A great friend and colleague would often come into the break room at work and describe the latest high-impact paper by ‘that’ professor as ‘really beautiful work’. It could be excused as hyperbolic or a touch dramatic, but they were both genuine and sincere. Being an incredibly experienced chemist, they had a deep appreciation of the work described and considered the research an important contribution to the field.
Emotional adjectives such as beautiful, elegant and inspired are seldom found in scientific literature. It is expected that a researcher will report their findings clearly and concisely, often toning down their passions into neutral adjectives such as modular or efficient. The humblebrag still communicates the same message but forgoes the emotional and fundamentally human critique of revolutionary work that can often be described as beautiful.
Other disciplines such as architecture and fashion take similarly great care in what they design but have less problem attaching emotion to their craft. Of course, these fields rely heavily on a subjective human eye to judge whether the work ‘pleases the senses or mind aesthetically’ – the definition of beauty. Further still, what pleases these audiences changes rapidly with the latest trends. We can all recall certain building styles or clothing fabrics which were all the rage, then never seen again. Is research in chemistry really any different?
Which bit pleases the audience?
Imagine seeing ‘that’ paper by ‘that’ professor for the total synthesis of ‘Destroyer-of-3-PhDs-&-4-PDRAs’. The authors report delicate and considered disconnections after multiple years of dedicated work pursuing dead ends with countless disappointments and failures. All was rewarded when the perfect crystal was isolated, confirming an intricate structure that matches the highest aspirations of the researchers. It can be cold-hearted to describe their achievement as anything other than beautiful. The question is which bit pleases the audience? Was it the elegant vision and conceptual intent of a multi-disciplinary team? The immense chemical knowledge and practical skills required to achieve such a milestone? Or is it just the physical crystal and analysed structure? If only the latter is true, does it matter how it was isolated?
What if we replaced that research team with something more autonomous? What if we programmed an artificial intelligence (AI) to determine an efficient and cunning retrosynthesis then use software to plan out the reactions for an automated group of machines to synthesise, purify, analyse and optimise? Certain factions of chemistry are already seeing the pioneering footsteps of automation and machine learning dramatically impact productivity in the lab. [2-7] The ability of these systems grows year-on-year as more reactions are added to the training sets, rapidly expanding the number of compounds possible to imagine and then synthesise.
Topically, the 2024 Nobel prizes in physics and chemistry were awarded for pioneering contributions that solidified AI as an indispensable tool in research. John Hopfield and Geoffrey Hinton developed crucial methods that are at the heart of today’s powerful machine learning algorithms while David Baker, Demis Hassabis and John Jumper applied these technologies to computational protein design and protein structure prediction. Their breakthroughs have dramatically marked a turning point where AI’s role is not just supplementary, but foundational.
Responses to these awards have varied. Obviously, these systems are deeply impressive and represent an evolution in the way science is conducted. However, others are keen to emphasise that the raw data that made these advanced models possible should not be forgotten, underscoring the importance of experimental work that enables machine learning’s rapid progress.
 The lines between human and machine creation are increasingly blurred
In the early days of AI, outputs were often clunky and unreliable, plagued by frequent glitches that highlighted its experimental nature. The technology seemed promising, but its shortcomings made it easy to distinguish between human and machine-generated outputs. Fast forward just a few years and AI has undergone a remarkable, and somewhat beautiful transformation. The progress has been so profound that today, the lines between human and machine creation are increasingly blurred. Well-trained AI systems can now parse procedures, analyse spectra and conduct new experiments with such speed and precision that the question is no longer whether they can match human quality, but whether we can even tell if something was human made at all.
This evolution challenges us to reconsider not just the capabilities of AI, but our ability to discern the origin of what we see and read in an era where the boundaries between human and artificial creativity are becoming indistinguishable. So far, these systems have been used to investigate research questions that we find interesting. Perhaps future generations will begin formulating their own hypotheses and conducting their own experiments.
When a new paper comes out by Professor R Obot, are we going to call the work beautiful too?