Discovering skill

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Skill acquisition can occur through direct instruction or independent discovery.
• Understanding the causal structure of a task is key to effective learning.
• Previous models focused on instruction, leaving discovery less explored.

Purpose of the Study:

• To investigate how skills are acquired through discovery, mirroring instruction.
• To extend computational models of skill acquisition to include learning by discovery.
• To compare learning outcomes between instruction and discovery in a complex task.

Main Methods:

• Extended Anderson et al.'s skill acquisition model (ACT-R) to incorporate discovery.
• Modeled the discovery process as environmental exploration and association building.
• Conducted an experiment where participants learned a video game via instruction or discovery.
• Compared learning speed and final performance between the two groups.

Main Results:

• Instruction led to faster initial learning, but discovery learners eventually matched performance.
• Successful discovery learners' behavior became indistinguishable from instructed learners after sufficient practice.
• The behavior of successful discovery learners mirrored that of the computational model.

Conclusions:

• Identical skills can emerge from both instruction and discovery if causal structure is understood.
• Discovery learning integrates implicit (associative learning) and explicit (causal inference) processes.
• Computational models can effectively simulate learning by discovery in complex environments.