Yes, but No

Well this is disappointing.

Back in November Google Deepmind claimed to have an AI/ML tool that discovered over 2 million crystals including 380k stable materials ‘that could power future technologies’

I was not fond of Materials Science (I blame the lecturer, yawn) but I thought this was very cool and a clever use of the technology. This is the kind of thing we should be using it for etc, etc.

But then Materials Researchers actually looked at it:

“…unfortunately finding scant evidence for compounds that fulfill 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.”

Which sounds like a very polite way of saying “An expert should have checked this”.

The tool did follow rules to discover new things - unfortunately making mistakes like including compounds that are only available in minute quantities and the rarest occasions on our planet.

It’s the problem to be solved - just because this technology can very quickly generate answers, and it seems impressive - doesn’t mean it’s correct. That’s an issue across a lot of domains.