Michael Levin on Multi-Scale Intelligence and Teleophobia

A conversation about goals, agency, and why they matter for democracy research

In a wide-ranging interview hosted by the European Institute of Science and Management (EISM), Professor Michael Levin from Tufts University explores a non-traditional view of intelligence, the role of goal-directed behaviour in biology, and why scientific progress often depends on frameworks that are empirically testable rather than purely philosophical.

Although the conversation begins with biology and regenerative medicine, it repeatedly returns to a broader question highly relevant to the EuropeanCity² (EC2) project: how can we understand complex collective behaviour across different scales, from simple agents to large social systems?

Intelligence, Goals and Teleophobia

Professor Michael Levin proposes a practical way of understanding intelligence: not as a property of a particular biological structure or technological system, but as the capacity of an agent to navigate “problem spaces”. These spaces may be biological, behavioural, social or informational.

In this perspective, intelligence emerges when a system can move toward preferred states under changing conditions. Importantly, intelligent systems show flexibility – they can reach similar outcomes through different pathways when disruptions occur.

A central theme of the discussion is what Prof Levin calls teleophobia – the discomfort many scientific communities have with describing natural systems as goal-directed. He argues that this hesitation often stems from an outdated binary view of the world: either inert matter without goals, or humans with conscious intentions. Fields such as cybernetics and control theory demonstrate that goal-directed systems can exist naturally without implying consciousness.

The Cognitive Light Cone

In the interview, Levin treats agency as something that exists in degrees. Some systems pursue only local and immediate goals, while others can represent goals extending across wider spatial and temporal horizons.

To compare very different types of agents, he proposes a conceptual tool inspired by physics diagrams: the “cognitive light cone.”

An agent’s cognitive light cone represents the scale and time horizon of the goals it can pursue. Simple organisms may operate within very limited ranges of influence, while humans can formulate goals that extend far into the future and beyond their own lifespan.

Scaling Intelligence and Collective Minds

Another key theme of the interview is the scaling of intelligence. Levin examines how higher-level systems can emerge from networks of smaller components. In biology, cells can organise into tissues and organs through connectivity and information exchange, forming coordinated systems capable of pursuing larger goals.

However, such multi-level organisation also introduces risks. What benefits a collective system may not always benefit each component within it. Misalignment between system-level and individual-level goals can arise in biological systems, technological systems, and social systems alike.

Levin also critiques the term “artificial intelligence”, suggesting that the distinction between natural and artificial intelligence may become increasingly meaningless as biological and technological systems merge into hybrid and compositional forms.

Why These Ideas Matter for EuropeanCity²

The concepts discussed in the interview resonate strongly with the research agenda of EuropeanCity², which studies democratic decision-making through computational modelling and simulation.

Levin’s understanding of intelligence as navigation through complex spaces aligns with EC2’s approach to democracy as a system of interacting agents operating within opinion landscapes, policy trade-offs and information environments.

Rather than treating democracy as a static institutional structure, EC2 studies it as a dynamic process shaped by citizen behaviour, incentives, and decision rules. Through agent-based modelling, the project can simulate scenarios and analyse how different voting mechanisms perform under varying assumptions about information flows and collective preferences.

This simulation-based perspective makes it possible to empirically explore questions that are often debated only theoretically – including how mechanisms such as Quadratic Voting compare with traditional voting systems when representing the intensity of citizens’ preferences.