Two Choice Task Learning Model
- One task we did both in CABot2 and 3 was learn the correct
action to choose.
- We ran this task on some further data from Friedman.
- Here the user was given two options and rewarded for the
correct choice.
- We added a random activation network and a reward network
to modify the Hebbian learning.
- Initially, when offered a choice the random activation network
chooses an action randomly.
- If it chooses correctly, that choice is strengthened.
- If it chooses incorrectly, another is randomly chosen and strengthened.
- In the long run, this leads to the correct behaviour.
- This is a simple form of reinforcement learning.
- People don't get the reward percentages correct, tending
to choose a value more toward 50%.
- This probably has something to do with exploration
vs. exploitation.
- Our system duplicates these results.
