AI and hidden algorithm of quantum world

Demis Hassabis , CEO of Google DeepMind , ecently proposed an intriguing conjecture in his talk Accelerating scientific discovery with AI :

"Any pattern that can be generated in nature can be efficiently discovered and modeled by a classic learning algorithm"

This bold claim hints at profound implications for the foundations of physics. Inspired by this, I’d like to cautiously explore an idea that, while speculative, may offer an intriguing perspective.

Let’s imagine that the outcome of any quantum measurement is determined by a hidden algorithm—an incredibly complex and evolved process that we are still centuries away from discovering. This algorithm, if it exists, could be analogous to those governing biological phenomena like protein folding or strategic decision-making in games like Go.

These problems share a critical characteristic: they involve navigating an unimaginably vast combinatorial search space. The sheer number of possibilities makes brute-force computation infeasible, requiring computational resources far beyond our civilization's current capabilities. Remarkably, generative AI has demonstrated an ability to find solutions in these domains. By identifying patterns and learning from data, AI encodes within its weights and biases an approximation of the underlying algorithm with striking accuracy.

If such an algorithm exists for quantum mechanics, could these AI techniques—designed to handle such immense complexity—help uncover it?

It is important to acknowledge that most physicists, for good reasons, hold the view that quantum measurement outcomes are fundamentally random, with no underlying algorithm to determine them. This randomness is a cornerstone of quantum theory and has withstood extensive experimental scrutiny.

However, applying generative AI to this problem offers an intriguing prospect:

• If AI fails to identify any patterns or rules that narrow the range of possible outcomes, it could provide indirect evidence supporting the notion that quantum randomness is indeed fundamental.
• Conversely, if AI discovers a hidden structure—something that systematically narrows the possibilities or identifies subclasses of outcomes—we might have our first glimpse behind the quantum veil.

A Starting Point: The Double-Slit Experiment

How could we approach this idea experimentally, assuming we are sufficiently curious (and perhaps skeptical) to test it?

A good starting point might be the double-slit experiment. In this setup, individual particles (e.g., electrons or photons) are sent through two slits, gradually forming an interference pattern on a screen. We could capture images of the process as it unfolds—each particle’s arrival contributing to the evolving pattern—and use these as training data for a neural network.

The goal would be to see if the AI can make meaningful predictions about the location of the next particle on the screen.

Broadening the Dataset

Even if a hidden algorithm exists, relying solely on data from the double-slit experiment may not provide sufficient insight for AI to uncover it. Expanding the dataset to include results from other interference experiments—especially those involving multiple paths, both classical and quantum, from the source to the screen—could help. These more complex scenarios might provide the AI with the additional information it needs to identify underlying patterns.