Will AI replace chemists? 

Author

  • Michael Parker, PhD

The short answer to this is no. Artificial intelligence can’t replace the insights and instincts that are unique to chemists and their expertise, but it will continue to be an increasingly valuable tool that teams can and should use to their advantage. 

I realise that there is an ever-growing library of articles that follow the form “will AI replace X”,  However, we have seen genuine concern in recent years about the increasing prevalence of AI tools in chemistry, and what that means for people’s careers, so I’m going to attempt a reasonably grown-up blog on the subject. 

ChatGPT can be great for the basics, but cannot replace expert human knowledge  

I’m going to quickly discount the most trivial idea, of replacing chemists with an LLM like ChatGPT. These models are great if the output you want is readily available in large quantities on the internet, for example clickbait articles (ironically, we’ve seen that the job of writing clickbait articles about AI has already been replaced with AI). Also, if you’re looking for instructions for how to boil an egg or wanting to generate trivial code, then these platforms can save you time.  

Importantly, you should only use these if you don’t care about the output being confidently wrong – see the example below. 

Fig 1. Google Gemini explaining how an LLM might incorrectly state the wrong number of Rs in strawberry, while also getting it wrong.  

If what you need requires specific, expert knowledge, and you care about the answers being correct, or you need some level of accountability, then even the best-performing LLMs are verging on useless unless accompanied by expert human supervision.  

The danger of misunderstanding AI capabilities 

Unfortunately, it’s easy for non-experts to get a false impression of how useful LLMs are for many tasks, because the cracks in their performance only start to show at the expert level. For example, I’m terrible at writing html. I’ve never bothered to learn it, and if I have to make a website, I’m happy to use AI for that. This works, because it’s a basic task that I am bad at. I could incorrectly conclude from this that web developers can be easily replaced with AI, a mistake that would only become apparent when an expert level problem occurs, and neither I nor the AI can solve it. Similarly, LLMs are great at doing high school chemistry homework, but trying to replace actual experts will fail embarrassingly in short order. 

The capabilities and limitations of advanced AI in chemistry 

What about more chemistry specific applications of AI/ML? These usually involve single-purpose models that have a specific task, such as property prediction, compound generation, or synthesis prediction, and they can dramatically outperform human beings at these specific tasks. However, in much the same way, a calculator can outperform mathematicians at long division while still lacking any conception of why we might want to divide long numbers, which numbers we might want to divide, and so on.  

Inspyra™ and Cerella: Case studies in AI and chemist collaboration 

Looking at a more concrete example, the next version of our Inspyra™ software is powered by a new generative AI model in combination with a custom ML algorithm, and is designed to learn what molecular properties the user is interested in and then generate compounds matching those goals. In principle, you could try and set up something like this to come up with compound ideas without needing chemists, but inevitably, the AI system would overlook something important outside the scope of the parameters it was trained on, but which a human expert would immediately spot. The net result of this is generation of a huge number of compounds that then need an expert chemist to sift through afterwards.  

Inspyra is designed to work in tandem with an expert chemist, harnessing the power of generative AI and the deep understanding of a human to get the best of both worlds. We use feedback from the chemist to iteratively improve the model’s understanding of the project goals, and that information is then fed into the generative AI to generate new, optimised compounds from the starting points identified by the chemist. This means that we can identify new candidate molecules faster and cheaper than either a human or an AI could on their own. To learn more, watch our webinar on generative chemistry.

A similar issue arises with our Cerella software, which, among other things, can recommend the optimal measurements to add to a dataset. This could be interpreted as taking a human’s role and automating it, but in reality, it’s more like giving chemists additional tools with which to make informed decisions. The AI can tell you that measuring a specific compound will most improve its ability to make predictions or that a specific assay can be used as a proxy for a more expensive measurement. But in the real world, that has to be considered alongside a host of other factors, such as how easy a compound is to synthesise, the cost of alternative assays, or the desirability of the molecular structure. Learn more about Cerella AI, read real-life case studies here.

The future of chemistry: AI as an empowerment tool 

My general conclusion is that while it might, in principle, be possible to replace chemists (or software developers, for that matter) with AI models, any company that did so would be dramatically outperformed by one that gave their chemists access to powerful AI tools. At its best, AI software can be used to empower users to solve problems faster and more cost-effectively while still retaining the experience, strategic understanding, and creativity of human beings.  

About the author

Michael Parker, PhD

Michael is a Principal AI Scientist at Optibrium, applying advanced AI techniques to accelerate drug discovery and improve decision-making. With a Ph.D. in Astronomy and Astrophysics from the University of Cambridge, he brings a data-driven approach to solving complex scientific challenges. Michael is also a thought leader, contributing to discussions on the impact of AI in pharmaceutical research.

Linkedin

Dr Michael parker, Principal AI Scientist, Optibrium

More AI resources