Suboptimal Criterion Learning in Static and Dynamic Environments.

Area of Science:

• Cognitive psychology
• Neuroscience
• Decision science

Background:

• Signal detection theory (SDT) traditionally assumes fixed decision criteria.
• Recent evidence suggests trial-by-trial updating of decision criteria.
• The mechanisms of criterion setting remain largely unknown.

Purpose of the Study:

• To investigate how observers learn to set decision criteria in orientation-discrimination tasks.
• To explore criterion dynamics under static and dynamic conditions.
• To compare explicit criterion setting with traditional discrimination tasks.

Main Methods:

• Introduced a novel overt-criterion task alongside a traditional covert-criterion task.
• Used stimuli (ellipses) with orientations from two Gaussian distributions.
• Compared human performance to ideal Bayesian and suboptimal computational models.

Main Results:

• Observers employed suboptimal learning rules in both static and dynamic conditions.
• A model using recent sample history to infer category means best explained observer data.
• Dynamic criterion adjustment was observed even after extensive training.

Conclusions:

• Human decision criteria are not fixed but dynamically updated.
• Recent sensory history significantly influences criterion setting.
• Suboptimal learning rules, influenced by past experiences, govern criterion dynamics.