What do our sampling assumptions affect: How we encode data or how we reason from it?

Area of Science:

• Cognitive Psychology
• Machine Learning Theory
• Human Learning

Background:

• Inductive inference theories traditionally focus on sample content for generalization.
• Recent research highlights the impact of data generation assumptions on inference.
• The interplay between sample content and generation assumptions remains underexplored.

Purpose of the Study:

• To investigate how and when sampling assumptions are combined with sample content during generalization.
• To determine if sampling assumptions influence encoding or retrieval processes.

Main Methods:

• Two experiments were conducted manipulating the timing of sampling cover stories (before vs. after training stimuli).
• Generalization from observed data to novel cases was measured.
• The influence of sampling assumptions on encoding versus retrieval was analyzed.

Main Results:

• Sampling cover stories significantly impacted generalization when presented before training stimuli.
• No significant effect was observed when sampling cover stories were presented after training stimuli.
• This suggests that sampling assumptions are integrated during the encoding phase.

Conclusions:

• Sampling assumptions are integrated during the encoding of information, not solely at retrieval.
• Understanding the timing of assumption integration is crucial for theories of inductive inference.
• This finding has implications for designing effective learning environments and AI algorithms.