Wednesday · Understand it

The robot team that found another question

Hello,

This week, a robot story that is really about judgement: when a machine spots a new lead, what should it be allowed to change?

Two things worth knowing

Eye screening: the result while the patient is still there. In a Tanzanian trial, more people with referable diabetic eye disease attended a specialist clinic after an AI-supported screening pathway than after standard screening. The AI pathway also gave people an immediate result and face-to-face counselling, rather than a later result by phone. We cannot credit the difference to the software alone. Read the researchers' account.

Talking to Gmail, Docs and Keep. Google says eligible users can ask Gmail to find information, talk through a draft in Docs and turn spoken thoughts into organised notes in Keep. In September, Google listed access for particular paid AI plans and said Workspace business access would come later. Check what's available in your account before considering a paid plan. Read Google's announcement.

What happened in Virginia

A drone was carrying out a search in Virginia when it found another area worth investigating. It already had a job. The question was whether to stick to that job or do something about the new lead.

In NASA's account of the field test, the system kept its original task going and brought in another robot to investigate. It is a small change of plan, but it gets at a difficult question for space exploration: what should a machine be allowed to decide when the people running a mission cannot direct every move in real time?

A team with different jobs

The test took place in July 2026 at the Virginia Tech Transportation Institute in Blacksburg. NASA and its partners used three robots: a drone to scout from above and two rovers to work on the ground. One rover had lidar to map its surroundings and help it navigate. The other had an arm for collecting samples.

People set the science goals. The drone was then assigned to choose an area of interest, and the software weighed which robot was best placed to investigate. Distance and risk mattered, but so did the potential value of what the team might learn. The robots also had to keep checking for hazards and report back to their human managers.

That matters because a promising patch of ground is not automatically worth chasing. A rover might be too far away, the route might be risky, or the new question might be less important than the work already under way. The system had to deal with those competing priorities inside a brief set by people.

When the drone found the second area of interest, it did not abandon its first task. According to NASA, the software paused to assess the opportunity and brought another robot in to look. That is the particular behaviour this test showed. It does not mean the robots became independent scientists or discovered something on another world.

Why test this on Earth?

NASA wants future robotic teams to be useful where a human instruction cannot arrive quickly enough, or where sending people would be too dangerous. Its examples include distant uncrewed missions and difficult parts of the Moon or Mars. For now, this was an Earth-based field test, with human managers and a controlled set of science goals.

NASA's report describes what the team observed, but it does not give a detailed public scorecard for how often the system chose well, how it handled failures, or whether it is ready for a mission. Those are different questions from whether one change of plan worked during a test.

The interesting part is the boundary. People decided what knowledge mattered and how much risk was acceptable. The robots were given room to make smaller choices within that frame and to tell the team what they had done. When you hear that an AI system can act on its own, those are useful things to ask about: who set its priorities, what can it change, and how will a person know when it has changed course?

If a story here raises a question or needs correcting, reply to Millie.

Millie

If you want help working out where AI could be useful in your business, tell me what you're trying to decide. Get in touch.

Someone forwarded this to you? Subscribe to One Eye on AI.

Sources and corrections

The linked eye-screening item follows the researchers' institutional account of a randomised trial, and the Google item follows Google's product announcement. The main story follows NASA's own field-test report. None is first-hand reporting by One Eye on AI. The field test is not proof of readiness for a space mission, and the eye-screening comparison tested a whole care pathway, not the software in isolation. Reply if a source or correction materially changes this account.