What a Squirrel Knows That Language Models Skip (Michael Tomasello)

M
Machine Learning Street Talk • Nov 19, 2026

Audio Brief

Show transcript
How did living organisms transition from simple feedback loops into sophisticated, shared social architectures capable of navigating a chaotic world? On this episode, evolutionary biologist Michael Tomasello explains how agency evolved as a vital survival strategy for environmental unpredictability. By the end of this audio summary, you will understand how decision making evolved from simple feedback loops to complex human societies, and why modern artificial intelligence is missing the core ingredient of real biological agency. The Biology of Agency Agency is a biological shield against a chaotic world. While predictable environments allow for hardwired, instinctual behaviors, chaotic environments require a flexible architecture that enables individuals to assess novel situations and make context sensitive decisions. Evolutionary biologist Michael Tomasello explains that when the environment is unpredictable, nature builds an architecture that allows an organism to pause, evaluate, and choose. The Thermostat Model of Agency At its simplest level, agency works just like the thermostat on your wall. This basic feedback loop requires three core components, which are a way to perceive the environment, an internal goal, and a decision making mechanism to modify behavior until reality aligns with that goal. If you want to make something as simple as a traffic light smart, it must be able to perceive, have a goal, and decide how to act based on what it senses. The Four Tiers of Agency Agency evolved hierarchically. It began with Goal Directed Agents, like reptiles operating on basic feedback loops, followed by Intentional Agents, which are mammals that use working memory to mentally simulate choices so their hypothesis dies instead of themselves. Next came Rational Agents, like great apes that use metacognition to monitor uncertainty, and finally, Normative or Shared Agents, which are humans capable of cooperative planning and social self regulation. Emotions as Biological Fingers on the Scale Your feelings are not irrational glitches in your decision making. Instead, emotions are evolutionary mechanisms designed to instantly prioritize survival goals. Fear, anger, or hunger act as homeostatic biases, which are internal systems that immediately redirect an agent's focus under stress, though advanced metacognition allows some organisms to consciously observe these biological signals and override them. Agency versus Learning in Artificial Intelligence Modern artificial intelligence is built on a fundamental misunderstanding of nature. In the biological world, learning is merely a supportive sub process meant to help an organism make real time decisions, rather than being the ultimate goal. Current AI models focus almost entirely on pattern recognition and language generation from massive datasets, missing the goal driven, feedback control architectures that define genuine biological agency. Perspectival Flexibility and Shared Agency The defining line between humans and other great apes is our ability to form a collective "we". Humans can put their heads together to form shared goals, joint commitments, and mutual social norms where each person has a specific role to play. This shared agency fosters perspectival flexibility, which is the unique capacity to view a single object through multiple conceptual lenses depending on the social context, forming the very foundation of human language and culture. Status Functions and Virtual Realities Human societies operate inside massive, cooperative illusions. We collectively assign status functions to physical things, such as treating a piece of paper as valuable money or a specific citizen as a president. These social fictions exist solely through shared agreement, creating a virtualized reality that allows humans to coordinate behavior on an unprecedented, massive scale. To leverage these insights, design artificial intelligence with a control loop framework of perception, goal, and action rather than relying solely on data driven learning patterns. Cultivate perspectival flexibility in your own communication by actively examining problems through different social, cultural, and contextual lenses, while explicitly defining joint goals and individual roles to maximize team success.

Episode Overview

  • Evolutionary biologist Michael Tomasello explores how agency evolved in living organisms as a survival strategy for navigating environmental unpredictability.
  • The discussion traces the evolutionary history of decision-making from simple feedback loops in reptiles to the sophisticated, shared social architectures of humans.
  • The conversation challenges current assumptions in artificial intelligence by framing learning as a subordinate tool of decision-making rather than its ultimate goal.
  • This episode is ideal for those interested in evolutionary biology, cognitive science, the nature of human cooperation, and the biological differences between animal, human, and artificial intelligence.

Key Concepts

  • The Biology of Agency: Agency is an evolutionary adaptation designed to help organisms navigate unpredictability. While predictable environments allow for hardwired, instinctual behaviors, chaotic environments require a flexible architecture that enables individuals to assess novel situations and make context-sensitive decisions.
  • The "Thermostat" Model of Agency: At its simplest level, agency functions as a basic feedback loop requiring three core components: a way to perceive the environment, an internal goal or reference point, and a decision-making mechanism to modify behavior until reality aligns with that goal.
  • The Four Tiers of Agency: Agency evolved hierarchically: Goal-Directed Agents (reptiles operating on basic feedback loops), Intentional Agents (mammals with working memory to simulate choices), Rational Agents (great apes with metacognition to monitor uncertainty), and Normative/Shared Agents (humans capable of cooperative planning and social self-regulation).
  • Emotions as Biological "Fingers on the Scale": Rather than being irrational disruptions, emotions are evolutionary mechanisms designed to prioritize survival goals. Fear, anger, or hunger act as homeostatic biases that instantly redirect an agent's focus, though advanced metacognition allows some organisms to consciously override these drives.
  • Agency vs. Learning in Artificial Intelligence: In nature, learning is merely a supportive sub-process meant to assist an organism in making adaptive, real-time decisions. Modern AI models focus almost exclusively on pattern recognition and language generation from massive datasets, missing the goal-driven, homeostatic control architectures that define genuine biological agency.
  • Perspectival Flexibility and Shared Agency: What distinguishes humans from other great apes is the ability to form a collective "we" with shared goals, joint commitments, and mutual social norms. This shared agency fosters perspectival flexibility—the capacity to view a single object through multiple conceptual lenses depending on social context—which serves as the foundation for language and culture.
  • Status Functions and Virtual Realities: Human culture functions by collectively assigning "status functions" to physical things, such as treating paper as money or a specific citizen as a president. These social fictions exist solely through shared agreement, creating a virtualized reality that allows humans to coordinate behavior on an unprecedented, massive scale.

Quotes

  • At 0:02:54 - "When you need a decision-making agent is when the environment is unpredictable... What mother nature builds is an architecture enabling the individual to assess the situation and make a context-sensitive decision with flexible behavior." - Explains why agency evolved as a survival mechanism for chaos rather than relying solely on pre-programmed instincts.
  • At 0:06:23 - "Agents are in control of what they're doing, of the causal effects they're having in the environment... Within the context of the organism's biological adaptations, there is space for making certain decisions." - Defines agency as a bounded but genuine capacity for self-directed control, separate from absolute philosophical free will.
  • At 0:08:18 - "How would you make [a traffic light] smart? Of course, there's only one solution: first of all, it has to have a way of perceiving... it has to have a goal... and it has to have a decision-making mechanism." - Demystifies intelligence and agency by reducing them to the fundamental components of a feedback control loop.
  • At 0:13:38 - "It's a whole another level of organization—I call it a tier of organization. And the executive tier observes, perceives, pays attention to... what's going on at the attention and action level, and monitors it." - Introduces the structural concept of metacognitive tiers in advanced animal brains.
  • At 0:17:36 - "In evolution, what happens all the time is what they call pre-adaptation... You have something adapted for one thing, and then now you're in a new context, and the organism—the agent—figures out creatively how to apply it in a new way." - Highlights how agency enables behavioral creativity and the evolutionary reuse of existing traits.
  • At 0:18:25 - "Learning only exists to help the agent make adaptive decisions. So learning is a sub-process... learning is not the be-all and end-all." - Challenges the heavy emphasis on learning in AI by framing it as merely a supportive tool for decision-making.
  • At 0:22:32 - "Nature puts her finger on the scales in various ways of your decision-making... and your sociology puts its finger on the scale... but you the agent have to resolve moral dilemmas." - Explains how biological drives and cultural norms influence us without completely stripping away our personal agency to choose.
  • At 0:27:17 - "In an evolutionary sense, the emotions are nature putting its finger on the scale of your decision-making." - Explains how emotions serve as evolutionary survival overrides, instantly re-prioritizing behavior during threats.
  • At 0:30:19 - "If you're thinking in terms of the natural world and other animals, general intelligence is not the best construct... there is a lot of specialization." - Highlights why comparing different species using a singular metric of intelligence misses the ecologically specialized nature of cognition.
  • At 0:34:25 - "Daniel Dennett said it one time... that with intelligence comes [the reality] that my hypothesis dies instead of myself." - Illustrates the survival value of intentional agency, which allows mammals to mentally simulate and reject failing plans before physically executing them.
  • At 0:37:42 - "What makes humans different from other great apes is that we can form shared agents... we put our heads together, we have a shared goal, and you have your role and I have my role." - Establishes the foundational human transition from individual strategic agency to collaborative social agency.

Takeaways

  • Look beyond standard IQ or general-intelligence metrics when evaluating capability, recognizing that both biological systems and teams thrive on highly specialized, context-specific cognitive traits.
  • Design artificial intelligence and decision-making software with a "control loop" framework (perception, goal, action) rather than relying solely on data-driven learning patterns.
  • Recognize emotions as functional biological signals that reprioritize focus under stress, allowing you to consciously observe these "fingers on the scale" rather than reacting automatically.
  • Cultivate "perspectival flexibility" in communication and conflict resolution by actively examining problems through different social, cultural, and contextual lenses.
  • Leverage the power of shared agency in teams by explicitly defining joint goals, mutual commitments, and individual roles to maximize collaborative success.
  • Beware the "dark side" of human cooperation, actively monitoring and mitigating the natural evolutionary bias to favor the in-group while displaying distrust toward the out-group.