Skip to content
Exit presentation

How AI Learns II: Neural Networks Intuition

The big picture

One caretaker's log ran through every lesson. Eight ideas, each with one number you can check.

How AI Learns II — eight ideas, in order
IdeaRememberExample
Holding a rule insteadTwo weights and a threshold, kept instead of a pile of days.Gauge 6, days 0 → rule says wait
The weighted sumEach input times its own weight, then added.6 × 6 + 1 × 4 = 40
The line it crossesAt or past the threshold calls; under it waits.28 of 63 days call
Measuring one wrongA count needs its denominator, from a real walk.Missed 2 of the 4 kept-back
Nudging the weightsEach weight moves by the nudge times its own input.6 + 6 = 12, 4 + 0 = 4
Counting what movedGuess, compare, count, nudge, again. A count over a few days can hide what a nudge broke.Kept-back misses 2, 1, 0 of 4. All 12 days: 2, 3, 1
Write the sum yourselfTwo functions: the weighted sum, then the at-or-past check.Sum 36, decide → wait
What the word neural buysA neuron here is one unit. It adds inputs times weights, then compares the sum to a threshold. A network is many units joined. This track built one unit.No brain claim, no product

The finished rule gets 11 of 12 days right. One day cannot be fixed. Gauge 5, days 3 sums the same both times. The caretaker gave those two days different words. Two weights and a threshold cannot give two answers for one sum.

You stored a rule and moved it yourself. Your next track is AI Ethics & Society. It asks who chose the weights, and who a guess affects.

Jump to lesson

This track

  1. 0 Start here
  2. 1 Holding a rule instead
  3. 2 The weighted sum
  4. 3 The line it crosses
  5. 4 Measuring one wrong
  6. 5 Nudging the weights
  7. 6 Counting what moved
  8. 7 Write the sum yourself
  9. 8 What the word neural buys
  10. 9 The big picture
  11. 10 Track quiz

Other tracks