AI executives are betting big on biology. They say their large language models that have already dramatically changed the fields of coding, software development, and mathematics can upend biological research in the same way, reports BritPanorama.
However, one company’s attempt to do so has already set off a wave of skepticism and controversy.
Anthropic, the AI and research giant behind Claude, last month announced that its biology research lab used AI agents to discover an unusual pattern of DNA in the genetic code of viruses, specifically a novel enzyme system. The signature is analogous to the microbial system harnessed in gene-editing tools such as CRISPR Cas-9, which is widely used to modify the DNA of living organisms. This technology won the Nobel Prize in Chemistry in 2020 for its inventors.
Dario Amodei, the chief executive of Anthropic, detailed on social media that this finding offers significant scientific potential. “Today we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism,” he wrote on X.
Anthropic stated that the discovery was made “with only high-level direction from our scientists,” leading to a decline in shares of companies developing gene-editing technology shortly after the announcement.
However, some experts not involved in the work expressed that the finding was preliminary and incremental, falling short of constituting a genuine breakthrough.
“These are very interesting preliminary results, but at this point, they don’t demonstrate gene editing or provide a mechanistic picture of exactly what’s happening,” commented Aaron Engelhart, an associate professor at the University of Minnesota’s Department of Genetics, Cell Biology, and Development.
The Anthropic researchers have yet to ascertain what the newly identified viral sequences do, and whether they represent a mechanism with practical applications akin to those of CRISPR tools. Notably, the research highlighting this discovery has not yet been published in a peer-reviewed scientific journal.
Scientists, including Engelhart, indicated that Anthropic’s finding certainly warrants additional investigation. A news release from the company included a quote from Feng Zhang—a professor at MIT and the Broad Institute and a pioneer of CRISPR genome editing—who described the work as “genuinely intriguing.”
However, shortly after Anthropic’s announcement, Mario Rodríguez Mestre, who relied on Claude for his recently completed doctoral research at the University of Copenhagen, asserted that his unpublished research describes the same viral signatures.
“It is essentially the same finding. This is not simply a case of two groups studying related protein families or similar biological systems,” Mestre explained via email.
Defining originality
Mestre’s claims echo those made by mathematicians after OpenAI, which developed ChatGPT, announced it had resolved a longstanding and high-profile math problem. They also raise questions regarding the process leading to Claude’s finding.
The biologists at Anthropic’s lab instructed Claude to scour a vast database of DNA sequences for “interesting new examples” of reverse transcriptase, an enzyme that plays a role in the spread of genetic information among organisms.
The company reported that 950 AI agents, capable of independently planning and executing tasks, dedicated 21 hours to this task, filtering a broad pool down to the 20 most compelling reverse transcriptases. An analysis that would typically consume weeks to months for human scientists.
“One of the agents spotted something remarkable: a repeating pattern of DNA sequences that occurs next to the gene for an odd-looking reverse transcriptase,” the company noted.
Following this, Anthropic’s scientists examined and tested the finding, which they characterized as a “previously uncharacterized” biological system in bacteriophages—a type of virus that targets bacteria—bearing a structural resemblance to CRISPR.
However, Mestre stated that he had shared unpublished research on his private Claude account, encompassing drafts of his dissertation and analyses that describe these systems.
“I am not claiming that Anthropic deliberately took our work. I cannot demonstrate that,” he clarified.
“Either information related to our work somehow reached the model and it was trained on it, or the second possibility, which I currently think is more plausible, is that there was considerably more prior scientific knowledge and human direction behind the search than the phrase ‘autonomous discovery’ suggests,” Mestre added, referencing the terminology used in Anthropic’s release.
Anthropic did not respond to a request for comment.
While it is common in scientific research for various groups to arrive at similar results, Mestre’s allegations carry significant weight, leading many scientists to reconsider the wisdom of inputting unpublished results into these tools.
New territory
Nonetheless, Anthropic’s announcement suggests that AI is starting to operate with greater autonomy within scientific realms, according to Gustavo Sudre, a professor of genomic neuroimaging and AI at King’s College London.
“It received broad direction and then had to use its judgment about what was interesting in the data. That represents a real shift from the usual ‘analyze this dataset for me,’” he noted.
Finkelstein also remarked that while AI excels at the “needle in a haystack search,” it is less effective at achieving “imaginative breakthroughs.” AI can efficiently explore domains where human researchers might tire or grow disinterested in identifying noteworthy patterns.
“They can pursue many unproductive paths rapidly,” he stated. “They are effective at grunt work when coordinated by domain experts.”
Finkelstein highlighted that the lead author of the Anthropic paper, Dr. Peter Yoon, previously worked in the lab of Jennifer Doudna, one of the architects of CRISPR, who shared the 2020 Nobel Prize with Emmanuelle Charpentier. This suggests that Anthropic’s AI was guided by prominent figures in the field.
“Four of six authors are senior molecular biologists and domain experts. If you give that group a heap of computational resources and frontier models, we can expect more exciting bioinformatic discoveries,” he noted in a blog post.
However, some researchers argue that claims of autonomous AI-driven discovery can be overstated, indicating that current models are not sufficiently capable of generating meaningful results independently.
While permitting a model to analyze data beyond human capacity does advance the field, it does not necessarily mark a significant breakthrough, Sudre remarked. “From experience, the gap between identifying an ‘interesting pattern’ and deriving ‘useful knowledge’ is where most of the work resides,” he noted.
Anthropic launched Claude Science, its research-focused AI tool, earlier this year. Its competitors have similarly unveiled advanced tools. OpenAI introduced GPT-Rosalind, Microsoft has Quine, and Google DeepMind has Co-Scientist. These tools function like large language models but utilize specific scientific datasets, including DNA sequences.
Nevertheless, the efficacy of models is inherently linked to the quality of training data. For instance, the strength of Google DeepMind’s AlphaFold, which won the 2024 Nobel Prize in Chemistry for determining protein structures, is derived from a specialized data source amassed over decades.
AI developers face intense public scrutiny as apprehension grows over the potential for rogue AI agents and the feasibility of properly managing them. Breakthroughs in biology may provide executives with an opportunity to cultivate a more positive narrative countering fears surrounding their products.
Yet, achieving substantial and tangible results in the field will likely necessitate the integration of AI models into practical lab work through robotics.
“Biology is not maths; at some point, intelligence must manifest in the physical realm through robots capable of executing intricate experimental tasks, akin to what a skilled student would perform instinctively,” remarked Yuval Elani, an associate professor in biochemical technologies at Imperial College London. “Based on current advancements in robotics and automation, we are not close to realizing the hype surrounding AI in biology.”
Sudre underscored that the role of human scientists will persist rather than diminish. “The true risk isn’t that scientists will be supplanted in discovery; it’s that they may become complacent and rely too heavily on AI to determine the significance of findings,” he cautioned.
The next stage for Anthropic’s scientists involves painstaking laboratory experiments aimed at determining the practical value of the pattern they have identified in biotechnological applications.
“Of course, in the future, some of that work could be delegated to robots,” Sudre remarked. “But the essence of inquiry in science doesn’t simply reside in databases awaiting discovery. It springs from our own interests and concerns.”
