Recently, Amazon announced a multi-year partnership with Orbital Materials to use their ‘proprietary AI platform’ to develop new materials that help decarbonise their data centres. Orbital is synthesising and designing advanced materials for data centre-integrated carbon removal and cooling technology for Amazon.
“Together, AWS and Orbital will evaluate the scalability and performance of these new technologies in removing carbon and increasing efficiency,” said Amazon in the report.
Orbital, like any other materials science lab using AI, said that the development of new materials has ‘traditionally been a slow process of trial and error in the lab,’ and generative AI has significantly improved the timeframe for discovering these materials.
“Orbital has achieved a 10x improvement in its material’s performance through the use of its AI platform—an order of magnitude faster than traditional development and breaking new ground in carbon removal efficacy,” the startup announced. The company also plans to deploy and test these carbon removal technologies by the end of 2025.
Orbital’s claims of a ‘radical’ improvement in the speed of material discovery are also backed by research findings from the Massachusetts Institute of Technology (MIT), suggesting we’re living in a golden age of material discovery.
The Golden Age of Materials Science?
Recent research from MIT outlines the potential of AI technologies and their assistance to scientists, specifically those working on materials science. The author revealed statistics after handing over ‘new materials discovery technology’ to 1018 scientists in an unnamed ‘R&D lab of a large US firm’.
The data showed a 44% increase in new material discovery, a 39% increase in patent filings, and a 17% increase in product prototypes that incorporated these newly discovered materials.
“Trained on the composition and characteristics of existing materials, the model generates recipes for novel compounds predicted to possess specified properties,” said Aidan Toner-Rodgers.
Interestingly, the research also revealed that 82% of scientists reported a reduced level of satisfaction with their work, as they found that leveraging AI decreased creativity and skill underutilisation.
The advent of AI revolutionising materials science began last year, when Google DeepMind released research titled ‘Scaling deep learning for material discovery’, where they discovered 2.2 million new crystals, equivalent to 800 years of work of knowledge. Out of which, 380,000 materials were stable materials.
“For example, 52,000 new layered compounds similar to graphene have the potential to revolutionise electronics with the development of superconductors,” said DeepMind in the announcement.
Google DeepMind went beyond discovery and also published another research stating that 41 out of 58 selected materials were successfully synthesised in just 17 days.
Meta marked its foray into material size by releasing a massive data set called Open Materials 2024, or OMat24. The dataset contains over 118 million examples of material simulations and structures, focusing on a wide range of inorganic bulk materials to improve AI-enabled material discovery.
New Materials, Really?
Deep learning has amplified materials research and discovery for a few years now. However, if there’s one way AI can save humanity, it lies in not just discovering new materials, but also finding ways to synthesise and manufacture them in commercial capabilities.
When Google Deepmind made the announcement last year, the company did not take long for contradictory results to come into the picture, where a researcher said that many of the materials discovered weren’t truly novel and were just minor variations of known compounds. It also said that Google Deepmind mostly focused on the stability of the materials, overlooking functionalities that are required for practical applications.
Moreover, Google’s claim of synthesising these newly discovered materials using an autonomous lab is also facing scrutiny, thanks to a study from Princeton University. The researchers here outright deny Google Deepmind’s claim. They said that none of the materials produced by the autonomous lab were new and that the majority were misclassified.
The authors also say that a majority of the synthetic products were wrongly characterised. It failed to distinguish between materials with random and organised atomic arrangements. In short, the authors said that AI can generate predictions but will struggle to interpret experimental results accurately.
When Do I Use It
Presuming these challenges are overcome in discovering and synthesising new materials, but when can we hold these new materials?
It is indeed a very long journey. Benjamin Reinhardt, CEO at Speculative Technologies, puts it best in his work: Getting materials out of the Lab.
“You can think of lab-scale materials as the most artisanal products in the world, painstakingly handcrafted by people with advanced degrees. Like any artisanal product, lab-scale materials are expensive,” he said.
“Trying to mass-produce these materials by simply increasing the number of fume hoods, test tubes, and pipette wielders would make them cost billions of dollars per kilogram.”
Mass-producing or scaling these newly discovered materials remains a challenge. From analysing the materials’ failure modes in controlled conditions to setting up an industrial-scale factory, following regulations, and justifying the choices to investors and customers, one is surely set to turn into Sisyphus.
Scientists, mostly working in universities, have strong incentives to focus on novelty and one-off demonstrations because these can lead to publications and positive media attention.
“Start-ups commercialising a new material face many headwinds, including often-unavoidable relationships with larger companies, trade-offs between speed and profit margins, and capital constraints,” said Reinhardt.
Moreover, newly discovered materials need to go through a lot of testing, and the world has seen quite a few tragedies when they’re put out in the real world. For instance, carbon fibre, and composite materials are all the rage. We don’t need to look far behind in the history books. In the recent OceanGate’s tragic implosion, the Titan submersible was using an ‘experimental carbon fibre composite hull’. It was the first instance of the material being used in a submersible.
An investigation revealed that the hull may have been prone to micro buckling and delamination under deepsea pressures. Small manufacturing imperfections in carbon-fiber material could amplify such issues.
In 2017, the Grenfell Tower fire in London, claimed 72 lives owing to the building’s exterior cladding. It was made of aluminium composite panel panels with a polyethylene core and flammable insulation materials. The combination led to a rapid spread of fire, and investigations revealed that manufacturers exploited loopholes in regulation.
Therefore, there will be an increased sense of caution before these materials come to life. In the essay, Reinhardt also provided a dataset that visualises the timeline from when materials were discovered and fully commercialised.
The numbers speak for themselves. The next time it seems exciting when AI has discovered new materials, it’s time to give ourselves a reality check. There’s no doubt AI has the potential to speed up the manufacturing process as well, but we’re yet to see how.
The post AI’s Material ‘Breakthroughs’ are Still Useless appeared first on Analytics India Magazine.
