The emergence of agentic AI in lab automation

 “The lab is a place with highly educated, smart people spending a lot of time doing repetitive tasks,” says Marco Ravot-Licheri, as we discuss the current challenges in medical research. According to Ravot-Licheri, laboratories have hardly evolved over the last decade, with people basing the structure of their work around technology. Tasks such as measuring reagents and monitoring experiment parameters are still done by humans, whose time could be better spent thinking strategically about the research, according to Ravot-Licheri. Enter “Agentic AI”, a technology that can ensure experiments run smoothly and detect any issues that might derail the research before they become critical. With this approach, many tasks are automated, such as handling liquids, which robots are able to do with accuracy and precision. On top of this is a layer of software that allows operators to oversee and run the automated systems.

“You have an analytics layer that sees how you use the fleet of robots, you can add the agentic capability to tell the operator how to use the instruments.” The software collects data, sends it to the cloud, where it is processed, and can be used by AI agents as part of the automation process. Tecan is working with NVIDIA, which has provided the framework and a toolkit that provides the interface allowing operators to create an agent that can assist with the day-to-day running of the lab. Agents typically monitor parameters within the lab and communicate with operators when situations arise that require an intervention. This can prevent situations such as having to rerun an expensive and time-consuming test because a cell sample has been contaminated or killed. A system that proactively monitors an experiment and warns when things are about to go awry could cut costs, improve productivity and save time, according to Ravot-Licheri. “Agentic AI informs you before something happens,” he explained.

Humidity warning

A practical example cited by Marco Ravot-Licheri is humidity, an issue that has affected experiments in his experience in the lab. An agentic AI system could monitor the humidity in the laboratory as an experiment is run, flagging if the conditions in the lab are approaching a point where the culture may be compromised. The system is capable of analysing hours of previous results in a database to find the correlation between humidity and the failure of an experiment. Setting an AI agent to use the lab systems to constantly monitor humidity and warn if it approaches certain levels allows researchers to perform other tasks while the experiment runs. At the same time, they are safe in the knowledge that alarms will sound should conditions become unfavourable.

AI Native

A key part of the philosophy is to build a lab that is “AI Native” – designed from the ground upwards to run using AI systems. Trying to integrate AI into an existing lab setup often leads to sub-optimal outcomes due to unforeseen technical issues. He explained: “If you want to have transformational use cases, you will hit roadblocks. If you  build a platform where you know AI will be front and centre and it has AI at its core, you can really develop use cases and do it fast.” An AI-based system will be constantly evolving in line with the tasks it has been set. In today’s highly competitive world, being able to iterate and change things quickly is an important consideration. “You need to constantly keep learning to actually test things and see what works and what needs to be improved before you release a product,” he said. The need to quickly develop new therapies and hypotheses was demonstrated by the response to the COVID-19 pandemic. Machine learning algorithms were used in the analysis to identify vaccine targets in a fraction of the time it would have taken human researchers.

Future of the lab

According to Ravot-Licheri AI will play an integral role in the development of the lab and scientific research in the coming decade. The whole system is “ripe for a significant improvement”, with the lab functioning as a fully integrated part of the research process. “I see the trend of the lab in the loop,” he said. AI can help with the process of developing ideas for research and as a tool to test them before they get to the laboratory. He said: “In the past, the bottleneck was developing the ideas. Now, it is so easy to design proteins and make hypotheses. Even if you put the scientific papers in the world, they have hidden errors inside. You still need to test them, if you have a small error, it can be amplified. In the next decade, I see a focus on making it much easier to test the hypothesis, the output feeds back into the AI and you have a closed loop solution.”

The human touch

Ravot-Licheri sees AI in the lab as an enabler, allowing laboratory teams to realise their full potential. AI shouldn’t be seen as a threat to jobs, or a way of cutting labour costs and replacing the role of lab assistants. He said: “There is the school of thought that the lab is going to look similar and be more integrated. There is another school that we will have labs without lights, with no humans.” Ravot-Licheri’s vision is based on the former school of thought, with scientists at the centre of research, enabled by AI and freed to do their job, thinking of hypotheses and testing them to develop new therapies and approaches to treating disease. Despite the automation, people should be there to run the lab and make the key decisions, based on the best available information.

He said: “At some point, something happens, there will be need for experts to make that lab work. The way I see it, AI is a very powerful tool, being so powerful it can be used for not so good reasons.” With AI already playing such a major role in research, the task now is to ensure that it is used responsibly and in ways that are beneficial. “Having use cases that are easy to implement and move the needle, with analytics and agentic AI tool to predict errors, it is the best bet to make sure people get familiar with AI and use it for meaningful purposes. You avoid unnecessary errors and let the team focus on what they are hired for, I see it more from that perspective.”