A Hacker News thread titled "Could a computer scientist build a brain?" surfaced last week and collected 38 points with 24 comments.
The discussion centers on the gap between current computational models and biological brains, with participants citing differences in energy use, learning rules, and verification methods.
Discussion: 38 points | 24 comments | Source: Hacker News thread
Core Technical Question
Participants asked whether standard computer science tools—algorithms, silicon hardware, and formal verification—can replicate brain function. Several comments noted that brains operate with roughly 20 watts while large language models require orders of magnitude more power for narrower tasks.
The thread highlighted that brains use spike-based, event-driven signaling rather than dense matrix multiplications common in current AI.
Key Arguments in the Thread
Early comments focused on verification limits. One line of discussion observed that formal proofs work for narrow algorithms but struggle with systems that must adapt continuously like biological tissue.
Other posts pointed to missing components: no equivalent of glial cells, no built-in developmental stages, and no native support for lifelong learning without catastrophic forgetting.
How Existing Projects Compare
Current large-scale brain simulations such as the Blue Brain Project and SpiNNaker use specialized hardware yet still model only small cortical columns at reduced biological fidelity.
| Approach | Scale Achieved | Power Draw | Verification Method |
|---|---|---|---|
| Blue Brain | ~1 million neurons | Megawatts (cluster) | Partial validation against slice data |
| SpiNNaker | Real-time spiking | ~1-10 watts per chip | Spike timing checks |
| Standard LLMs | Billions of parameters | 100+ kW training | Benchmark accuracy only |
Practical Barriers
Commenters listed concrete obstacles: lack of a unified theory of brain computation, inability to observe all relevant variables in vivo, and the absence of developmental feedback loops that shape biological networks.
One recurring point was that any engineered system would need continuous, unsupervised plasticity rules that current gradient descent methods do not provide.
Who Should Pay Attention
Researchers working on neuromorphic chips or continual learning frameworks can extract testable hypotheses from the thread. Practitioners building production AI systems will find little immediate engineering guidance.
The discussion is most useful for teams already exploring non-von-Neumann architectures or energy-constrained inference.
Bottom Line
The thread surfaces persistent gaps between computational theory and biological reality without offering a near-term construction path.
Early participants converged on the view that verification remains the hardest missing piece: without it, any candidate brain model stays unprovable at scale.
Future progress will likely require tighter integration between theoretical neuroscience and hardware that natively supports sparse, event-driven updates rather than further scaling of existing transformer stacks.
Top comments (0)