How Much Computational Power Does It Take to Match the Human Brain?

Open Philanthropy is interested in when AI systems will be able to perform various tasks that humans can perform (“AI timelines”). To inform our thinking, I investigated what evidence the human brain provides about the computational power sufficient to match its capabilities. This is the full report on what I learned. A medium-depth summary is available here. The executive summary below gives a shorter overview. Table of Contents Introduction Executive summary Caveats Context FLOP/s basics Neuroscience basics Uncertainty in neuroscience Clarifying the question Existing literature Mechanistic method estimates Functional method estimates Limit method estimates Communication method estimates The mechanistic method Standard neuron signaling Synaptic transmission Spikes through synapses per second FLOPs per spike through synapse A simple model Possible complications Firing decisions Predicting neuron behavior Standards of accuracy Existing results Dendritic computation Crabs, locusts, and other considerations Expert opinion and practice Overall FLOP/s for firing decisions Learning Timescales Existing models Energy costs Expert opinion Overall FLOP/s for learning Other signaling mechanisms Other chemical signals Glia Electrical synapses Ephaptic effects Other forms of axon signaling Blood flow Overall FLOP/s for other signaling mechanisms Overall mechanistic method FLOP/s Too low? Too high? Neuron populations and manifolds Transistors and emulation costs Do we need the whole brain? Constraints faced by evolution Beyond the mechanistic method The functional method The retina Retina FLOP/s From retina to brain Visual cortex What’s happening in the visual cortex? What’s human level? Making up some numbers Other functional method estimates The limit method Bit-erasures in the brain Landauer’s principle Overall bit-erasures From bit-erasures to FLOP/s Algorithmic arguments Hardware arguments Overall weight for the limit method The communication method Communication in the brain From communication to FLOP/s Conclusion Possible further investigations Appendix: Concepts of brain FLOP/s No constraints Brain-like-ness Findability Other computer analogies Summing up Sources Endnotes Introduction Back to TopExecutive summary

Let’s grant that in principle, sufficiently powerful computers can perform any cognitive task that the human brain can. How powerful is sufficiently powerful? I investigated what we can learn from the brain about this. I consulted with more than 30 experts, and considered four methods of generating estimates, focusing on floating point operations per second (FLOP/s) as a metric of computational power.

Leave a Comment