AI - Climate Destroyer or Saviour?

📌 Introduction

Since the dawn of artificial intelligence and the publication of Google’s famous “All you need is attention” paper, the rise of AI throughout every corner of industry has been almost instantaneous. From terrible AI-generated poems to protein structure determination by AlphaFold 2.2, AI seems set to replace dogs as man’s best friend.

In 2024, Nobel Prizes in both Physics and Chemistry were awarded for significant research related to AI. During his acceptance speech, “the father of AI” Geoffrey Hinton issued a grave warning regarding the potential dangers of artificial intelligence.

Chemical research and material design are no exception to this paradigm shift — from Scooby-Doo to C3PO. The question that these rapidly evolving developments bring to my mind, and to those of my fellow sceptic researchers, is:

“Can this computer really do what it would take me a week to accomplish in a matter of hours?”

In short, the evidence seems to point towards yes, it can. However, as science and technology are so very seldom black and white, the real answer is slightly longer.

🏗️ Metal-Organic Frameworks (MOFs)

In today’s article, I will attempt to discuss possible answers to the above question, highlighting some of the complications and benefits that using AI for metal-organic framework (MOF) design presents.

As I’ve covered in my first article at The Periodic Pulse, MOFs are a favourite of mine. These highly porous framework materials can be designed and applied in a variety of important areas such as catalysis, energy storage, and carbon capture. The incorporation of AI into the design process of such materials has shown promise of accelerating the development of new MOF-centred technologies.

My own infatuation with the MOF concept is largely due to the many, many (many) different building blocks from which they can be designed.

While this is one of the key aspects that interests me so, it also poses a significant issue when it comes to sitting down and designing these things. Out of all of the ever-expanding shelf of metal nodes and their connecting bridges, where do we start? How do we select materials best suited to purpose?

🤝 AI Meets MOFs

To help shed some light on potential routes through this vast bazaar of viable linkers and metals, researchers have explored the use of machine learning (ML) to predict the performance and viability of specific MOF structures. This is done by simulating how molecules will fit inside the MOF pores and the effects this will have on the framework structure. As the conceptual design of MOFs is constantly directed towards specific applications, it is imperative that researchers explore effective ways of selecting optimal materials for a chosen purpose.

Preliminary applications of AI in MOF design have generally been in the automation of literature reviews and extracting data to streamline the research process.

More recent studies have looked at using machine learning to direct and run computational simulations of conceptual MOF structures in order to gain a closer look into material properties at the molecular level. These “High-throughput Computational Screening” (HTCS) studies, although useful, are slow and expensive, taking anywhere between 1–7 days, with even longer timescales for the generation of framework models by AI.

Therefore, instead of simulating every possible material combination, a cooperative approach has been adopted where ML simulations analyse specific MOF databases to establish key structural motifs associated with optimal properties for a given application.

📊 Industry Outlook

Aside from all this conceptual, blue-sky fugazi, the question still remains: to what end is AI contributing to the use of real-life MOFs?

Market research suggests that the niche sub-market of AI-assisted MOFs is expected to rise from an estimated 1.2 billion USD in 2024 to 3.5 billion by 2033, with a CAGR of 12.3%. Although these numbers are approximate and may include other sub-branches of MOF research, given how scarce and blurred market research can be when it comes to bleeding-edge technology they still paint a picture of rapid growth.

When answering the question of AI-MOF relevance, one application stands out among the rest, the ironically environmentally friendly field of CO₂ capture has been the focus of most initial AI studies.

🌱 AI, MOFs & The Climate Crisis

In the age of climate crisis, the field of carbon capture has experienced significant growth in recent years (a diplomatic, professional way of saying a mad scramble to save our planet), and MOF research has been at the forefront. Promising examples of these developmental technologies are being put to use by companies across the globe, such as BASF in Canada.

However, trial-and-error testing, slow kinetics, and traditional high-throughput techniques make the development of these carbon-capturing, world-saving MOFs tedious and not fast enough to save a small village, much less the planet.

Therefore to help bypass these limitations, the integration of machine learning simulations for the selection of MOF materials may provide the apocalypse-dodging solutions that research requires. Only time will tell.

🏁 Final Take

Although the robotic march of artificial intelligence continues to spark that primordial part of my brain that warns of Skynet or HAL, even I must admit that yes — until those bay doors refuse to open — there is a place for AI in the lab.

Thank you for reading, goodbye for now.
Dylan Joseph Shaw

💡 P.S.

For the astute and judgemental among you: I am aware that I’ve used the terms AI and machine learning interchangeably throughout this article, and that these are strictly speaking not the same thing.

To you I would like to highlight that me and my luddite ways are still amazed at the fact that these AIs are drawing strange cartoons, much less doing chemistry. Have patience.