AI Masters Cosmology - Then Gets Stuck in Its Old Assumptions

For decades, physicists ran ever more expensive simulations to test theories about dark energy, neutrino mass, and possible tweaks to gravity. Now a team has shown that artificial intelligence can absorb the standard model of cosmology in a fraction of the usual time. But there's a catch: once trained on the familiar, the AI sometimes refuses to recognize anything else.

In a paper published this month in the Journal of Cosmology and Astroparticle Physics, Princeton undergraduate Veena Krishnaraj and colleagues Adrian Bayer, Christian Kragh Jespersen, and Peter Melchior used transfer learning to train neural networks. The networks first studied cheap simulations of the ΛCDM model - the standard cosmology. Only afterward did the researchers expose them to data from theories that go beyond it.

Transfer learning cut the number of costly beyond-standard-model runs by more than ten times. That matters. Upcoming surveys from the Vera C. Rubin Observatory and Euclid will flood researchers with data, and anything that reduces computational costs looks attractive. Yet the networks carried forward biases. They performed well when new physics sat far from the standard parameter space but stumbled when new signals overlapped with directions already explored inside ΛCDM.

"The negative transfer is not random," Krishnaraj told Phys.org. "It is driven by underlying physical degeneracies in the model." Different combinations of parameters can produce nearly identical observable effects on the cosmic web. The AI latched onto those associations during pre-training. Later it treated slight deviations as noise instead of evidence for something new.

Degeneracies have long plagued cosmology. Neutrino mass and the amplitude of matter fluctuations trade off against each other in ways that look similar in large-scale structure data. The standard model already encodes these trade-offs. Pre-trained networks simply amplified them. In tests, the AI grew overconfident, down-weighting genuine signals that sat inside the degenerate regions.

Co-author Francisco Villaescusa-Navarro, of the Flatiron Institute, helped frame the broader stakes. Transfer learning can accelerate discovery, he noted, but without safeguards the tools may steer the field away from the anomalies they were built to find. Some degeneracies are physical. Others may be baked into the network weights themselves. Distinguishing the two will demand new techniques - perhaps networks that actively unlearn or ensembles trained on deliberately contradictory priors.

The paper stops short of prescribing fixes, instead mapping the terrain. Transfer learning works. It also fails in predictable ways tied to the physics. That predictability offers hope. Physicists can design around it. They cannot afford to ignore it.

Source: Webpronews