SCI: AI and the art of the possible

BY JEREMY FREY

Artificial intelligence (AI) has the potential to facilitate radical change in everything, from fundamental research and application development, through to supply chains.

However, for those not yet engaged with the AI revolution, what does it have to offer and what do they need to do to ensure that they can sensibly join in?

SCI’s Data Digitalisation and AI group recently hosted a webinar entitled ‘AI -The Art of the Possible’ ahead of a conference to be held at SCI on 8-9 September 2026.

The aim of the webinar and conference is to provide examples of how AI can make a difference to productivity and efficiency and profitability for companies, small and large, operating in the chemicals sector.

While, according to a recent Gartner report we are already in a ‘trough of disillusionment’ with AI and there is a lot of talk of AI bubbles bursting, I am much more optimistic.

There may need to be some re-assessment, but AI has huge potential and offers much to industry. Historically there have been several periods of AI hype followed by AI winters (at least for AI researchers). AI and machine learning (ML) are much more than ChatGPT, and Large Language Models (LLMs) in general.

Researchers may well already be using some of the basic ideas of AI. After all, chemists have been involved with very similar technologies to ML since the early 1960s, mainly in the drug discovery area, and toxicology predictions with the ideas of QSAR and Design of Experiments (DoE), with some of the issues formalised in the Organisation for Economic Collaboration and Development (OECD) rules.

The evolution of AI from expert systems to LLMs, which derive from very different fundamental approaches to ML, has generated huge public and government interest. The UK Government, along with many other countries, is counting on AI to promote increased economic productivity and growth.

One such pathway is through enhanced fundamental scientific and engineering research and more rapid and innovative development and commercialisation of all aspects of the research and product supply chain. The UK investment in AI compute facilities is one obvious outcome.

The key question for many if not most companies - and indeed academic research groups - is what do they need to do to be ready to apply AI successfully? Understanding what can achieved and matching this to the appropriate AI and ML techniques is essential.

In my view data is the key underlying aspect. AI models need quality data on which to work. Collecting up historic data and making it available in digital form is essential. This will often initiate a review and rationalisation of processes and procedures.

The heavy lifting to get all the company data organised is the first area that AI, in the form of LLMs, can help with alongside more traditional programming. LLMs are versatile and adaptive feature extraction engines which are steadily improving as the community trains them with more chemical knowledge.

Current ML approaches are ravenous data consumers, and the quality of the data is critical. Missing data and or bad data can lead to terrible bias. Once a company has started on the AI track, then the need for more data can become the dominant point on the critical path. Successful uses of AI will often go hand-in-hand with automation and increased throughput of synthesis and manufacture.

In the R&D arena, this also has consequences for analytical and characterisation techniques for which there is a need to work with smaller and smaller amounts of materials and develop good proxies for the ultimate product function.

Just as Artificial General Intelligence (AGI) is a long way off, the idea of a fully self-driving autonomous lab, that develops its own hypothesis and then carries out relevant experiments, is still in my view a distant dream (or nightmare). But systems to augment R&D with human-in-the-loop systems will lead to increased efficiency and very likely more interesting jobs. The re-enforcement learning with humans in the loop is one way to tap the tacit knowledge of experts.

A recent meeting of the AIChemy Hub highlighted that in the materials discovery area, the growing strength of the LLMs enable chemical reasoning to be included in generating experiments. Interestingly this seems to lead to more scatter in the resulting experiment but a better final outcome, due perhaps to more adventurous non-chemical mathematically-driven approaches.

Watch the webinar and join us for the conference to understand how your company could benefit from AI. Explore the stages of planning needed, what data needs to be gathered, what processes need to be aligned, and what testing needs to be done to exploit AI successfully.

Jeremy Frey is a professor of physical chemistry at the University of Southampton, UK, and chair of SCI’s Data Digitalisation and AI group. Watch the webinar on YouTube.