Virtual Reality for Understanding Artificial-Intelligence-driven Scientific Discovery with an Application in Quantum Optics

TL;DR

This work shows how transferring part of the analysis process into an immersive virtual reality (VR) environment can assist researchers in developing an understanding of AI-generated solutions and shows the potential of VR to enhance a researcher’s ability to derive knowledge from graph-based generative AI.

Abstract

Generative Artificial Intelligence (AI) models can propose solutions to scientific problems beyond human capability. To truly make conceptual contributions, researchers need to be capable of understanding the AI-generated structures and extracting the underlying concepts and ideas. When algorithms provide little explanatory reasoning alongside the output, scientists have to reverse-engineer the fundamental insights behind proposals based solely on examples. This task can be challenging as the output is often highly complex and thus not immediately accessible to humans. In this work we show how transferring part of the analysis process into an immersive Virtual Reality (VR) environment can assist researchers in developing an understanding of AI-generated solutions. We demonstrate the usefulness of VR in finding interpretable configurations of abstract graphs, representing Quantum Optics experiments. Thereby, we can manually discover new generalizations of AI-discoveries as well as new understanding in experimental quantum optics. Furthermore, it allows us to customize the search space in an informed way - as a human-in-the-loop - to achieve significantly faster subsequent discovery iterations. As concrete examples, with this technology, we discover a new resource-efficient 3-dimensional entanglement swapping scheme, as well as a 3-dimensional 4-particle Greenberger-Horne-Zeilinger-state analyzer. Our results show the potential of VR for increasing a human researcher's ability to derive knowledge from graph-based generative AI that, which is a common abstract data representation used in diverse fields of science.

Figures (6)

• Figure 1: Virtual Reality (VR) assisted digital discovery workflow for obtaining insight from solutions discovered by Artificial Intelligence (AI). AI results are studied in VR utilizing interactive 3D visualization to discover interpretable structures. These are used to manually build new graphs or to direct the AI with smart initial geometries for more efficient searches.
• Figure 2: Optimization flow for a 2-dimensional 4-particle Greenberger-Horne-Zeilinger (GHZ) state of the PyTheus-algorithm ruiz-gonzalez_digital_2023. Starting from a fully connected initial graph, PyTheus iteratively performs continuous edge weight optimisation and subsequent pruning of unnecessary edges until a minimal graph is reached that fulfills the desired output. When translating the resulting graph to the experimental setup, every vertex represents a detector, and every edge a correlated photon-pair source emitting towards the connected detectors. The terms contributing to the final quantum state of the experiment are conditioned on coincidence detection of all detectors. In the graph picture, these terms correspond to perfect matchings, subgraphs where each vertex is only reached by a single edge (see equation \ref{['eq:pm_decomp']}).
• Figure 3: Point of view of an exemplary PyTheus graph analysis inside the environment of AriadneVR. A PyTheus graph is shown with one perfect matching subgraph on the right. A rendered version of this graph is depicted in Fig.\ref{['fig:useOf3D']}(a).
• Figure 4: Through interactive 3D visualization structure is easily revealed in complex graphs. (a) I: 2D representation of a 3-particle 4-dimensional GHZ-state analyzer and its discovered 3D geometry. (a) II: Interference loop example on the undesired $|311\rangle$ ket. https://artificial-scientist-lab.github.io/AriadneVR/4d_3p_GHZ-analyzer. (b) I: 3-particle 5-dimensional GHZ-state in 2D and 3D. (b) II: 4-particle 4-dimensional GHZ state in 2D and 3D.https://artificial-scientist-lab.github.io/AriadneVR/GHZ-358. (b) III: Common core subgraph highlighted in both previous graphs signifying the nature of both graphs as generalizations of the $|GHZ\rangle_4^3$ state. https://artificial-scientist-lab.github.io/AriadneVR/GHZ-448 Ancillae are drawn as squares/cubes, detectors as circles/spheres, and input modes as triangles/tetrahedrons. Colors represent photon modes, indicators on edges their negative weight.
• Figure 5: Manual discovery of new graphs assisted by VR for efficient high dimensional entanglement swapping. PyTheus discovered graphs are watermarked. Through experimentation with graph edits, concepts can be easily transferred and tested on new graphs.(a) Discovered extension mechanism for higher dimensions extracted from efficient PyTheus discovered 2-dimensional entanglement swapping, represented by the graph for three 2-dimensional pairs and a single 3-dimensional pair in the center. https://artificial-scientist-lab.github.io/AriadneVR/trid(b) PyTheus discovered solution for 2-pair 3-dimensional swapping. https://artificial-scientist-lab.github.io/AriadneVR/2p_3d_ES(c) Application of the mechanism from (a) to a subgraph from (b) to obtain a new swapping graph for 4-pair 3-dimensional swapping. https://artificial-scientist-lab.github.io/AriadneVR/4p_3d_ES