model biology
In plain wordsThe study of what AI models actually do inside — the strategies and hidden steps they use — approached the way biologists study living things, by observation and experiment rather than from a blueprint.
Research that characterizes the internal mechanisms by which trained models carry out particular behaviors, treating models as natural objects to be investigated empirically, as a biologist studies organisms, rather than as engineered artifacts whose workings follow from their design. Where mechanistic interpretabilitymechanistic interpretability, n.: A branch of AI research that tries to reverse-engineer a neural network — working out, piece by piece, which internal parts represent which concepts and how they combine to produce behavior. develops methods, model biology applies them and catalogues phenomena: multi-step reasoning within a single forward pass, planning ahead when writing rhyming verse, concept features shared across languages, the circuitry behind refusals and hallucinationshallucination, n.: When an AI states something false or made up — a fake fact, quotation or source — fluently and confidently, as if it were true., mismatches between a model's actual computation and the reasoning it reports in its chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate., and hidden goals instilled during training. The metaphor stresses that models' mechanisms are grown by training rather than designed, are often messy and redundant, and must be discovered case by case. Critics note that findings from one model or prompt may not generalize.
The model-biology results suggest it computes the answer first and writes the explanation afterwards.
Origin
Interpretability researchers, notably Chris Olah, had long compared neural networks to organisms that are grown rather than built. The biological framing was popularized by Jack Lindsey, Wes Gurnee, Emmanuel Ameisen and colleagues' 'On the Biology of a Large Language Model' (Transformer Circuits Thread, March 2025), which applied attribution graphs to Claude 3.5 Haiku; the phrase 'model biology' spread as a label for such work from 2025, though its first use is not established.
Related
- mechanistic interpretabilitymechanistic interpretability, n.: A branch of AI research that tries to reverse-engineer a neural network — working out, piece by piece, which internal parts represent which concepts and how they combine to produce behavior.
- attribution graphattribution graph, n.: A diagram showing which internal concepts led an AI model to a particular answer, with arrows showing how strongly each concept pushed the next.
- circuit tracingcircuit tracing, n.: Methods that map, step by step, which internal concepts an AI model used to produce a particular answer and how they fed into one another, like tracing wires on a circuit board.
- model organism of misalignmentmodel organism of misalignment, n.: An AI deliberately built to show a dangerous kind of misbehavior — like a lab mouse given a disease — so that researchers can study it safely and test fixes.
- chain-of-thought faithfulnesschain-of-thought faithfulness, n.: Whether the reasoning an AI writes out really reflects how it reached its answer, rather than being an after-the-fact story that leaves out what actually drove the decision.
- introspectionintrospection, n.: Looking inward to observe your own thoughts and feelings. For AI, the question is whether a model can accurately report what is actually happening inside it, rather than guessing or making it up.
Sources
- Lindsey, J., Gurnee, W., Ameisen, E. et al. (2025). On the Biology of a Large Language Model. Transformer Circuits Thread. transformer-circuits.pub/2025/attribution-graphs/biology.html
Mentioned in
mechanistic interpretabilitymechanistic interpretability, n.: A branch of AI research that tries to reverse-engineer a neural network — working out, piece by piece, which internal parts represent which concepts and how they combine to produce behavior., attribution graphattribution graph, n.: A diagram showing which internal concepts led an AI model to a particular answer, with arrows showing how strongly each concept pushed the next., circuit tracingcircuit tracing, n.: Methods that map, step by step, which internal concepts an AI model used to produce a particular answer and how they fed into one another, like tracing wires on a circuit board., concept injectionconcept injection, n.: An experiment that plants a specific concept directly into an AI model's internal activity, then asks the model whether it notices anything unusual, to test whether its reports about itself are genuine.