Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
AI

AI Boosts Materials Discovery By 44% at Major US Lab (nature.com) 27

Posted by msmash from the already-paying-dividends dept.
AI-powered teams at a major U.S. materials company discovered 44% more new materials and filed 39% more patents compared to teams using standard methods, according to a study by MIT economist Aidan Toner-Rodgers.

The research, conducted at an unnamed corporate laboratory with over 1,000 scientists, tracked the implementation of a custom machine-learning system combining graph neural networks with reinforcement learning. The AI tool, deployed in 2022, was pre-trained on crystal and molecular structure databases. Top-performing scientists showed the greatest gains with AI assistance, while lower-ranked researchers saw minimal benefits. The AI-designed materials demonstrated higher novelty compared to human designs, based on patent text analysis.

The company's secrecy limits independent verification of results, according to University College London chemist Robert Palgrave. Researchers using AI reported lower job satisfaction, citing reduced creative involvement in the discovery process.
This discussion has been archived. No new comments can be posted.

AI Boosts Materials Discovery By 44% at Major US Lab More

AI Boosts Materials Discovery By 44% at Major US Lab

Comments Filter:
  • If anyone has an alternative link it would be appreciated.

  • Patents are invalid (Score:5, Insightful)

    by laughingskeptic ( 1004414 ) on Monday December 09, 2024 @11:47AM (#65001051)
    Patents are required by law to disclose sufficient information to reproduce the invention. It is simple, if they haven't done this then the patent office must reject these patent applications.

    The company's secrecy limits independent verification of results

    This is the tip of the iceberg. If the patent office doesn't get its shit together we are about to all be swept away by a deluge of AI generated, never produced, non-reproducible, and conflicting patents generated by AI. There will be hundreds of patents granted that essentially might cover the same concept -- that will be the ultimate mockery of our current "purposefully vague legalese is engineering" approach to technical patents. Arguments about "priority" will be critical to whether a company's "IP" stands up in court and every "engineering" company and patent troll organization is going to go crazy generating as many patents applications as possible to cover everything they can think of ... and that an AI can think of. This is going to be a mess.

    The U.S. patent office's 8,000 examiners get roughly one man-week over years to decide if a given patent application out of 400,000 per year is worthy of becoming a patent. This is no way sufficient time to determine if most patents contain sufficient information to really define a reproducible and distinct concept. It only costs $400 to file a patent. The entire patent system is about to collapse under an avalanche of AI generated patents if they do not start doing what they should have been doing all along and requiring an actual implementation of the patent.

    • Publishing is Dead (Score:1, Informative)

      by gabebear ( 251933 )
      Agreed, but it's bigger than just patents.

      AI feeds in a compendium of information and regurgitates variations of what it was fed. Everything that relies on publishing things is being overrun by AI regurgitation.

      Intellectual property laws(copyright/patents) aren't protecting the people who generated the compendiums of information(whether you believe they should be protected or not). IP laws have existed for ~400 years with little change... I think we're going to see some big changes since they

      • Re: (Score:2)

        by Luckyo ( 1726890 )

        The opposite. If anything, we'll see relaxing of the laws to enable more AI possibilities (AI doesn't "regurgitate" by design"). Because whoever doesn't will be left behind by those that do.

        This is very much existential, because it's one of those basic science level breakthroughs. Deny it and you deny all that comes with it. And then lose to civilizations that allow it, and can benefit from it.

      • At least with patents, if the U.S. Patent Office had simply been following 35 U.S.C. 112(a) all along, patents would not be so susceptible to this craziness.

        The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated b

    • Re:Patents are invalid (Score:5, Informative)

      by timeOday ( 582209 ) on Monday December 09, 2024 @12:26PM (#65001145)
      I wouldn't assume the things they are patenting and the things they are keeping as trade secrets are the same things.

      Also, materials discovery doesn't mean "Our AI predicts this might work," and I'd be surprised if they lowered the bar of a patent that far. The goal of the algorithm is to increase the novelty and hit rate of your physical experiments, since how many things you can try will always be limited by resources. But even with a good AI the hit rate will still be far under 50%.

      In some cases the physical experiment can also be performed robotically and controlled by the AI, but it still isn't going to just pop out a patent submission automatically at the end.

      • Also the patent office will have AI of their own, it will help them search for prior art more efficiently. Traditional search finds whatever is topically related, but AI is better at narrowing down a search to functional relationships, or analogical processes that just use different words.

      • Re: (Score:2)

        by HiThere ( 15173 )

        They already HAVE lowered the bar. I believe that once upon a time you had to display a working copy of the invention.

    • You can use an AI to filter out some otherwise time-consuming branches of the search tree. But the parent doesn't care how it was discovered, only how it is made. So this is a perfectly fine use of AI.

    • The deluge of crap in this field is just an extension of the crap that we have released into the world. You won't be able trust patents any more than you can trust anything else. You won't be able trust video evidence, nor will you be able trust that video evidence is fake. Same with audio. Same with any record of any event, ever. It's the death of "proof" of any kind. It's perfect deniability, and the inability to deny anything.

      I don't know what the fuck we're going to do about it. In fact, I've decided to

  • We interrupt the crapfest... (Score:5, Insightful)

    by MpVpRb ( 1423381 ) on Monday December 09, 2024 @12:00PM (#65001073)

    ...to do useful work
    This is the promise of AI, not stupid crap like "summarizing emails"

    • Re: (Score:2, Insightful)

      by nightflameauto ( 6607976 )

      ...to do useful work This is the promise of AI, not stupid crap like "summarizing emails"

      A patent troll using AI to generate still more patents is useful work? Huh. I suppose, in the coming crapfloodpocalypse, these folks will probably be seen as "useful" in comparison to those using AI to generate crapfloods about even less reality based things.

      • Re: (Score:2, Insightful)

        by Turkinolith ( 7180598 )
        "Researchers input requirements for a material’s desired properties into the neural network, and the system suggests structures for new materials that could have those properties. The teams then weed out potential duds — such as formulae that would not lead to a stable compound — using their own specialist knowledge and computer simulations. They then attempt to synthesize the candidate structures and, if successful, test them in experiments and even in prototypes of finished products. The

        • Re: (Score:3)

          by HiThere ( 15173 )

          I think you're believing more of the PR than he is. The process one group used will not be the process used by all groups, and as there's no requirement that they "show the working invention", it can easily just be fiction. Many patents are known to be "concept inventions". (*Perhaps* those wouldn't hold up in court, but that only benefits those who can afford to use the patent court...which is a lot more costly than an regular court.)

    • Re:We interrupt the crapfest... (Score:4, Interesting)

      by Luckyo ( 1726890 ) on Monday December 09, 2024 @01:29PM (#65001335)

      Both are useful work. Enormous amount of useful work time is rendered worthless because most people can't write concise and precise emails. Myself included.

      A good collation service is great for me, because I don't have to send too verbose emails in English any more. I can just have LLM collate me a less verbose version with same information, and send that. Saves time for people reading it.

      Bonus points for it slowly teaching me how to write in a more condensed way.

    • Summarizing emails is a great use case even if it is a business function. Very early next year the powers that be for our conferencing platform is going to be integrated with an AI feature. If a call is recorded, it'll write up a summary based on not just was said but anything that was done over a screen share. The tech side, I've used it to troubleshoot some code and especially complex regex.
  • I mean if you set a script to churn out all possible combinations, eventually one will be a useful result. This is no different except with Al database relationships.

    That's fine. But patenting anything from the result of this is a slap to humanity. Patents are to protect implementations of an idea, not to protect some mashup from a database lookup. Patents are also not to protect code or language, which should make DNA un-patentable; but here we are with a broken patent system run by big Corp.
    • Try playing a game of chess against AlphaZero and then against an algorithm that chooses any valid move with equal probability. They're no different... except for the quality of the AI heuristics they use, of course. But other than that, no different.
  • Anybody, whose productivity was increased 40% by using AI was probably fucking off a lot to begin with.
  • Alphafold AI predicts the structure for folding proteins. For material microstructures, AI could use the same processes to determine optimal cohesion.
  • Stuff like this is the best usage of AI. Most scientific discoveries happen either by accident or after years of researchers meticulously trying every possible scenario using brute force methods. Folding at Home is a good example of brute force science. AI should in theory be able to find solutions to these big problems faster and more intelligently than brute force.

  • Now this is going to be great for mankind.

  • to fight over all the patents written and filed by AI

Slashdot Top Deals

A computer lets you make more mistakes faster than any other invention, with the possible exceptions of handguns and Tequilla. -- Mitch Ratcliffe

Close