What AI can bring to materials science

The dazzling progress of what is commonly called Artificial Intelligence today leads to deep fears, the fear of a dystopian future. We fear a double movement of progressive lobotomy of the human being who has become passive by dint of abdicating and concomitantly his dependence or even his erasure in favor of an artificial intelligence that has become sovereign.

These fears are legitimate and well-founded, accentuated moreover by the ambient discourse of injunction to progress (and therefore to AI) under penalty of disappearance. But this healthy distrust must nevertheless not obscure the real promises of AI in certain circumscribed fields, such as for example medical research or innovations in materials science.

I had the chance to visit the Berkeley laboratories (Materials Project) twelve years ago. The researchers there are said to be developing new computer tools that would be capable of scanning millions of data on the characteristics of materials. Then, from this encyclopedic knowledge, to propose avenues for developing new materials and alloys, following recipes established by these digital alchemists.

Since that time, many projects, firstly those by Microsoft (MatterGen) and Google (DeepMind), have developed artificial intelligence applications capable of creating materials that have never been imagined or tested. Thanks to the massive integration and analysis of data on chemical compositions, characteristics, measurements, and reactive principles, the laws of physics and chemistry become a large virtual playground and experimentation field. Millions of combinations are quickly drawn up and compared before finding the few rare pearls that correspond to the prerequisites. The software considers both the different possible marriages of multiple chemical compounds and their structuring and optimal architecture. These revolutionary AIs can even predict the qualities and mechanical characteristics of these assemblies that are envisaged but not yet produced.

In short, a researcher's dream, because until now the process of creating or improving materials was long and tedious, made up of multiple and often blind tests, complex studies and analyses, failures, dead ends, sometimes fortuitous advances. Change of logic and change of scale with the advent of AI, the entire process is swept exhaustively and with dazzling speed. This allows, from a specification of different desired characteristics, to favor two or three priority avenues because they are very promising and which simply require testing before being validated. We are then witnessing an unprecedented acceleration of the discovery process, even a paradigm shift, because we are moving through this means from groping in the depths of a dark cave to purposeful and informed experiments. It is becoming much easier, faster and less expensive to develop custom technical ceramics with extraordinary capabilities, or polymers that integrate enzymes and bacteria judiciously incorporated within them to trigger rapid, total and non-polluting biodegradation of plastic at the right time.

Imagine, for example, a metal alloy developed from these AI technologies, which, following draconian specifications, would make it possible to manufacture batteries that recharge in just a few minutes, thus allowing a drastic reduction in their weight and volume, which would only be composed of common and easily available materials, without critical metals, and which finally, while we're at it, would be non-polluting and recyclable... enough to revolutionize the transport sector, with more spacious, lighter, more durable and reliable vehicles, fewer in number because they are shared. The repercussions can then be significant and go far beyond the narrow framework of innovation in materials and techniques, having major impacts on the economic, environmental, or political levels: by a happy side effect, the emergence of such a disruptive innovation in the battery sector would make it possible to make obsolete one of D. Trump's latest cynical whims, namely to get his hands on Ukraine's precious mineral resources by taking lamentable advantage of the war that this country is currently waging, since de facto the "strategic resources" hoard would lose its value.

This simple projection exercise is just one example among thousands of equally promising research avenues; it's not forbidden to dream yet, right?

Quentin Hirsinger (matériO')

To find our newsletter #152 in its entirety, click here(link is external)