The default assumption runs deep: cognition is what brains do. Neurons fire, representations form, behavior emerges. Everything else in biology—cells dividing, tissues folding, organs maintaining homeostasis—is mere mechanism. Chemistry following gradients. Physics doing what physics does. This assumption is so embedded in how we think about thinking that it barely registers as an assumption at all. Michael Levin’s work in developmental biology and bioelectricity has been systematically dismantling it for two decades, and the implications reach far beyond biology.
The core insight is this: goal-directed behavior exists at every biological scale, and it has existed there long before neurons evolved. Individual cells navigate “morphospace”—the space of possible anatomical configurations—toward target morphologies. They do not simply execute genetic programs. When you cut a planarian into pieces, each fragment rebuilds a complete organism with correct proportions. The cells are not reading a blueprint; they are solving a problem. They detect the current state, compare it against a stored target, and close the gap through coordinated action. Bioelectric networks—voltage gradients across cell membranes—encode this target morphology as a kind of pattern memory. Change the bioelectric pattern and you change the target: Levin’s lab has induced frogs to grow eyes on their tails, created two-headed planaria from single-headed species, and reprogrammed tumors back into normal tissue—not by editing genes, but by changing the bioelectric conversation.
What makes this a framework rather than a curiosity is Levin’s insistence on asking the right question. The binary “does it think or doesn’t it?” is the wrong question. The right question is: what does this system think about, and how well? A thermostat has a trivially small cognitive horizon—it represents one variable (temperature) and pursues one goal (setpoint). A cell navigating morphospace represents a richer set of variables and pursues more complex goals. An ant colony computes solutions to optimization problems that no individual ant represents. The question of cognition becomes quantitative rather than qualitative—a spectrum, not a binary switch.
Levin proposes what he calls a “cognitive lens”—a set of criteria for evaluating where any system sits on this spectrum. The criteria are not arbitrary. Each one indexes a specific computational capacity that, when present, indicates the system is doing something more than passively responding to stimuli. Together, they form a multi-dimensional map of cognitive sophistication that can be applied to thermostats, bacteria, neural networks, and everything in between.
Compare two systems on Levin’s six dimensions. Select preset systems or add your own evaluation.
The cognitive lens is not an invitation to anthropomorphize. Saying that a cell “pursues goals” does not mean it has subjective experience, desires, or anything resembling human consciousness. The claim is more precise and more radical: the computational structure of goal-directed problem-solving is present at scales where we habitually deny it. The same formal properties—error correction, internal representation, flexible response to novel perturbation—appear in systems that have no neurons, no central processor, no apparent locus of control. The spectrum is continuous. The boundaries we draw between “real” cognition and “mere” mechanism say more about our own pattern-recognition failures than about the systems themselves. Mind-blindness is not a philosophical position. It is a perceptual deficit—and like all perceptual deficits, the first step is noticing what you have been failing to see.