Researchers at the University of Liverpool have developed an AI-powered mobile robot that can carry out chemical synthesis research very efficiently.
In a study published in the journal natureResearchers show how a mobile robot that uses AI logic to make decisions was able to perform exploratory chemical research tasks at the same level as humans, but much faster.
The 1.75m tall mobile robot was designed by the Liverpool team to solve three key problems in exploratory chemistry: carrying out reactions, analyzing products and deciding what to do next based on the data.
The two robots performed these tasks in a collaborative manner, solving problems in three different areas of chemical synthesis: structural diversification chemistry (relevant for drug discovery), supramolecular host-guest chemistry, and photochemical synthesis.
The results found that AI capabilities enabled mobile robots to make the same or similar decisions as human researchers, but that these decisions were made on a much faster time scale than humans, which can take hours.
Professor Andrew Cooper, from the University of Liverpool’s Chemistry and Materials Innovation Plant, who led the project, explained:
“Chemical synthesis research is time-consuming and expensive, both in terms of physical experiments and decisions about which experiments to perform next, so using intelligent robots provides a way to accelerate this process.
“When people think about robotics and chemical automation, they tend to think about mixing solutions, heating reactions, etc. That’s part of it, but decision-making can be at least time-consuming. This is especially true for exploratory chemistry. You can’t be confident in the results. In cases where this is not possible, it requires nuanced, context-sensitive decisions about whether something is interesting based on multiple data sets, a time-consuming task for research chemists but a difficult problem for AI.
Decision making is a key problem in exploratory chemistry. For example, a researcher may run several test reactions and then decide to scale up only those reactions that give good reaction yields or interesting products. AI is difficult to do this because the question of whether something is ‘interesting’ and worth pursuing can have many different contexts, such as the novelty of the reaction product or the cost and complexity of the synthetic route.
Dr Sriram Vijayakrishnan, a former PhD student at the University of Liverpool and a postdoctoral researcher in the Department of Chemistry who led the synthetic work, explained: “When I was doing my PhD, I performed many chemical reactions by hand. While setting up experiments, it took a long time to generate analytical data. When you start automating chemistry, you can end up drowning in the data, so these data analysis problems becomes more serious.
“Here, we solved this problem by building AI logic for the robot, which processes analytical data sets to make autonomous decisions, such as whether to proceed to the next step in the response. These decisions are essentially instantaneous. “Performing an analysis at 3 a.m. will determine what reaction will proceed at 3:01 a.m. whereas processing the same data set may take a chemist’s time.”
Professor Cooper said: “Robots have less situational breadth than trained researchers, so in their current form they can’t ‘Eureka!’ There won’t be a moment, but for the task presented here, the AI logic makes the same decisions as a synthetic chemist for these three different chemistry problems, and makes these decisions in the blink of an eye. There is also great scope to expand our contextual understanding of AI, for example by linking it using large-scale language models, and linking directly to relevant scientific literature.”
In the future, the Liverpool team hopes to use this technology to discover new materials for applications such as carbon dioxide capture, as well as chemical reactions involved in the synthesis of pharmaceuticals.
Although we used two mobile robots in this study, there is no limit to the size of the robot team that can be used. This approach can therefore be scaled up to the largest industrial laboratories.
