Levels of Analysis: From Molecules to Minds

Why one behaviour can be explained at many levels at once — genes, molecules, synapses, neurons, circuits, brain systems and behaviour — and how David Marr's three levels changed the way neuroscientists ask questions.

A single event — say, jumping when a door slams — can be described at many levels. Ion channels opened in hair cells of the inner ear. Auditory neurons fired a burst of action potentials. A fast pathway through the brainstem triggered a startle reflex. The amygdala flagged a possible threat. You felt a jolt of fear. All of these are true; they are descriptions at different levels of analysis.

The ladder of scale

Neuroscientists often picture a ladder: genes and molecules (proteins, transmitters), synapses, neurons, microcircuits (small groups of connected cells), brain regions and systems (for example the visual system), and behaviour and cognition. Each rung spans several orders of magnitude in size and in time — from nanometres and microseconds up to the whole body and a lifetime of learning.

A full explanation usually needs more than one rung. Knowing which gene is altered in Huntington's disease does not, on its own, explain why patients develop involuntary movements; that needs the circuit level (the basal ganglia) as well.

Marr's three levels

In his 1982 book Vision, the British neuroscientist David Marr proposed another way of slicing the problem. Any information-processing system can be understood at the computational level (what problem is it solving, and why?), the algorithmic level (what steps or representations does it use?), and the implementational level (how is that physically built in neurons?).

Marr's point was that you cannot understand a bird's flight by studying feathers alone. You also need the aerodynamics. His framework is still used across neuroscience and artificial intelligence.

Reduction and emergence

A recurring debate is how far "lower" levels can explain "higher" ones. Most neuroscientists accept that the mind depends entirely on the brain, while recognising that new properties emerge when many parts interact — a traffic jam is made of cars, but you won't find the jam inside any one car.