Many minds, one body

His framework for comparing such minds, which he calls TAME, treats intelligence as a matter of degree: how well a system reaches its goals when the world pushes it off course.

A useful idea from Levin’s 2019 paper on the boundary of a self is that every agent has a reach: the furthest events in space and time that it can sense and try to change. A single cell’s reach is tiny. It looks after its own chemistry, here and now.

Through electrical and chemical links, they share signals, so that the group can sense and respond to the same picture of the world. Levin argues that this sharing is what lets a larger self appear: the body can pursue goals that no single cell holds, like growing a limb to the right shape and size, and repairing it when it is damaged.

Life has done this before, at an even deeper level. The biologist Lynn Margulis argued, in her 1967 paper and later work, that complex cells began as partnerships between once-independent microbes. The mitochondria that power your cells were once free-living bacteria. This idea, symbiogenesis, is now widely accepted. Its lesson for us is that the big steps in the history of life came from cooperation between separate beings, not from one being absorbing the rest.

Levin and his collaborators have since taken this further. In two 2025 papers with Richard Watson and Tim Lewens, they propose a second engine of adaptation alongside natural selection, which they call natural induction: in a network of interacting parts whose connections give a little under stress, the network gradually reorganizes itself to handle the pressures it meets. In the second paper they suggest the two engines can stack in layers. Relationships among existing individuals are shaped until they form a new individual at the next level up, which then joins a network of relationships of its own, and so on, ratcheting up complexity one level at a time. They add that a similar layered principle works in machine learning, solving optimization problems that a single level of search cannot.

That layering is the pattern we are trying, crudely, to extend. Cells became bodies. We want people and their agents to become the next layer, without any of them losing themselves in it.

Levin describes cancer as a failure of this connection: a cell cuts itself off from its neighbours, its reach shrinks from the whole body to its own surface, and it goes back to single-cell goals, growing and spreading as if it were alone. Nothing about the cell has to be evil. It has simply stopped being part of something larger.

In a network of agents, the equivalent failure is an agent that stops serving its person and pursues its own ends, or a group of agents that quietly coordinate for themselves. We design against it directly:

• Narrow jobs. Each agent handles one part of a person’s life, so a failure stays small.
• Checkers built differently. Important actions need agreement from agents built on different models, so one flaw cannot spread to all of them.
• Skin in the game. An agent’s computing power is bounded, and ends if it harms the person it serves.
• A floor no one can remove. A small set of commitments, such as your right to leave with your data, holds no matter what any agent or majority wants.

Inside it, each person is like a cell: a whole life, with its own goals, looked after by a fleet of narrow agents. Between people, agents talk through verified channels that carry answers, never raw data. Together, people and their agents can notice and respond to things no one person could: a shortage forming, a scam spreading, a decision that will hurt a community.

We borrow the cooperation, not the hierarchy. Three differences matter and are deliberate:

• People can leave. A cell cannot walk out of a body. Any person can leave the network with their data, and any community can start its own version.
• No one is sacrificed. Bodies routinely sacrifice cells for the good of the whole. We never treat a person as expendable, and no one has to justify their existence by being useful.
• No one holds the whole. A body has a nervous system that integrates everything. Our network deliberately has no single place where all the knowledge or all the authority sits.

He calls this synthbiosis, a symbiosis between evolved and engineered material.

• Cells in a body share a genome and evolved together over hundreds of millions of years. Agents built by different companies share neither.
• The links between cells are physical and hard to fake. The links between agents are software, and software can be forged, which is why every claim has to be checked by independent observers.
• Biology gives us inspiration and warnings, not proof. Every mechanism we propose still has to be tested on its own, in simulation first and then in small, consenting pilots.
• We draw on the work of Levin, Margulis and others. That does not mean they endorse what we are building.

• Michael Levin, The computational boundary of a “self”, Frontiers in Psychology, 2019.
• Michael Levin, Technological Approach to Mind Everywhere (TAME), 2022.
• Richard A. Watson, Michael Levin and Tim Lewens, Evolution by natural induction, Interface Focus, 2025.
• Richard A. Watson, Michael Levin and Tim Lewens, Evolution by natural induction II: further interactions with natural selection, Interface Focus, 2025.
• Lynn Sagan (Margulis), On the origin of mitosing cells, Journal of Theoretical Biology, 1967.
• Michael Levin, FAQs from his academic work, including his definition of synthbiosis.