Agents
Living
Artificial
Michael Levin draws every agent as a cone in space and time: how far away the things are that it can pursue, how far back it remembers, how far ahead it works toward. A bacterium cares about sugar a few microns away, a few seconds from now. A human can work for a planet a century from now. Cells that join into a tissue share one bigger cone. Cells that break away, as in cancer, shrink back to their own.
One cell on its own. Its goals are its own metabolism and the patch of medium around it.
A single chat turn. It remembers this conversation and works toward one reply.
Left and right: how far away the goals can be, on a log scale from microns to the planet. Up: how far into the future the agent works toward. Down: how far back it remembers. Every gridline is ten times the one before it.
The pink dashed cone is the one you are building. Positions are order-of-magnitude estimates for discussion, not measurements.
Levin argues that the boundary of a self is set by what its parts can share: signals, stress, memories. Couple cells and their goals merge into a body-scale goal like "build a hand". The same logic applies to any collective.
A cell that loses its electrical coupling forgets the body's goals and goes back to single-cell goals: grow, divide, migrate. Levin's lab has reversed tumor-like growth in frogs by restoring the bioelectric connection, not by killing cells.
A single model call has a tiny cone. Memory, tools, schedules and teammates stretch it. A human-AI team can pursue goals neither holds alone. The open question for this class: what makes the cone of a team more than the sum of its members?