Skip to content

What It Would Actually Take for a Machine to Have Agency (Michael Tomasello)

Machine Learning Street Talk (MLST)

Hosted by Tim Scarfe

1h 6m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

In this episode, developmental psychologist Michael Tomasello discusses the evolutionary origins and structure of agency, contrasting biological organisms with artificial intelligence and large language models. The conversation traces agency across a four-level taxonomy, moving from basic control systems in reptiles to goal-directed mammals, metacognitive great apes, and uniquely human shared and cultural agency. The guest and host debate whether machines can ever possess genuine agency, how human goals and emotions bias decision-making, and why human cooperation and perspectival communication serve as the missing infrastructure for true artificial general intelligence.

  • Define agency as a control system mechanism that handles unpredictable environments through active goal pursuit rather than simple stimulus-response behavior. 3:54
  • Recognize that nature uses emotions to influence decisions, serving as biological biases that help organisms navigate immediate survival threats. 28:17
  • Understand that human cognition relies on shared intentionality and perspective-taking, which develop through joint social interaction rather than isolated computation. 38:41
  • View cultural conventions and linguistic structures as coordination tools that emerge from shared agency and collective social infrastructure. 47:53

Chapters

  1. 0:00
    Podcast Advertisements

    The episode opens with brief promotional spots for SAP Concur and Sephora.

  2. 0:30
    Host Introduction and Guest Overview

    Host Tim Scarfe introduces the episode and the guest, Michael Tomasello, and previews the discussion on machine agency.

  3. 1:23
    Levels of Agency in Evolution

    The conversation moves to Tomasello’s taxonomy of agency, from basic control systems to metacognitive agency in humans.

  4. 8:55
    AI and the Possibility of Machine Agency

    The hosts explore whether current AI can develop true agency, considering executive layers, learning, and the need for shared goals.

  5. 16:00
    Language, Conventions, and Perspective

    Tomasello highlights how language conventions and perspectival grammar arise from shared agency and are essential for complex communication.

  6. 52:26
    Perspective Grammar and Communication

    The discussion delves into how grammatical constructions reflect perspective-taking and correct misunderstandings in dialogue.

  7. 53:00
    Future Directions for AI Agency

    The episode concludes with speculation on how AI might evolve shared agency, create conventions, and achieve more human‑like cognition.

  8. 54:12
    Chomsky’s Critique of LLMs

    Discusses Chomsky’s view that LLMs can learn impossible languages, underscoring their lack of agentic storytelling.

  9. 54:48
    Infrastructure of Language and Pointing Gestures

    Examines how shared gestures and perspective-taking form the foundational infrastructure for language and agency.

  10. 1:00:26
    Cooperation, Lying, and Collective Intentionality

    Explores the cooperative nature of humans, how lying exploits this, and the emergence of collective intentionality.

  11. 1:02:26
    Virtualization of Culture and Fictional Entities

    Analyzes how cultural virtualization creates fictional entities like corporations and presidents, and its implications for agency.

  12. 1:04:53
    Market as Superorganism and Policy Implications

    Considers the market as a superorganism, discusses emergent phenomena, and the need for policies linking individual decisions to outcomes.

Highlights

  • 3:54

    Agency Handles Unpredictable Environmentsinsight

    Mother nature builds decision-making agent architectures specifically to deal with unpredictable situations where rigid, pre-programmed behaviors fail.

  • 10:31

    Evolution Focuses on Controlinsight

    Evolutionary complexity is not about building intricate structures like spider webs, but about having the control to guide flexible actions based on immediate perception.

  • 20:21

    LLMs Start at the Endinsight

    Large language models start at the pinnacle of human achievement with language, skipping the fundamental cognitive steps that made language possible.

  • 38:09

    Four Levels of Agencystrategy

    The evolution of agency moves through distinct tiers: basic control systems in reptiles, goal-directed intentional agency in mammals, metacognitive agency in great apes, and shared cultural agency in humans.

  • 40:47

    The Desert Island Thought Experimentstory

    A child raised on a desert island would never develop the notion of perspective because they would lack contrasting viewpoints present in social interaction.

  • 42:43

    Chimpanzees Versus Human Childrendata

    Experimental studies show that human children collaborate with joint commitments, correcting partners and sharing rewards, whereas chimpanzees lack this shared agency.

  • 57:49

    Fictional Status Functions in Cultureinsight

    Human culture builds powerful virtualization through status functions, turning pieces of paper into money and creating fictional corporate entities that run the world.

Quotes

  • “When you need a decision‑making agent is when the environment is unpredictable.”
    Unnamed speaker · 1:00
  • “I prefer talking about cognitive capacities, knowledge, planning, foresight, hypothesis testing, all the language of cognitive science, I prefer that to the language of intelligence, just because of its historical associations.”
    Unnamed speaker · 32:57
  • “The dog is chasing the cat, the cat is being chased by the dog, the passive construction. Actives and passives come from the perspective of the dog or from the perspective of the cat.”
    Unnamed speaker · 52:26
  • “Humans not only use language, they created languages, they created them.”
    Unnamed speaker · 43:34
  • “And I think, again, that's the, that's the infrastructure.”
    Unnamed speaker · 54:48
  • “And the most powerful forces on the world, Harari argues, is like corporations.”
    Unnamed speaker · 58:30
  • “I mean, I would say if it, if it walks like a duck and it quacks like a duck, it's a duck.”
    Unnamed speaker · 1:06:10
  • Cyber Fund 8:19–9:00

About this episode

Michael Tomasello is a developmental psychologist who spent twenty years at the Max Planck Institute in Germany putting children and great apes into the same experiments, and is now back at Duke University. His book The Evolution of Agency asks what it takes for a living thing, or a machine, to act for its own reasons. SPONSOR: --- Cyber Fund built the Monastery to help founders ship products that were impossible a year ago. Apply now: https://cyber.fund --- His answer starts with unpredictability. Where the environment is predictable, nature builds the behaviour in; where it isn't, it builds a decision maker. Tim and Tomasello test that on a dumb traffic light, a thermostat and a robot vacuum, then follow the line to humans: reptiles running simple control loops, a squirrel that imagines itself falling and takes another branch, and great apes that go looking for more information before they commit. The second half is about shared agency. Humans form joint goals, attend to the same things together, and learn that one object can be seen from different perspectives. Children who lose the sound on a video call invent gestures and turn them into conventions. Large language models, Tomasello argues, started at the endpoint and skipped that infrastructure. On the way: the stag hunt, why lying only works among cooperators, whether capital is an agent, why a Turing test isn't close to sufficient, and Dennett's intentional stance. TABLE OF CONTENTS: 0:00 Cold open: agency begins with unpredictability 0:29 Michael Tomasello and The Evolution of Agency 0:59 From chimpanzees to two-year-olds 2:05 Agents exist because the world is unpredictable 3:56 Defining agency: control loops, will and causal power 7:19 Sponsor: Cyber Fund 7:55 How would you make a traffic light smart? 9:57 Thermostats, robot vacuums and what is missing 11:32 Could a machine have more agency than we do? 16:26 Darwin, emergence and the limits of learning 21:04 Selfish genes, Chomsky and where goals come from 24:17 Modules, emotions and nature's finger on the scale 28:25 Why Tomasello avoids the word intelligence 32:06 Four levels of agency: reptiles, squirrels, great apes 37:04 Shared agency, joint attention and perspective 42:25 Why language models do not create conventions 44:08 The stag hunt and conventions as coordination devices 47:41 Perspective, grammar and why discourse matters 53:12 Pointing and building agents from the bottom up 55:44 Money, presidents and status functions 59:26 Lying, gullibility and the dark side of cooperation 1:01:27 Is capital an agent? Emergence and the invisible hand 1:04:48 Acting like an agent: the intentional stance 1:06:16 Closing BOOK: Michael Tomasello, The Evolution of Agency: Behavioral Organization from Lizards to Humans (MIT Press, 2022) https://mitpress.mit.edu/9780262047005/the-evolution-of-agency/ KEY REFERENCES: [11:41] Benjamin Recht, The Irrational Decision (Princeton UP, 2026) https://doi.org/10.1515/9780691272467 [28:07] Mark Solms, The Hidden Spring (Profile Books, 2021) https://profilebooks.com/work/the-hidden-spring/ [37:42] Michael Tomasello, The role of roles in uniquely human cognition and sociality (2020) https://doi.org/10.1111/jtsb.12223 [38:15] Cheng, Gao, Tomasello & Kushnir, The Agency Gap (CogSci 2026) https://escholarship.org/uc/item/8mv6v0s1 [38:59] Tomasello, Melis, Tennie, Wyman & Herrmann, Two Key Steps in the Evolution of Human Cooperation (2012) https://doi.org/10.1086/668207 [40:28] Michael Tomasello, The Normative Turn in Early Moral Development (2018) https://doi.org/10.1159/000492802 SUPPORT MLST: Patreon: https://www.patreon.com/mlst Discord: https://discord.gg/HNnAwSduud

Transcript

0:00
1:06:33

11,484 words · 7 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • Sephora
  • Capital
  • SAP Concur
  • The Evolution of Agency
  • Invisible hand
  • Turing test
  • Emory University
  • York East Primate Center
  • Max Planck
  • Duke University

Page updated · transcript and summary are machine-generated