HN Explorer

Ask HN: Is it OK to equate “experience” to “machine learning”?

2 points by zhirzh ↗ HN
I don't think my question explains itself very well, so allow me to elaborate.

If we talk about poker, the rules are few and simple. Yet, it takes some time to really master the game. The experience gained helps in making better decisions by having some sort of "insight" about the game that a beginner lacks.

Can we compare "experience" is the edge weights in a neural network? In my understanding, AI has 2 roles thus far: 1. draw some form of statistical inference from data by finding patterns - the experience 2. use the gained experience to solve a problem

2 comments

[ 137 ms ] story [ 124 ms ] thread
kafkaesq
A qood question - I'd say the best answer is "no" simply because "experience" and "insight" imply some higher-order cognition at play ("insight" for sure).

What ML algorithms do is more akin to curve-fitting, or (chains of) "if-this-then-that" type of association. Yes this is comparable to "experience", but only at very simplistic level.

zhirzh
i'd argue that the "if, then" logic is present in "insight" too.

Like, if you play any card game, you can't really calculate the probabilities of cards turning up, but there are certain patterns that give an estimate of what might happen. Like in texas poker, if the cards on the table are 8H 7H 5H, and a player has 6H 10D, there's a high probability that it is a winning hand, but we can't know for sure whether it will win since we don't know what cards other players have.

I believe a poker AI might reason the same way.