Loading slide

What Understanding Actually Is

  1. 01What Understanding Actually Is
  2. 02Back to the feeling
  3. 03The Chinese Room
  4. 04What Searle was saying
  5. 05Memory card
  6. 06The systems reply
  7. 07The Searlese Room
  8. 08It cuts both ways
  9. 09No right answer
  10. 10What humans actually do
  11. 11Does it understand?
  12. 12The line keeps moving
  13. 13Two kinds of intelligence
  14. 14The question Turing left us
  15. 15Reinforce your understanding
  16. 16Question: Chinese Room
  17. 17Question: The systems reply
  18. 18Question: Narrow vs general
  19. 19Quiz: answer
  20. 20Still open
  21. 21Want to go deeper?
13 / 20
BackNext

Two kinds of intelligence

There's a useful distinction here.

Narrow intelligence is the ability to do one specific thing very well. A chess engine beats world champions. A translation program translates better than most humans. An image classifier spots tumours in X-rays that radiologists miss.

General intelligence is what humans have: the ability to do many different things, to learn a new task from minimal instruction, to apply knowledge from one domain to a completely different one.

Most AI systems people use are narrow. Extremely powerful in their lane. Unable to leave it.

The chess engine that beats Magnus Carlsen cannot tie its shoes, understand a children's book, or navigate a new room without being specifically trained to do so.

Whether is something machines can ever have, or whether it requires something machines don't have, is the question this whole module has been preparing you to ask.

Did you know?

That gap you just read about, a machine that wins at chess but can't pick up the pieces, has a name: Moravec's paradox.
The surprise is that our oldest skills, like seeing and moving, run the deepest, and turn out to be the hardest of all to copy.
Citations(4)↓
  1. 1. ibm.com
  2. 2. ibm.com
  3. 3. plato.stanford.edu
  4. 4. mitpress.mit.edu