In 1875, Thaddeus M. Stevens described one of the first examples of automated laboratory equipment. It was a sealed vessel containing a controlled drip rate to wash filter papers. One and a half centuries later, academic scientists still carry out many tasks manually.
Challenges in Creating Automated Laboratories
Compared to industry and clinical research, scientific research still mainly relies on the low-cost labor offered by postgraduate and postdoctoral students, leaving academic science with an efficiency problem. Laboratories usually sit empty at night and on weekends, while the short contracts and high turnover of junior staff cause the loss of valuable experience and knowledge.
Scientists across various disciplines seek ways to integrate machine learning into their experimental setups. These self-driving laboratories combine AI and automation to design and execute repetitive steps, analyze data, and tweak the next cycle of experiments to build on the outcomes.
Self-driving laboratories have already been used to design smart materials and find enzymes that human scientists would have struggled to produce. Meanwhile, universities and the private sector invest in cloud laboratories, which provide remote access to robot workers and laboratory equipment.
However, actual self-driving laboratories with fully automated systems remain rare. This is because academic research has not traditionally lent itself to loads of automation, as frequently changing protocols are harder to perform with robots.
Another challenge for many academic laboratories is experiments that depend on multiple steps that are hard to link. Few automated tools can link together commercially available automated experiments, so connecting each one of the tools or modules remains a problem in developing a fully automated workflow.
Combining AI and Automation
While artificial intelligence receives most of the attention, improved automation is needed to leverage machine learning in science, according to Curtis Berlinguette, a chemistry researcher who runs a self-driving laboratory at the University of British Columbia in Vancouver, Canada. Robots are the only feasible way for many laboratories to boost the production of reliable data that AI feeds on.
When done properly, machine learning can enable researchers to catalog and process data far more effectively than they could through manual experiments. These better-quality and annotated data streams permit artificial intelligence to analyze the results and decide what experiment the robot should do next. This will produce more data for AI to analyze, and so on.
According to Berlinguette, this separates a self-driving laboratory from a high-throughput experimentation. The latter creates massive datasets, but they are not necessarily processing that data on the fly.
Berlinguette's laboratory features an automated system called Ada, which uses a robot arm to produce thin-film materials for perovskite solar cells. Ada works by changing process conditions like additive concentration, annealing time, and temperature. It also compares the properties of the resulting films to a set goal and repeats and reiterates the experiment until it makes the best-performing material.
Other self-driving setups can have similar successes by combining automation and artificial intelligence. Last year, scientists from the Lawrence Berkeley National Laboratory (LBNL) and Google DeepMind claimed to have paired an algorithm that conceives and designs the structures of new inorganic materials with a robot system known as A-Lab to make and analyze them.
Check out more news and information on Laboratory in Science Times.
