Google says artificial intelligence has helped discover “millions of new materials,” but scientists disagree

Google’s DeepMind published a blog post titled “Millions of new materials discovered through deep learning.” But researchers have analyzed the discoveries and found no strikingly new compounds in Google’s research, 404 media reports.

Google writes that its GNoME artificial intelligence tool has found 2.2 million new crystals, 380,000 of which are stable materials that could form the basis of future technologies. The company also boasted that this is equivalent to almost 800 years of knowledge.

The corresponding article was published in the journal Nature and received a very wide response in the press as an example of the incredible prospects of artificial intelligence in science.

Researchers at Lawrence Berkeley National Laboratory have also partnered with Google DeepMind to create the Autonomous Laboratory (A-Lab), which uses “computing, machine learning, and active learning to plan and interpret the results of robotically-assisted experiments.”

In fact, the researchers used the capabilities of artificial intelligence and robots to “eliminate” humans from the lab. The researchers claim to have discovered and synthesized new materials, which “demonstrates the effectiveness of AI-driven platforms for autonomous material discovery.”

However, two independent groups of researchers analyzed the work of DeepMind and Berkeley and published their own analyses, which show that the research is grossly overestimated. Many materials science experts say that AI has great prospects in discovering new types of materials. But Google and its deep learning methods have not made an incredible breakthrough in the world of materials science.

In a paper published in the journal Chemical Materials, Anthony Cheetham and Ram Seshadri of the University of California, Santa Barbara, selected a random sample of 380,000 proposed structures released by DeepMind and argue that none of them meet the three-part test for whether a proposed material is “reliable,” “useful,” and “novel.”

The researchers believe that what DeepMind found are crystalline inorganic compounds and should be described as such, rather than using the more general term “material,” which they believe should be used for things that “demonstrate some utility.”

“We have yet to find any strikingly novel compounds in the GNoME and Stable Structure listings, although we anticipate that there must be some among the 384,870 compositions. We also note that, while many of the new compositions are trivial adaptations of known materials, the computational approach delivers credible overall compositions, which gives us confidence that the underlying approach is sound,” they write.