Twenty watts
Sapiens-1 runs on about 20 watts. Where the energy goes, and what the upgrade cost.
A twenty-watt runtime
Sapiens-1 runs inference on roughly 20 watts. The brain is about 2 per cent of body mass and uses about 20 per cent of the body’s energy (Balasubramanian 2021). A direct count in adult male brains found about 86 billion neurons and a similar number of non-neuronal cells (Azevedo et al. 2009).
Scaling the primate architecture
Is that budget unusual? Herculano-Houzel argues that it is not: the human brain is a scaled-up primate brain, in its cellular composition and in its metabolic cost, and it is remarkable mainly because it has an extremely large number of neurons (Herculano-Houzel 2012). On this view there is no special overhead per neuron. There are simply a great many of them.
Where the budget goes
Most of the energy goes to signalling. An energy budget for grey matter in the rodent brain predicts that action potentials account for 47 per cent of signalling costs, postsynaptic effects of glutamate 34 per cent, resting potentials 13 per cent and glutamate recycling 3 per cent (Attwell & Laughlin 2001). The same analysis predicts distributed codes with no more than about 15 per cent of neurons active at once, which saves energy and allows more computing from a fixed number of neurons (Attwell & Laughlin 2001).
Training has a different cost
Training is where the bill comes due. Combining PET and MRI data, Kuzawa and colleagues found that the brain’s glucose use peaks not at birth but in childhood, at 66.3 and 65.0 per cent of resting metabolic rate in boys and girls, and that the ages of peak demand coincide with the slowest body growth (Kuzawa et al. 2014). The growing model appears to be paid for by growing the device more slowly.
The training budget peaks in childhood
Brain glucose / resting metabolism (%)
Published estimates combining PET and MRI data. Brain glucose uptake and resting metabolic rate are expressed in glucose-gram equivalents.
Birth and childhood-peak values are separate reported endpoints; the figure does not interpolate an age trajectory.
How the upgrade was funded
How the species afforded the upgrade is disputed. The expensive-tissue hypothesis proposed that a larger brain was paid for with a smaller gut (Aiello & Wheeler 1995). A test across 100 mammal species, including 23 primates, found no negative correlation between brain size and the mass of the digestive tract or any other expensive organ, but did find a trade-off between brains and fat storage (Navarrete et al. 2011). Its authors propose that human brain expansion was made possible by more stable energy inputs and energy redirected from locomotion, growth and reproduction (Navarrete et al. 2011).
References
- Aiello, L. C. & Wheeler, P. The expensive-tissue hypothesis: The brain and the digestive system in human and primate evolution. Current Anthropology 36, 199–221 (1995). doi:10.1086/204350
- Attwell, D. & Laughlin, S. B. An energy budget for signaling in the grey matter of the brain. Journal of Cerebral Blood Flow & Metabolism 21, 1133–1145 (2001). doi:10.1097/00004647-200110000-00001
- Azevedo, F. A. C. et al. Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. Journal of Comparative Neurology 513, 532–541 (2009). doi:10.1002/cne.21974
- Balasubramanian, V. Brain power. Proceedings of the National Academy of Sciences 118, e2107022118 (2021). doi:10.1073/pnas.2107022118
- Herculano-Houzel, S. The remarkable, yet not extraordinary, human brain as a scaled-up primate brain and its associated cost. Proceedings of the National Academy of Sciences 109 (Suppl. 1), 10661–10668 (2012). doi:10.1073/pnas.1201895109
- Kuzawa, C. W. et al. Metabolic costs and evolutionary implications of human brain development. Proceedings of the National Academy of Sciences 111, 13010–13015 (2014). doi:10.1073/pnas.1323099111