École d'été | 4 juin 2026 : Decision making in cognitive architectures

Value of the Information & Strength of the Argument

The talk provides a valuable synthesis of decision-making in cognitive architectures, offering clear definitions and a structured taxonomy of reasoning types. The speaker effectively argues for the importance of cognitive architectures as a unifying framework, drawing on foundational literature. However, the argumentation is largely descriptive rather than critical, and the speaker does not deeply engage with potential counterarguments or limitations of the presented approaches. The value lies in its comprehensive overview and the connection between classical cognitive architectures and modern LLMs.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates scientific rigor by grounding concepts in seminal works (Newell, Marr) and referencing specific cognitive architectures (e.g., SOAR, ACT-R, Companion). However, the speaker does not provide explicit citations for many claims, and the reliance on her own book as a primary source may introduce bias. The title accurately reflects the content, which focuses on decision-making mechanisms in cognitive architectures, contrasting rule-based and data-driven approaches.

The title accurately reflects the content, which focuses on decision-making mechanisms in cognitive architectures, contrasting rule-based and data-driven approaches.

8/10

The talk is based on a recent book by the speaker, providing a structured overview of decision-making in cognitive architectures. It references foundational works (Newell, Marr) and specific architectures, but lacks detailed citations or empirical validation.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction: speaker introduces herself and the topic of decision-making in cognitive architectures.
• 05:00Definition of cognitive architectures as blueprints for intelligence, referencing Newell's 1973 paper.
• 10:00Discussion of challenges in unifying cognitive theories: definitional issues, experimental precision, and falsifiability.
• 15:00Introduction of Marr's levels of abstraction and how they apply to cognitive architectures.
• 20:00Distinction between reasoning about beliefs and reasoning about actions.
• 25:00Explanation of monotonic vs. non-monotonic reasoning and types of logical reasoning.
• 30:00Discussion of action selection: deliberative planning vs. reactive approaches.
• 35:00Introduction of rule-based systems and their role in decision-making.
• 40:00Discussion of LLMs as data-driven approaches to decision-making.
• 45:00Emphasis on the importance of evaluation in cognitive architectures.

• You can't play 20 questions with nature and win — Referenced as the 1973 paper by Allen Newell that introduced the concept of cognitive architectures.
• Unified Theories of Cognition — Referenced as the 1990 book by Allen Newell that fleshed out the concept of cognitive architectures.
• Vision — Referenced as the 1982 book by David Marr that introduced levels of abstraction.

• Cognitive architectures: Research issues and challenges — This paper provides a comprehensive review of cognitive architectures, aligning with the talk's overview.

• The Emperor's New Mind — Penrose argues that computational models of cognition are insufficient, contrasting with the talk's assumption that cognitive architectures can capture intelligence.