Loading slide

The Question

  1. 01The Question
  2. 02The machines arrive
  3. 03The uncomfortable question
  4. 04A peculiar habit of mind
  5. 05What Turing had already built
  6. 06Unsatisfied
  7. 07What does it mean to think?
  8. 08The definition problem
  9. 09The Imitation Game
  10. 10Why this sidesteps the problem
  11. 11Why this was radical
  12. 12Meet ELIZA
  13. 13What did you notice?
  14. 14The question is still open
  15. 15Reinforce your understanding
  16. 16Question: Turing's move
  17. 17Question: Behavior
  18. 18Question: Limits of the test
  19. 19Quiz: answer
  20. 20The question follows you
  21. 21Want to go deeper?
13 / 20
BackNext

What did you notice?

You were watching for something: a slip, a hesitation, a response that felt too smooth. Something that didn't quite ring true.

But "ring true" is not a definition. It's a feeling. Passing the test doesn't prove a machine thinks. It proves it fooled someone. Those aren't the same thing.

A great actor can convince an audience they're devastated without feeling anything. ELIZA convinced people it understood them using nothing but : no learning, no inner life, just rules.

Did you notice how often ELIZA just repeated your words back? "Your boyfriend made you come here." "Your father." It rarely said anything new. It mirrored.

Here's how it worked: ELIZA scanned your message for certain words. If it found "mother", it asked about your family. If it found "I am", it reflected it back: "Why are you..."

No memory. Each response came from that single message alone, with no knowledge of what came before. It couldn't remember your name, your story, or anything you'd said. And yet people felt heard.

Turing never claimed the test would resolve this. Passing it doesn't prove understanding. It proves fooling. Turing wasn't claiming otherwise. He was saying it was a reasonable stand-in, given that we can't define the real thing. Today's AI works very differently from ELIZA, but whether it understands anything is a question we'll come back to.

Citations(1)↓
  1. 1. dl.acm.org