TALAMANA · A READY-MADE PATH

Bias, judgment, and trust

21 ideas, in order. about 35 minutes of listening. It ends at "Confident is not correct".

The path

  1. Machines learn from examples — The examples are the teacher, so bad examples teach badly.
  2. Prediction, not understanding — The machine produces what is likely, and likely is not the same as true.
  3. Machines get things wrong — A machine gives a wrong answer as fast and neatly as a right one.
  4. Rules that draw: the kolam — Rules can make form, and you have watched it happen before breakfast.
  5. Early algorithms: the Sulba Sutras — A rule that works is knowledge, even when nobody has written why.
  6. Scale, ratio, and proportion — A door two metres wide is drawn two centimetres wide, and every length does the same.
  7. Talamana: proportion as a rule system — A proportion rule is somebody's judgment written down so it can be repeated.
  8. Chance, and how to name it — A number between 0 and 1 names chance, and machines run on it.
  9. Same input, same output? — A rule is checked by reading it, and a draw by looking.
  10. Learned, not programmed — Nobody wrote this behaviour line by line, so there is no rulebook to read.
  11. Small models, large models — Size is one side of a trade, and size alone does not rank models.
  12. What "model" really means — A model is a trained mathematical function, and the manners come from the app.
  13. Every output is a draw — Judging a tool by one output is like judging a die by one roll.
  14. The name is from 1956 — The name was written down in 1955, and the field dates from the 1956 workshop.
  15. Promises, then winters — A student who knows the winters can tell a demonstration from a delivery.
  16. 2012: the examples won — The ideas were decades old, and enough labelled data and cheap computing arrived around 2012.
  17. Attention becomes the architecture — Most chatbots and many image tools share one 2017 ancestor, and so share its limits.
  18. Language models, the next-word engine — Nothing in the next-word loop checks whether the argument is true.
  19. Sampling: from a spread to a token — The model produces a spread of probabilities, and a decoding rule picks one token.
  20. Uncertainty has more than one meaning — An answer carries four kinds of confidence, and each is checked in a different place.
  21. Confident is not correct — Confidence is a property of the writing, not evidence about the world.

Take this path on the map · The complete map · Logika · RBDS AI Lab, India.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — LINKS
DONE
OPENS NEXT
SOLID — STANDS ON