Google Deepmind and UC Berkeley have raised doubts among the chemist group with their claim of creating new materials using robots controlled by artificial intelligence.
In an initial study published in NATURE magazine, the use of a robotic laboratory A-Lab was highlighted. This lab employed the GNOME model from Google DeepMind to automatically synthesize new connections.
The program generated millions of recipes for new inorganic crystalline compounds, which could potentially be useful in future electronics. Over a period of 17 days, the robotic device reportedly created more than 40 new materials, with 35 of them being predicted by GNOME. The robot mixed and heated various powders to create these materials, and their structures were analyzed using x-ray diffraction.
A machine learning algorithm analyzed the experimental data and compared it to predicted models to confirm the success of compound synthesis. This experiment was considered to be evidence of the potential contributions of AI-controlled robots to scientific discoveries.
However, the results are now being disputed. In a separate article published on chemrxiv, seven researchers from Princeton University and University College London argue that A-Lab failed to create any new inorganic materials.
“Unfortunately, we found that the main claim of the A-Lab article regarding the synthesis of a large number of previously unknown materials is false,” states the analysis. When studying the x-ray diffraction data of each material, the researchers discovered that most of them were misclassified.
X-ray diffraction templates are used by scientists to determine the arrangement of atoms within a material. Different materials produce distinct diffraction patterns, and scientists analyze the peaks and patterns in the data to interpret the structure of each material.
The data from the A-Lab article indicate that the majority of the 35 templates for new materials predicted by GNOME resemble mixtures of already known compounds, while three of them are not new at all. According to Robert Palgrave, a professor of inorganic and material science chemistry at University College London, these errors occurred due to an attempt to utilize AI to determine whether a new material had been created or not.