[FRIAM] AI Musings
Prof David West
profwest at fastmail.fm
Thu Mar 30 10:50:32 EDT 2023
The "AI Pause" made national TV news yesterday (long after those on this list noted and reacted to it) and that made me revisit a theme I have thought about since Newell, Simon, and Shaw created Logic Theorist.
Advocates take a caricature (perhaps too strong a word) of human intelligence, write a program to emulate it and declare the program "intelligent."
The original conceit: true intelligence was the kind of thinking exhibited by college professors and scientists. Almost trivial to emulate (Newell and Simon programmed Logic Theorist on 3x5 cards before Shaw was able to implement on a computer).
Maybe reading—correctly converting text to sound, like a child—was more indicative of human intelligence, and Sejnowski created NetTalk. that, somewhat eerily, produced discoveries of sounds, and errors, and achieved near perfect ability to "read." Listen to the tapes sometime and contrast them with tapes of a human child learning to read. Of course, comprehension of what was read did not make the cut.
State of the art improved dramatically and the caricatures of human intelligence are more sophisticated and the achievements of the programs more interesting.
But, it seems to me there is still a critical gap. We can program an AI (or let one learn) to fly a commercial jet as well or better than a human pilot—BUT, could even the best of of breed of such an AI pull a Shullenberger and land on the Hudson River?
Another factor behind the "hysteria" (sorry for the sexism) over AIs causing massive unemployment is a corollary to the caricaturization of human intelligence. Since the Industrial Revolution, and certainly since the age of Taylorism and the rise of automation; work itself has been dehumanizing.
If you define human work in terms of what can be done by a computer then it is tautological to claim an AI is intelligent because it can perform human work.
I was contemplating ChatAIs and quickly realized that my profession—college professor—was one at immense risk of replacement. I would bet good money that a ChatAI could produce, and maybe deliver, lectures far better than any I created in 30 years teaching. And probably most, if not all, of the presentations I made at professional conferences over the years.
I am still vain enough to think that some of the papers and books I have written are beyond an AI, and certain that no AI could do as well in spontaneious Q&A after a presentation than I.
Bottom line, I still believe that AI can and does equate to HI, only when some aspect of HI is ommitted from the equation. This is not essentialism, but analogous to the digitization of a sine wave, no matter the finite sampling rate, there is always some missing information.
davew
More information about the Friam
mailing list