Why were so many artificial intelligence founders so optimistic?

Question

• 1954: The Georgetown experiment in 1954 involved fully automatic translation of more than sixty Russian sentences into English. The authors claimed that within three or five years, machine translation would be a solved problem.
• 1956 - Herbert Simon: "machines will be capable, within twenty years, of doing any work a man can do"
• 1967 - Marvin Minsky: “within a generation ... the problem of creating ‘artificial intelligence’ will substantially be solved.”
• ...

Why were so many AI founders so optimistic in the early stages of AI?

Answer 1

I suggest that the tacit assumption in the question---that all founders of A.I. were optimistic about the goal being reached very soon---is historically false.

Some were optimistic, but just as many were not, although they hoped that their work was a step to the eventual realization of this goal far in the future.

I'm not sure it has anything to do with the generation they were part of being particularly optimistic. Many scientists in the same generation, including many noted physiologists, were sceptical about such claims, yet worked to make contributions toward this goal.

Many researchers in the same generation denied that artificial intelligence in the sense of pure Turing Machines COULD EVER do ALL the work biological brains were doing. M. Lighthill, R. Gerard, N. Rashevsky (originator of inhibition in neural network models), R. Rosen, Popper, Penrose, Hayek, Lucas, Pribram, etc., etc.

E.g., in a paper (Church’s thesis and its relation to the concept of realizability in biology and physics) in 1962, Rosen, argued that most arbitrary neural networks are not computable. Which does not mean artificial intelligence is impossible, but that it requires other methods of physically realizing computation.

Most suggested schemes are neural networks conceptually, but require specialized physical systems to be realized, due to the number of simultaneous interconnections. Indeed, one can couple Turing machines in a particular way to make a "as-a-whole" noncomputable (hypercomputable) system. Such discussions were present at the same time as much of A.I. was being developed.

(This has been discussed by J. Copeland in the 1990's---and the one catch is that Turing computation is the only computation that has a universal computer. Most hypercomputation concepts lack universal realizations, but depend on their construction on what they can and cannot do, are more specialized.)

McCulluch and Pitts were students of Rashevsky for instance, and they published on related subjects all the way through the 1950's.

Answer 2

According to Generations, by William Strauss and Neil Howe (S&H), the so-called World War II, or "Greatest" Generation, born 1901-1924 in America, were a particularly optimistic generation, in large part because of their life cycle. They won World War II as young men, landed a man on the moon in their middle age, and enjoyed the Reagan (economic) boom as oldsters. Simon (born 1914), Minsky (born 1919) and Reagan himself (born 1911) were all members of this generation. "Artificial intelligence" emerged during the period of these men's middle age, and that's why their optimism is sometimes associated with their science.

They had been shaped this way as children by parents of the two preceding generations: FDR's "Rendezvous With Destiny" Generation (or what S&H called "Missionaries,") and the Lost Generation of FitzGerald and Hemingway. Mr. Howe (and Mr. Strauss before he passed away) believe that today's young people, the so-called Millennials, born 1982-2001, are another optimistic generation, raised by the Boomers (the modern "Rendezvous") and Generation X (the "new Lost"). Hopefully without a "World War" attached to their designation. This would be particularly true of Americans, somewhat true of others in a "global" economy.

These prognosticators made one fundamental mistake; that of assuming that succeeding generations will be like them and continue their trajectory. S&H postulated four different generations with different priorities, with the fourth remotest (the Millennials), taking up the "baton" of the World War II generation. So the vision of these World War II generation men will be realized in the middle age of the Millennials, in the 2020s-2030s, not during the middle age of the so-called Silent generation (of Alan Greenspan and Neil Armstrong), (1965-1985).

Answer 3

There is a possible answer to this which is pretty generic. I came across this idea in an essay by Robert Heinlein -- can't find the title or other info right now, unfortunately. The idea is that technology tends to follow an exponential curve, but when people imagine the future, they tend to assume linear progress.

The result is something like the graph drawn above. In the short term, the line predicts progress that's too rapid, but in the long term it's the other way around.

As an example, if you look at science fiction books from the 20th century, they project that human space travel will be much more common and economically viable than it really turned out to be. For example, in the 1960s they projected lunar colonies in the 1970s.

AI could just be another example of the same phenomenon.

Answer 4

In short, they under-estimated the prime importance of context.

AI can be superior to human effort in some small, well-defined, self-contained worlds, e.g. small vehicles and machines running around a warehouse. However, in the real world, they are lost, and can be completely out-performed by small children.

I can't imagine any AI in the near future being able to understand teenagers' language, slang and argot. Not only does it change quickly, but it includes irony, sarcasm, implied negatives, and much, much more, things that a computer program, no matter how sophisticated, won't be able to keep up with, for the main reason that teenagers don't necessarily publicize the manner in which their slang is to be understood. You have to be part of the "in-crowd". They do, in fact, go out of their way to ensure that others don't understand the subtleties.

There's much more, of course, but human language is one aspect where computer programs are completely inferior, and will remain so, possibly forever.

Answer 5

Had they not been so optimistic, they would have pursued other endeavors instead, and not became the founders of AI.

There's a selection bias that things may go the other way around from what seems to be behind the question.

Also, they were solving the easier problems first, so progress seemed fast, and they just projected that speed of progress out even beyond where it was accurate to do so, because the harder parts take longer to solve. Thanks to much better tools and the work that's been done already, progress is being made faster than before, but the challenges being tackled are also quite hard.

Answer 6

One effect that is visible e.g. in programming, is that one of the holy grails was "automatic programming", i.e., having the computer write a program from some high(er) level description. Nobody calls that "artificial intelligence" anymore, it is programming language design, compiler building (In the sixties there was a project to build a compiler giving better code than experienced assembly language programmers... and they succeeded. Doing this is bread-and-butter for modern optimizing compilers.).

The curse of AI is that when they find out how to solve a problem, it suddenly isn't AI anymore. Heuristics, search in humongous spaces getting cut down to manageable sizes, all that just melts away.