Probabilistic Classification Learning (Weather Prediction) Task

Description

On each trial show a combination of cues only probabilistically linked to two outcomes, ask for a prediction, give feedback, and track how accuracy grows over many trials.

In the weather prediction task four cue cards are each linked to 'sun' or 'rain' with probabilities of roughly 60 to 80 per cent. Each trial shows a combination of cards; the participant predicts the weather and is told the right answer. Because no single trial is reliable, memorising episodes does not help and learning accrues gradually, which made the task a standard probe of habit-like learning: amnesic patients learned it nearly normally while failing recognition tests. For UX, the task is a model of learning noisy cue-outcome relations from feedback, such as which signals predict a successful transaction, a reliable seller or a correct AI suggestion. Comparing the learning curve with the best achievable accuracy, and with explicit knowledge tests, shows how well and how consciously people learn.

Attributes

Loss type
Unmigrated
Effort
Days

Steps

  1. Define a small set of cues (typically four) and two outcomes, giving each cue a probability of predicting each outcome between about 60% and 80%.
  2. Generate trials as cue combinations with outcomes drawn from the combined probabilities, and compute the best achievable accuracy.
  3. On each trial show the combination, take a forced choice, then show the correct outcome.
  4. Run many trials (often 100-200 or more) and record accuracy by block.
  5. Afterwards test explicit knowledge: recognition of cues and questions about each cue's predictive strength.
  6. Compare learning curves across conditions (for example feedback formats or explanation displays) against the optimum.

Sources

  • Handbook of Cognition (Lamberts & Goldstone, 2005)

Also known as

  • Weather prediction task
  • Probabilistic category learning task
  • Multiple-cue probabilistic learning with trial feedback

Categories

  • User Research
  • Cognitive Psychology & Perception