New AI algorithms are 95% better at showing how the universe changes over time

A newly developed technique could teach AI algorithms to see the universe with unprecedented clarity — potentially exposing the cracks in our understanding of the cosmos.

Our cosmic rulebook, known as the standard cosmological model, has done an unparalleled job of describing the universe, accounting for everything from its accelerating expansion to galaxy formation. But even the best explanations need robust, independent checks, and that's where genetic algorithms come in.

These clever techniques, inspired by natural selection, let us reconstruct cosmic functions directly from data, without forcing them into preconceived models. They're like tireless explorers, always seeking the best results.

But even the most powerful tools have their blind spots. For genetic algorithms, that blind spot has always been about seeing subtle changes in the cosmos. The overall picture might look good, but the derivatives ‪—‬ crucial measurements of how quickly things are changing ‪—‬ get wobbly.

For traditional genetic algorithms, these insights into rates of change are incredibly fragile. The "best-fit" function, which nails the observed data, often struggles with nonobservable quantities involving these derivatives. It's like trying to navigate a dense fog; the algorithm risks getting stuck in a deceptive patch that seems like the perfect solution but isn't the real deal.

So how do we equip our cosmic explorers with sharper vision?In a paper published to the preprint server arXiv in February, researchers propose an answer: a new strategy to teach our algorithms to see the universe with unprecedented clarity.

Known as GAME (short for "Genetic Algorithms with Marginalised Ensembles), this ingenious update doesn't rely on a single algorithm. Instead, it unleashes a whole squad. Imagine a council of cosmic detectives, each tackling the puzzle slightly differently. Then, GAME applies ensemble averaging, taking a weighted average of their solutions. Each algorithm's answer gets a weight based on its data fit and the smoothness of the resulting function.

The results are nothing short of spectacular. For reconstructing a test function, GAME showed a solid 20% improvement in overall accuracy. But here's the real punch: For those elusive derivatives, GAME delivered a jaw-dropping 95% improvement in accuracy. That's like trading blurry binoculars for the sharpest cosmic lens imaginable, especially for watching the universe change.

The methodology is already reconstructing the universe's expansion rate, known as the Hubble constant, using data from cosmic chronometers, which are essentially natural clocks. And the early results are perfectly compatible with our existing cosmological model. With this newfound precision, GAME is like a telescope into the future of cosmology.

Imagine what this sharper vision means for unraveling the universe's grandest puzzles. As new data from observatories like the Dark Energy Spectroscopic Instrument floods in, GAME is poised to become an even more competitive tool. It will help us discriminate between different models of how the cosmos works, allowing for clearer answers and crucial model-independent consistency tests.

While the full impact of data correlations is still a frontier, the journey has only just begun.