Stay up so far with free updates
Simply log in Artificial intelligence myFT Digest – delivered straight to your inbox.
Google DeepMind and BioNTech are developing AI lab assistants to assist researchers plan scientific experiments and higher predict their results, as corporations compete to search out specialized applications for energy- and data-intensive artificial intelligence models.
Sir Demis Hassabis, head of Google's AI division, is leading the corporate's efforts to develop a specialized AI model that acts as a research assistant, helping scientists collaborate across disciplines and make unexpected connections more easily.
At a recent Nobel Foundation event, he said that biology is “undergoing a revolution” due to AI software.
“We're working on a big scientific language model that may very well be like a research assistant and possibly aid you predict. . . the results of an experiment,” said Hassabis.
He said that in the approaching years, the tools DeepMind developed could suggest and design experiments based on a particular hypothesis, giving scientists a probabilistic view of the potential success or failure of a proposed experiment.
Meanwhile, German drugmaker BioNTech and its London-based AI subsidiary InstaDeep announced on Tuesday that that they had developed a specialized AI assistant called Laila with “detailed knowledge of biology” based on Meta's open-source Llama 3.1 model construct up.
In a live demonstration, researcher Arnu Pretorius showed how the AI agent could automate routine scientific tasks in experimental biology, corresponding to the evaluation and segmentation of DNA sequences and the visualization of experimental results.
Scientists on the BioNTech laboratory in Mainz also demonstrated how Laila can connect with laboratory equipment and monitor ongoing experiments or tasks performed by robots by having the assistant detect a mechanical fault on a BioNTech machine during a live demonstration.
“We don’t consider the long run is full AI automation any time soon. We see AI agents like Laila as a productivity accelerator, allowing scientists and engineers to spend their limited time on what really matters,” Karim Beguir, managing director of InstaDeep, told the Financial Times.
In the primary presentation of its technology for the reason that Covid-19 vaccine maker acquired InstaDeep in 2023 for as much as £500 million, InstaDeep also unveiled AI models that would help BioNTech discover or discover recent targets to fight cancer .
While Beguir acknowledged that competitors like DeepMind could also develop AI assistants, he said the incontrovertible fact that InstaDeep's technology is “under one roof” with BioNTech's expertise in biology is an “accelerator” for implementation of AI and is “unique” within the pharmaceutical sector.
The recent scientific assistants come at a time when tech corporations are spending billions of dollars on AI models and products because they consider the technology can transform industries from healthcare to energy and education.
The wave of AI innovation in science has up to now focused on predicting recent and useful drug candidates. However, the bottleneck in bringing recent treatments to market stays conducting real-world experiments, which is the gold standard in scientific research.
The goal of an AI research assistant could be to simplify this process by planning experiments more effectively, for instance by choosing probably the most promising ones from a set of possible experiments.
Companies like Google and Microsoft are adapting large language models—software that may generate text, code, images, and even DNA or molecular sequences based on large training data sets—to enable scientific breakthroughs.
In 2022, DeepMind developed an AI system called AlphaFold that would predict the form of just about every known protein, solving a 50-year-old scientific challenge and potentially significantly reducing the time required for biological discoveries.
Nobel Prize-winning geneticist Paul Nurse said on the Nobel Prize event in March that members of his lab use “AlphaFold on a regular basis” in biochemical experiments, adding that the AI model's results “aren’t at all times right, but they’re right enough to to be it”. a implausible tool.”
Hassabis has since spun off that work into an AI drug offshoot called Isomorphic Labs, a bunch on whose advisory board Nurse sits and which has agreed to partnerships price as much as $3 billion with Eli Lilly and Novartis.
Microsoft's AI4Science Research division also uses LLMs to speed up scientific discoveries. Its director, Chris Bishop, said at a research forum this yr that considered one of the notable properties of LLMs was that “they will function as effective reasoning engines,” which is especially useful in science.
Bishop said the team worked with the Global Health Drug Discovery Institute to make use of LLMs to find recent molecules to treat tuberculosis more effectively.
