Study finds AI helps scientists analyze data but lab work slows progress

In short: Scientists are using AI widely to analyze research data, but many of the time savings do not translate into faster real-world discoveries because lab work and verification still take time.

What's going on

A new study led by researchers at Google looked at how scientists use Gemini, Google’s large language model (an AI that can work with text and data, like a very fast assistant). It combined usage data with a detailed survey of researchers in the US and UK.

The study found that AI is already a regular part of scientific work. Almost half of surveyed scientists said they use AI daily, and about 80 percent said they use it at least weekly. The most common use was analyzing or modeling research data.

But the survey also suggests there are limits to how much this speeds up science overall. Scientists said they save almost seven hours a week on average, but some of that time goes to checking AI output carefully. The researchers called this extra checking work a “verification tax” (like having to double-check a coworker who sometimes makes mistakes).

Another major slowdown is the physical world. Many researchers said the biggest bottlenecks are running experiments and collecting data. Some said their backlog of untested ideas has grown, meaning AI can help generate more theories faster than labs can test them.

What to watch

If labs, testing capacity, and regulations do not speed up too, AI may mostly increase the number of ideas on paper rather than the number of proven results. Another open question is whether research groups will hire more people and build better processes for verification, so scientists can trust AI help without spending so much time auditing it.

Source: Financial Times