Has progress in AI been disappointing?
I like to consider AI to be the field that uses computational modeling methods to investigate one of the great intellectual problems of all time: What is Mind? And how can a physical object have a mind?
I like to compare AI with Physics, which is the field that investigates another of those great intellectual problems: What is the World made of? And how did it come to be?
We use Summer 1956 to mark the beginning of the modern age in Artificial Intelligence, commemorating the Dartmouth Workshop, officially the Dartmouth Summer Research Project on Artificial Intelligence.
Naturally, there was important previous work, including the 1943 McCulloch-Pitts neuron model, Alan Turing’s seminal 1950 paper, and Marvin Minsky’s SNARC neural network machine in 1951.
According to the Wikipedia History of Artificial Intelligence article,
In 1958, Herbert Simon and Allen Newell wrote, “within ten years a digital computer will be the world’s chess champion”, and “within ten years a digital computer will discover and prove an important new mathematical theorem.”
In 1965, Herbert Simon predicted that “machines will be capable, within twenty years, of doing any work a man can do.”
In 1967, Marvin Minsky predicted that “Within a generation . . . the problem of creating `artificial intelligence’ will substantially be solved.”
In 1970, Marvin Minsky, in an interview with Life Magazine, stated “In from three to eight years we will have a machine with the general intelligence of an average human being.”
These predictions seem dramatically over-optimistic. However, . . .
It is quite understandable that young scientists, having discovered new and fruitful approaches to a long-standing, extremely hard and important problem, may be quite excited and optimistic about the promise of future progress. (Simon, Newell, and Minsky were in their 30s and early 40s when they made those predictions.)
Consider the first prediction, from 1958, that within ten years, a digital computer will be the world’s chess champion. In 1968, computer chess was nowhere near the world-champion level, but by 1966 Richard Greenblatt's Mac Hack had played in numerous human amateur chess tournaments, and had achieved a US Chess Federation rating of 1243.
In fact, it was not until 1997, four decades after the original prediction, that IBM’s Deep Blue supercomputer defeated Gary Kasparov, the world chess champion, in tournament play. By 2017, after another two decades, AlphaZero from DeepMind was able to learn, though self-play, to defeat all other human or computer players at chess, shogi, and Go. Although there are vigorous chess competitions among computers, and among hybrid human-computer collaborations, computers are not eligible to play for the human world chess championship (any more than forklift trucks are allowed to compete at weight-lifting in the Olympics).
One must ask: when predicting the time-line for achieving a future scientific and technological goal, that many experts predicted could never be achieved at all, how large an error is a factor of four?
If I had to pick a beginning for the modern age of physics, I would nominate the 1660s and 1670s when the differential and integral calculus, that is, differential equations, were invented by Newton and Leibniz.
Differential equations are the appropriate mathematical language for describing the continuous dynamical behavior of the physical world. By providing a powerful and expressive language for scientific hypotheses, and a set of solution methods capable of making strong predictions from those hypotheses, differential equations enabled new kinds of scientific progress.
Physics has made tremendous progress in the 350 years since the creation of differential equations, but few people would say that the problem of physics is “solved”. (Although in 1878, Max Planck's thesis advisor told him not to go into physics, because "almost everything is already discovered".)
It is, of course, impossible to state reliably whether the “Problem of Physics” or the “Problem of the Mind” is more difficult.
However, in my opinion, the Problem of the Mind is the more difficult one.
Physical laws are believed to be spatially and temporally invariant. They do present differently at different scales of certain parameter spaces like velocity, mass, etc, because terms that are negligible at one scale may become dominant at others. The invariance that unifies the classical domain with the quantum domain is still a matter of faith rather than science, but it is generally believed, even if we don’t yet understand how it might work.
The phenomenon of mind (as far as we know today), does not extend over the same numbers of orders of magnitude as the phenomena of physics, but the nature of intelligence, including perception, cognition, action, communication, culture, etc, seem to be significantly more variable.
Perhaps this apparent complexity is due to our relative ignorance about the mind, and in coming centuries, an underlying simplicity will emerge. But I doubt it. Physics describes several layers at the foundation of a hierarchy of scientific theories. On top of Physics are built (very roughly speaking) Chemistry, Biology, Psychology, and Sociology-Anthropology, with increasing variability and complexity as we proceed upward.
This suggests (with an extremely broad brush) that the Problem of the Mind is significantly more complex than the Problem of Physics, and arguably the Problem of Culture will be yet more complex.
Physics has made impressive progress in 350 years, but it is not yet “done”.
Only 60 years after starting to apply computational modeling methods to the Problem of the Mind, it is not surprising that we are far from “done”.
In my opinion, the last three prediction bullets above are over-optimistic by at least one order of magnitude. That is, Artificial General Intelligence (AGI) will take at least one, and perhaps several, centuries to achieve. And that assumes that we can keep our scientific, technological, social, political, and ecological infrastructures on an even keel during that period.
These are some of the most fundamental problems that humanity is trying to answer. It is hardly surprising that they are problems for the centuries and millennia, not just for years and decades.