issues — that promised a very great deal." [mijn nadruk] (434)
"It seemed too good to be true. It was. A young and fragile science was being asked to perform miracles that were then beyond its powers. And so the industrialization of AI in the early 1980s came to prefigure, more modestly, the industrialization of the Internet a decade later (and railways and automobiles a century and more earlier) in feverish boom, followed inevitably by deflated bust, a typical pattern as new technologies are commercialized. Which is not to say it was all smoke and mirrors. The science was young and fragile, not fraudulent." [mijn nadruk] (435)
[Heel die overschatting van AI is er vanaf het begin. "Not fraudulent"? Hm, ik denk dat grootspraak ook een vorm van bedrog is.]
"As the 1980s drew to an end, yet another AI paradox arose. Business people, politicians, and the military were clamoring for AI; the media had made it into a nine-day wonder, but the field’s new big ideas seemed fewer and further between."(442)
"By the start of the 1980s, other philosophers [naast Dreyfus en Weizenbaum - GdG] came to address the question of whether or not AI is possible. John Searle, also of Berkeley’s philosophy department, had no objection to “weak AI,” (...) Searle provoked many responses, including one from another philosopher, Daniel Dennett, of Tufts University. He and his com- puter scientist coauthor, Douglas Hofstadter, of the University of Indiana, replied to Searle"(443-444)
"In 1985, Marvin Minsky published his The Society of Mind, a summary of what he’d learned, been thinking of, and speculating about for many years ..."(445)
"By 1989, another scoffer had emerged, this time a distinguished physicist, Roger Penrose, who argued in his best seller, The Emperor’s New Mind, that real intelligence grows out of some undiscovered behavior at the quantum level of the human brain, and AI will never be able to achieve that with computers constructed using current silicon technology."(448)
"The increasing power of computation in the 1990s, plus an accumulating body of knowledge, permitted some robotics researchers to turn their attention to socially interactive robots, robots that helped and were at home with people and each other."(463)
[En opnieuw krijgen we over Kismet te horen, der meest overschatte robot ter wereld.]
"The elders who've alpha- and beta-tested the Nursebot like her just fine. Why not? Enormous care is going into her design and development. Is eldercare a job that ought to be done by humans? Maybe, if they’re well trained and experienced, but the realities of world demographics say it isn’t going to happen. In this case, better to have skilled and reliable robotic care than nothing, or dismally unskilled human care." [mijn nadruk] (468)
[Weer eens een erg normatief stukje. Ja, hoor, die oudjes vinden dat geweldig leuk. Wie zou dat onderzocht hebben? En zouden ze ook het alternatief - ervaringen met mensen - onderzocht hebben en dat vergeleken hebben met de ervaringen met robots? Maar nog belangrijker: dit is een verkeerde manier van denken. Het zet de zaken op zijn kop: het accepteert wat niet geaccepteerd moet worden en is enthousiast voor wat er niet toe doet.]
"In addition to using formal logical systems to achieve AI goals, much of the progress of AI in the 1980s and 1990s was based on mathematical techniques drawn from fields such as probability, statistics, and decision theory. Using these techniques, AI successfully found its way into all kinds of nooks and crannies, from airline reservation systems, where pleasant humanoid voices offer information about arriving and departing flights, and even book your seat, to video games, where virtual opponents (or teammates) anticipate, understand, or cooperate with the human player, to remarkable summaries of news bulletins from disparate sources, to machine translations done on the fly by automatic translators at- tached to search engines like Google — a search engine which itself uses many AI techniques, and is begging AI researchers for more. These approaches achieved results (some worked very well indeed) and their designers felt no particular obligation to be faithful to human models of intelligence." [mijn nadruk] (487)
[Dat eerste is typisch voor de reductionistische aanpak in AI, voor de beperkingen ervan. Dat laatste is juist de goede benadering. Het hoeft allemaal niet op mensen te lijken, waarom zou het? Maar hou dan ook op met in mensentermen over machines te praten. Waarom zou je die "intelligent" noemen? Dat is een zinloze en verwarrende aanpak.]
"For no sooner did we breathe easy about the potential cultural meltdown caused by Y2K than Bill Joy, the cofounder and then chief scientist of Sun Microsystems, one of the largest and most successful computer firms, raised his own alarms in the April 2000 issue of Wired (Joy, 2000). A prepublication version of Ray Kurzweil’s The Age of Spiritual Machines had crystallized a growing unease he’d felt for a while (Kurzweil, 1999)."(494)
Bespreking van het artikel en het debat dat volgde op de bladzijden hierna.
[Dat is een interessant debat. Ik moet eree eens een keer induiken.]
"AI has simply become background technology in a complicated and complex world. Good. Let artificial intelligence, cognitive intelligence, computational intelligence, whatever its label, slip into our lives ever further. In tasks that are beyond human wit, too complex, too fast, too detailed, AI steps in and allows us to do them anyway. We are building in our own image, but better, we hope. If not, we’ll find out soon enough."(515)
[Wat een naïef standpunt.]
"AI has taught us more about human intelligence, that it is not only the function of brain activity (search, prune, recombine, all the techniques mentioned here), but that general intelligence exists by being embedded in a milieu, an environment, cultural, social, and physical, that contains not only other human beings (and now partially intelligent machines), but also the accumulated human knowledge and artifacts of the past, as well as nature itself." [mijn nadruk] (518)
[Daar hadden we AI helemaal niet voor nodig, dat was honderd jaar geleden al duidelijk.]
"AI is a godlike enterprise, I said in the first edition, and I stand by that. To repeat one of Stewart Brand’s aphorisms, we are as the gods, and we may as well get good at it. My own faith, that our smart machines are going to help us do that, can rest on much more evidence than when I first declared it — more, but not yet altogether conclusive." [mijn nadruk] (520)
[Tja, als je zoiets kunt zeggen is er geen hoop meer voor je.]
Verder aangevulde tijdlijn vergeleken met de tijdlijn in de eerste editie.