Quo vadis, planning?

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Neuroscience

Background:

• Deep meta-learning is a significant driver of current AI progress.
• It presents a compelling framework for understanding flexible cognition in natural intelligence.
• Classical models often explain behaviors that may be better understood through meta-learning.

Purpose of the Study:

• To highlight the importance of deep meta-learning in AI and cognitive science.
• To propose meta-learning as an alternative explanation for "model-based" behaviors.
• To advocate for a re-evaluation of neural theories concerning problem-solving and goal-directed planning.

Main Methods:

• Conceptual analysis and theoretical argumentation.
• Review and interpretation of existing research on meta-learning and model-based behaviors.
• Synthesis of findings to propose a new theoretical framework.

Main Results:

• Deep meta-learning provides a more parsimonious explanation for a range of complex behaviors.
• The distinction between meta-learning and classical model-based approaches requires re-examination.
• Evidence suggests that meta-learning principles underlie flexible cognitive functions.

Conclusions:

• The study argues for the central role of meta-learning in both artificial and natural intelligence.
• It calls for a paradigm shift in understanding problem-solving and planning mechanisms.
• Revisiting neural theories through the lens of meta-learning is crucial for future research.