The Fermi-Hubbard game is a pattern-matching game, but in this case, the patterns are generated by simulations of fermionic atoms in an optical lattice [1]. Researchers are still trying to understand the underlying physics that governs how these atoms behave, and this has important consequences for quantum-based technologies like high-temperature superconductivity.

Players can interact with a variety of different game modes that describe different physical scenarios. Players do not need any prior knowledge of physics, as the game has a training mode designed to guide them in image classification. During or after gameplay, players are given the option to submit feedback outlining their strategies. This classification problem has been addressed with a neural network [2], but we hope to better understand how humans learn to classify the images and their thought processes during the task.

As players learn to classify images, they can enter their insights using the “Strategies” button.

Different game interfaces

Players can undergo training in image classification.

Afterwards, they can:

(a) Classify images as one of three theories (called pi-flux, sprinkled holes or strings)
(b) Compare images from the two theories
(c) Classify images based on the atoms’ temperature
(d) Build images describing the theories based on their insights.

REFERENCES

1. C.S. Chiu et al, Science 365, 251, (2019).
2.  A. Bohrdt et al, Nat. Phys. 15, 921, (2019).