undefined
Meta has been accused of using faulty data to train an artificial intelligence climate tool, with scientists claiming the Big Tech group raised false hopes about the feasibility of removing carbon dioxide from the atmosphere at scale.
The tech company said last year that it had helped researchers with the elusive problem of identifying materials to efficiently remove carbon dioxide from air, by publishing a “groundbreaking” data set on which it also trained free-to-use machine learning models.
But none of the 135 materials that Meta’s research said could bind CO₂ “strongly” had that characteristic, while some did not exist, according to researchers from Heriot-Watt University and the Swiss Federal Institute of Technology in Lausanne (EPFL).
“I wish they had computed a bit less and thought a bit more,” said Berend Smit, a professor of chemical engineering at EPFL, describing some of Meta’s results as “nonsense”. “You get the impression that the Big Tech mentality is do first, think later.”
Meta said in response that its data set was based on “valid calculations useful for training machine learning models”. It added that it had “long believed in open sourcing technology . . . because that approach promotes collaboration and innovation”.
Tech companies have invested in bringing down the cost of CO₂ removal techniques — such as direct air capture (DAC) — as they prepare to offset their own carbon footprints, which have grown as they race to build generative AI tools.
Although DAC is not yet operating at scale, Microsoft is one of the sector’s biggest investors, while Meta has committed to buying credits from start-ups that are developing it.
The findings by scientists from Meta and from the Georgia Institute of Technology were published last year in a peer-reviewed journal of the American Chemical Society.
A.J. Medford, an associate professor at Georgia Tech who helped author the paper, said the critique had missed the point of the research. He said the team did not set out “to conclusively identify new materials”, but intended to experiment with more sophisticated materials screening techniques, as well as identify new challenges and questions for the field.
After performing 40mn calculations in the field of quantum mechanics, Meta’s team created a database of chemical structures known as metal-organic frameworks, which it predicted could strongly attract CO₂ while not attracting other components of air like water vapour.
The project required hundreds of times more computing than the average academic computing lab could do in a year, Meta said. The data was then used to train an “AI model that is orders of magnitude faster than existing chemistry simulations”, to identify further promising materials, it said.
Researchers who sought to reproduce the results said Meta’s team had overestimated the materials’ binding capacities to CO₂ and that its open-source AI tools were not fit for purpose. A key problem was the use of erroneously described chemical elements from an academic database that had since been updated, they argued.
Meta said the materials had been identified as only “promising and deserving of more thorough inspection”.
It added that in some cases the calculations in the data set “correspond to implausible or highly unstable structures because of their electronic configurations or otherwise” and that the authors had disclosed this.
US President Donald Trump’s slashing of subsidies for clean energy projects in the US has caused anxiety in the carbon capture sector, although tax breaks for the technology were largely protected in his tax and spending bill.
Climeworks, a frontrunner in direct air capture technology, which does not rely on the materials identified by Meta’s team, said in May that it had cut more than 100 jobs. On Wednesday, however, Climeworks said it had surpassed $1bn in equity funding, which the company said signalled investor confidence.
Wijnand Stoefs, a policy lead on carbon removals at the non-profit Carbon Market Watch, said DAC technology had been “enormously overhyped” by tech groups including Meta. The technology’s high expense and energy needs compared to replacing fossil fuels with clean energy meant its proponents faced a “deep crisis”.
Scientists nonetheless welcomed Meta’s attempts to tackle an expensive and complex research question that has gripped climate-focused chemists.
“By publishing everything openly, Meta enabled the research community to dig in and build better tools,” said Susana Garcia, professor of chemical and process engineering at Heriot-Watt University.
Additional reporting by Martha Muir in New York
undefined