EP.004 · Research talk · Nature 2022

The Simplified Mental Maps We Plan With

Mark K. Ho · Cognitive scientist

Hosted by Dr. Omar Costilla Reyes

First published On Interpretable Updated 58 minCognitive ScienceDecision-MakingInterpretable AI

Listen or watch

Download MP3(21 MB)
0:0058:05

Audio edition, 58 min. Keyboard: K play/pause, J back 15 s, L forward 30 s.

Full video, 58 min.

Chapters(18)

About this episode

Planning is computationally demanding, yet people plan quickly and flexibly every day. In this talk, cognitive scientist Mark Ho presents value-guided construal, a computational account proposing that people form simplified, task-specific representations of a problem in order to plan and act. Using maze-navigation experiments, he shows that what people notice and remember is shaped by what is useful for the plan, which opens a window onto metacognition, structured representations, and goal-directed behavior. The findings were published in Nature in 2022. On Interpretable, this episode turns the lens from machines to minds. Value-guided construal is an interpretable model of human reasoning, one whose assumptions can be stated, tested, and compared with what people actually do. That kind of understanding is what AI built to work alongside people will need.

Listen if: You are interested in decision-making, cognitive science, or AI that reasons in ways people can follow.

The research behind the episode

Nature2022Peer reviewed

People construct simplified mental representations to plan

Mark K. Ho, David Abel, Carlos G. Correa, Michael L. Littman, et al.

606: 129–136 · doi:10.1038/s41586-022-04743-9

In plain language

Research on planning in psychology, economics, and AI has usually assumed that people plan over a complete, fixed model of a task. This paper proposes instead that people control how they represent a task, simplifying it before they plan, and models that choice computationally.

Why it matters for health

A precise account of how people simplify decisions is a foundation for studying decision-making in health and illness, and for building AI that reasons in ways people can follow.

Key findings

  1. 01In preregistered experiments, how people simplified a task was subject to online cognitive control.
  2. 02People balanced the complexity of a representation against how useful it was for planning and acting.
  3. 03Strategically choosing how to see a problem helps people use limited cognitive resources well.

What you'll learn

  • Why planning is expensive, and how people work within limited cognitive resources
  • What maze experiments reveal about what people notice, ignore, and remember
  • How a formal model of construal connects attention, memory, and decision-making

Working on something like this?

Suggest a guest or a paper for a future episode, or talk to equ about AI for health in your own work.