Moving the Goalposts. Sagan and the Machines, lecture 2. Terence Kok and Michael Low, with a faint portrait of Carl Sagan.
ML TK
The reflex

A machine does the thing. The thing stops counting.

Researchers call this the AI effect.

ML TK
Turing, 1950

“You will never be able to make one to do X.”

The objection he expected, from "Computing Machinery and Intelligence"

His reply: “No support is usually offered for these statements.”

Sagan and the Machines · Lecture 2
ML TK
Sagan, 1979

He caught the reflex in the middle of the game.

The put-down

Machines were demeaned, often with relief, because chess was an area where humans were still superior.

His reply

A machine at middle-range human level is “a very capable machine.”

And on Turing's test: he was “not sure how many humans would pass.”

Sagan, Broca's Brain (1979), ch. 20.

Sagan and the Machines · Lecture 2
ML TK
The 2025 result

A language model was judged human more often than the human.

GPT-4.5 with persona prompt: 73% GPT-4.5 without it: 36% Chance: 50%

The authors call it “a test of humanlikeness,” not of intelligence.

Jones and Bergen (2026), PNAS, five-minute three-party games.

Sagan and the Machines · Lecture 2
ML TK
Same shape, each time

A game is held up as a measure. A machine wins. The method is examined.

1997

Deep Blue beats Kasparov, 3.5 to 2.5. "It is only brute force."

2016

AlphaGo beats Lee Sedol, 4 to 1. "It is only a game with rules."

2025

A model passes a three-party Turing test. "It measures humanlikeness."

IBM; Google DeepMind; Jones and Bergen (2026). The quoted replies are our summary of the pattern.

Sagan and the Machines · Lecture 2
ML TK
The saying

“Intelligence is whatever machines haven't done yet.”

Larry Tesler

Dated it to about 1970. Hofstadter's shorter version is “AI is whatever hasn't been done yet.”

First print found

Hofstadter, 1979, the same year as Broca's Brain. The origin stays disputed.

Tesler's website; Quote Investigator.

Sagan and the Machines · Lecture 2
ML TK
Not every moved line is vanity

“Solely measuring skill at any given task falls short of measuring intelligence.”

François Chollet, On the Measure of Intelligence, 2019

Unlimited training data buys any level of skill. The buyer wants to know how it copes with something new.

Sagan and the Machines · Lecture 2
ML TK
Warning signs

A defensive test.

Sagan and the Machines · Lecture 2
ML TK
The five checks

A fair test.

Our own checklist, a heuristic from reading, not a validated instrument.

Sagan and the Machines · Lecture 2
ML TK
Michael Low's view

Decide what to protect, then test for it.

Defining human value as whatever machines cannot yet do guarantees a retreat. Name the capabilities worth developing, say why, and test for those.

Michael Low monogram

Michael Low, Founder, SENSE Collective

ML TK
Questions for leaders

Before you judge an AI system this quarter.

Sagan and the Machines · Lecture 2
ML TK
Next

Part 3: The Terminal That Listens

Sagan imagined therapy terminals in 1979. What changes when the terminal listens back?

1The Man Who Wasn't Afraid of Robots
6Who Meets Us First
2Moving the Goalposts
7The Candle in the Dark
3The Terminal That Listens
8A Society Nobody Understands
4Memory Outside the Body
9Who Holds the Button
5Robots on Other Worlds
10Who Speaks for Earth
Michael Low monogram

Terence Kok and Michael Low · terencekok.com/blog/series/sagan-and-the-machines