Artificial intelligence has become remarkably good at answering questions.

Now researchers are asking whether it can help discover questions that nobody thought to ask.

This week, AI company Anthropic announced that its Claude model had helped identify a previously uncharacterised biological system in bacteriophages — viruses that infect bacteria. The company has named the system array-associated reverse transcriptases, or ART.

The discovery is particularly interesting because part of the system resembles something already famous in biology: CRISPR.

CRISPR began life as a strange repeating pattern in bacterial DNA. Scientists eventually discovered that CRISPR systems could be adapted into extraordinarily powerful tools for editing genes.

ART is not a new CRISPR. At least, not yet.

Scientists do not currently know exactly what the system does, and the research is still at an early stage. Anthropic has published its findings as a preprint and says further experiments are underway.

What makes the discovery remarkable is how it happened.

Anthropic's researchers asked Claude to search enormous quantities of genetic information for unusual examples of reverse transcriptases. These are enzymes capable of copying information from RNA into DNA.

The AI agents gathered more than 200,000 reverse transcriptases and identified around 3,500 candidate partner families. These were eventually narrowed down to 20 particularly compelling candidates for closer investigation.

The search involved roughly 950 AI agents working for around 21 hours, using approximately 210 million tokens.

Eventually, Claude noticed something unusual.

One reverse transcriptase was located next to a long sequence of regularly spaced DNA repeats. The pattern had similarities to the repeat arrays associated with CRISPR systems.

The underlying reverse transcriptase was not itself new. Researchers had identified it in previous studies. What Claude appears to have noticed was the wider system surrounding it: the repeat array and an additional partner protein whose function remains unknown.

The AI then did something important.

It did not simply announce that it had found something interesting and stop.

It counted the repeats, examined their spacing, compared the structure with known biological systems and searched the scientific literature to establish whether the pattern had already been described.

Human scientists then took over the laboratory work.

The system they investigated consists of three main components: the reverse transcriptase, a neighbouring partner gene and a long array of evenly spaced DNA repeats. Early experiments suggest that the repeat array is transcribed into a series of short RNA molecules.

Exactly what those molecules do remains unknown.

It would be tempting to describe ART as "the next CRISPR", but there is currently no evidence that it can be used as a programmable gene-editing tool. The resemblance is a reason for scientists to investigate the system, not proof that it will become a medical technology.

And that may be what makes the story most significant.

Scientific discovery has traditionally involved humans deciding what is interesting, designing an experiment and interpreting the results. AI can now search through datasets on a scale that would be extremely difficult for an individual researcher.

It does not mean scientists have become unnecessary. In this case, humans still designed the research programme, reviewed the candidates and performed the laboratory experiments.

Instead, the division of labour is changing.

The AI can look at an enormous amount of biological information and say:

This bit is strange — have you looked at it?

A human scientist can then ask the more important question:

Why?

That question is now being applied to ART.

If further experiments reveal that the system has a useful biological function, today's obscure collection of DNA repeats could eventually become the starting point for a new area of biotechnology.

Or it could turn out to be a fascinating biological curiosity with no practical application at all.

Both outcomes would still tell us something valuable.

Because the real breakthrough may not be ART itself.

It may be the demonstration that AI can help scientists find things hidden in the enormous amount of biological information that humans have already collected — things that were sitting there all along, waiting for someone, or something, to notice them.