We have algorithms that can dive into mountains of personal data in order to associate personalities with extremely well-targeted advertisements. How about algorithms that would dive into mountains of data about the cosmos to find hidden patterns?
The big difficulty, summarizes the New Scientist, is that in theoretical physics, a “hidden trend” is called an equation, and it is not easy to teach this “language” to an artificial intelligence. Admittedly, in the 16th century, Johannes Kepler was able to write his own mathematical equations on the movements of the planets, because he had immersed himself in the numerous data of astronomical observations of his time. But today, the field of analysis and the amount of data are infinitely broader, and continue to grow with each new observation from the James-Webb or Gaia space telescopes. The more variables there are, the more the possible number of equations multiplies: even an artificial intelligence would not know where to start.
One of the tracks is to go there piece by piece: a few years ago, a team from Princeton University “fed” an algorithm with the movements of the moons around Jupiter and Saturn, and waited for it to “discover” the laws of gravity. Since then, other researchers have been trying to take it to the next level: their algorithms have succeeded in deduce the mass of black holes from data on gravitational waves, for example. Could an even “higher” target be dark matter, which permeates the cosmos but cannot be seen or detected? If it should turn out that its distribution is linked to systems which are observable, we would have a key for a mathematical equation that an artificial intelligence could deduce, even if we are talking about a phenomenon on a cosmic scale.
But even if we get there, there will remain a problem, which is summed up by the New Scientist : all these successes of the last few years are “empirical equations. In other words, they are descriptive and adequate to reproduce experimental data”, but not to “offer a theoretical explanation”. Just as it happened four centuries ago: Kepler had discovered the laws of planetary motion, which means he could describe “how it works”; but it would be necessary to wait for Newton to understand the “why”, that is to say a law of gravity which united all these celestial bodies.
Will an artificial intelligence ever make it that far? Impossible to say, but theoretical physicists would like to be offered this helping hand.