Image: © vchalup/Stock.adobe.com
The study claims there is ‘scant evidence’ the DeepMind AI found compounds that ‘fulfil the trifecta of novelty, credibility and utility’, but DeepMind stands by its claims.
A new study analysing a recent claim from Google-owned DeepMind suggests there is little evidence the company managed to find novel materials using AI.
The company claimed last November that one of its AI models found 2.2m new crystals that could potentially be used to create new materials. DeepMind said 380,000 of these crystals were “stable” and are the best candidates for creating new materials that could boost various forms of technology.
The company also said it would contribute the 380,000 materials it predicts to be stable to the Materials Project, an open-access database that aims to support the creation of new materials.
But a group of researchers have taken issue with some of the claims made by DeepMind and claim they found “scant evidence” for compounds that “fulfil the trifecta of novelty, credibility, and utility”.
“While the methods adopted in this work appear to hold promise, there is clearly a great need to incorporate domain expertise in materials synthesis and crystallography,” the researchers said.
DeepMind did not respond to a request for comment, but a spokesperson told The Register that the company stands by “all claims” made in its Gnome paper.
In the study, Prof Anthony Cheetham and Ram Seshadri of UC Santa Barbara claim the DeepMind AI’s finding consist “solely of crystalline inorganic compounds” and argue that they should be labelled as such, instead of using “the more generic label ‘material’”.
“Polymers, glasses, metal-organic frameworks, heterostructures and composites are only a few excluded materials classes that come to mind and that are each infinite in their scope,” the researchers said.
The study looked at the first 250 entries of the Gnome Explorer database and claims many of the entries are based on the “ordering of metal ions that are unlikely to be ordered in the real world”.
The researchers also criticised the way the results were organised, describing them as a “seemingly random walk through the periodic table”.
“While we are confident that the tools of artificial intelligence and machine learning have a bright future in the field of materials discovery, more work needs to be done before that promise is fulfilled,” the researchers said.
Find out how emerging tech trends are transforming tomorrow with our new podcast, Future Human: The Series. Listen now on Spotify, on Apple or wherever you get your podcasts.
