Do we not have a thread on AI?

well admittedly whatever models etc. they were using didn't actually emerge out of the machines they were running on, synthesise the material, build the lab experiment to test the candidates. Nor did it replace any of the 8 scientists otherwise needed to change the light bulb. I'd suggest that is probably a good thing though. 


On the other hand, 

"...used AI and supercomputers to narrow down 32 million potential inorganic materials to 18 promising candidates in less than a week - a screening process that could have taken more than two decades to carry out using traditional lab research methods.

So saying it "just" sped things up is a bit like saying the steam engine "just" sped things up, or "just" reduced the manpower required. Which in most circumstances amount to the same thing. I'd suggest the 'gain' is a bit more than just a bit of a speed up. 

What is not said in there is how these 'potential' materials were known/discovered. The article doesn't go into any details. But if I was to hazard a wild guess, I'm thinking it is fairly likely that the AI contribution here was something like:

- train the model (or get it to teach itself) to understand what can make a good battery material
- train the model (or get it to teach itself) about molecular structures/shapes/chemical behaviours to understand what would happen if you doped this with that/combined the other with this
- get it to postulate and synthesise new materials (model wise) based on this knowledge
- and a shed load more that I'm even less qualified to postulate and prattle on about
- analyse the resulting materials to determine whether probablistically they were good/viable candidates or not

In other words it created novel materials based on knowledge gained, and "tested" whether they were likely good candidates or not. To get a short list to be presented to real world actors 

So sure, it hasn't lit any physical bunsen burners. But if the 'gain' is a couple of decades compressed into (less than) a week, that is really not much different from saying it has done something which may have have otherwise been done. And this might be a dud of course. But hell, there have been plenty of those over the years. Fail fast, just a shed load faster, learn and improve.