AI system self-organises to develop features of brains of complex organisms

Nature and significance of the model

• Some commenters stress this is a model of neurons and network structure, not of “intelligence” per se.
• The core setup is described as adding spatial distance and wiring-cost penalties to a recurrent neural network; under these constraints, hub-like structures and multiplexed representations emerge.
• A few argue this is almost trivial or previously known: if you minimize wiring cost, hubs appear. Others counter that the paper is careful, well-designed, and valuable particularly for neuroscience, not just AI.
• There is debate over whether such brain-like constraints improve performance or are merely artifacts of biology that artificial systems need not copy.

“Playing God” and existential risk

• One long branch debates whether advances like this bring us closer to “playing God” — defined by some as creating uncontrollable, proliferating technologies with irreversible large-scale harms.
• Several argue humans have been “playing God” for millennia (agriculture, fossil fuels, climate change, money, property) and that this is inherent to civilization and science.
• Others warn that our cognitive limits and history of unintended consequences make pushing toward AGI especially dangerous, invoking nuclear-arms-race dynamics and asymmetric power for early adopters.
• There is concern that global coordination is weak: even if most actors enforce safety, a small minority can race ahead; analogies to mutually assured destruction and to engineered pathogens are raised.
• Some suggest multiple competing AIs for balance; others argue self-improvement makes “first mover” dominance and rapid loss of control plausible.

Technology, quality of life, and death

• A strong anti-tech strand claims modern technology (AI, smartphones, fossil fuels, industrial food) has reduced human well-being despite material gains; ideal is a simpler, nature-connected life.
• Opponents counter that modern comforts, health care, and survival rates depend on high tech and petrochemicals.
• There is dispute over whether people in the past were happier, with pushback noting selection bias in historical records and the prevalence of disease and early death.
• Another subthread debates curing all diseases: critics fear overpopulation and “dehumanizing” immortality; others find it morally abhorrent to accept avoidable suffering.

Neural architectures vs biological brains

• Several comments note modern ANNs are mostly feed-forward and 2D, whereas brains are 3D, recurrent, continuously active, and heavily feedback-driven.
• Recurrent networks, attention, spiking networks, and spatial constraints are mentioned as partial moves toward biological realism, but they are often harder to train and less practical than standard backprop-based, layered models.
• Some argue human-like intelligence may require brain-like constraints (spatiality, delays, embodiment, multi-modal inputs); others see those constraints as unnecessary for useful machine intelligence and mainly relevant to brain modeling.

Mathematical and structural side notes

• Discussion touches on grid-cell–like representations, multiplexed encodings, and analogies to Fourier transforms and other basis transforms where each unit encodes global structure.
• There is brief mention of distance-regularized networks for interpretability and of alternative transforms (e.g., Chebyshev-based methods) in graph and signal processing.