Chemistry deals with that most fundamental subject: matter. New drugs, materials and batteries all depend on our ability to make new molecules. But discovery of new substances is slow, expensive and fragile. Each molecule is treated as a bespoke craft project. If a synthesis works in one lab, it often fails in another.

The problem is that any single molecule could have an almost infinite number of routes to creation. These routes are published as static text, stripped of the context, timing and error correction that made them work in the first place. So while chemistry is often presented as one of the most advanced sciences, its day-to-day practice remains surprisingly manual.

For centuries prior to the emergence of modern chemistry, alchemists worked by hand, mixing substances, adjusting conditions by feel, passing knowledge from teacher to student while keeping many secrets. Today’s chemists use far more analytical tools, yet the core workflow has barely changed.

We still design molecules manually using the rules of chemistry, then ask highly trained humans to translate these ideas into reality in the laboratory, step by step, reaction by reaction.

At the same time, we are living through an explosion in artificial intelligence and robotics – and chemists have rushed to apply these tools to molecular discovery. AI systems can propose millions of candidate molecules, rate and optimise them, and even suggest reaction pathways.

But frustratingly, these tools frequently hallucinate........

© The Conversation