Meta is accused of using inaccurate data in a climatic program that was presented as a “revolutionary”

The Meta was accused by several researchers of having used erroneous data to “train” an artificial intelligence designed to combat climatic changes, the database in that “tool” referred to possible methods of widely eliminating carbon dioxide from the atmosphere and contained a list of useful materials.

The company had announced in 2024 that it has helped researchers approach the difficult problem of identifying materials capable of effectively extracting carbon dioxide, publishing a data set as “revolutionary”, a set also for training for free “Machine Learning” models, writes Financial Times.

However, according to researchers at Heriot-Watt University and the Federal Technology Institute in Lausanne (EPFL), none of the 135 materials that Meta had presented in research had the ability to attract carbon dioxide, and some materials in the list did not even exist.

In reply, Meta said that her data set was based on “valid, useful calculations for training AI”. The company added that it has always been the Open-Source adept, because this approach “stimulates collaboration and innovation.”

Tech companies have invested a lot in reducing the costs of carbon dioxide capture technologies, such as direct capture from the air, to compensate for their own carbon emissions, emissions that have increased as the generative tools quickly developed.

The results of Meta research, carried out with the Georgia Institute of Technology, were published last year in a specialized magazine of the American Chemical Society.

Meta said that the project required hundreds of times more computing power than it has, in a year, an ordinary academic laboratory. The data set was used to train a “model of the order of tens or hundreds times faster than existing chemical simulations”, intended to identify other promising materials.

But the researchers who have tried to reproduce the results claim that the Meta team has overestimated the ability of the materials to attract carbon dioxide and that the open-source tools are not suitable for the proposed purpose. One of the big problems was the use of incorrectly described chemical elements in an academic database, a basis that has been updated in the meantime.

Meta replied that the respective materials had been only “identified as promising, requiring additional checks”. He also added that, in some cases, the calculations targeted “unlikely or extremely unstable structures, due to their electronic configurations or for other reasons.”

Photo source: dreamstime.com