He answered the AI debate in 1979. We are in the middle of it.
years between Sagan's defence of intelligent machines and today
The builders of frontier AI now say the takeoff has started. See The Future of Humanity, Part 1, terencekok.com.
years between Sagan's defence of intelligent machines and today
The builders of frontier AI now say the takeoff has started. See The Future of Humanity, Part 1, terencekok.com.
Sagan, Broca's Brain (1979), pp. 295 to 308. Photo: NASA/JPL, public domain.
The assumption that no being is as capable or reliable as a human, applied to machines before the evidence is in.
“A conclusion drawn before all the facts are in.”
Carl Sagan, Broca's Brain, 1979, p. 296
Sagan defined the prejudice as prejudgment. Applying it to trust as well as dismissal is our reading.
Turing's test: can a machine pass for a human in a typed exchange? Sagan said none was that good yet.
Winograd's program stacked children's blocks on typed instructions and explained why it moved each one.
Weizenbaum's program turned a patient's statements back as questions. Sagan called one reply astonishing.
The reply was preprogrammed, he wrote, but so are human therapists' replies.
Short exposures could induce "powerful delusional thinking in quite normal people."
Chat is an illustrative exchange in ELIZA's reflect-the-statement style, not Sagan's transcript. Weizenbaum quote via History.com.
Sagan writes that the ten best chess players have nothing to fear from any present computer, and that the situation is changing.
Carl Sagan dies on 20 December.
Deep Blue beats world champion Garry Kasparov, 3.5 to 2.5.
Sagan, Broca's Brain, p. 301. History.com on Deep Blue; Cornell Chronicle, 20 December 1996.
Sagan, Broca's Brain, p. 304. NASA, Perseverance Sol 200 AutoNav drive.
Viking, he wrote, was about as intelligent as a dragonfly.
Sagan, Broca's Brain, pp. 305 to 306. The rung layout is our summary.
| Sagan in 1979 | Since then | Our reading |
|---|---|---|
| Ten best players have nothing to fear | Deep Blue wins in 1997 | Right, early |
| Rovers will need to decide for themselves | Perseverance picks its own path | Right |
| Machines will be specialised | Frontier models are general-purpose | Missed |
| Machine therapy is possible; confidentiality to be worked out | Companion chatbots exist; privacy still open | Light on risk |
| Displaced workers re-employed within a generation | Contested | Untested |
Our own comparison. Sagan: Broca's Brain, pp. 301 to 307.
Weizenbaum drew the opposite lesson from the same program: people read understanding into a machine.
Re-employment, he wrote, should not be too difficult. He gave it a few sentences and little evidence.
A label like speciesism can end an argument as easily as win it. His own test says to ask for facts.
Every time a machine passes a test of intelligence, the test changes. Who gets to move it?
Terence Kok and Michael Low · terencekok.com/blog/series/sagan-and-the-machines