A reliable railway and a laboratory discovery might seem worlds apart. Both show how AI can help people make sense of complicated patterns without handing over the final decision.
AI could help make Britain’s railways more reliable
Researchers at Royal Holloway, University of London are exploring AI to predict rail delays and reduce the time level crossings need to remain closed. The university highlighted the work on 26 September.
One strand, presented earlier in the year by Dr Xu (Sean) Feng, learns from four years of railway records to make delay forecasts with a measure of confidence. Rather than just saying a train might be seven minutes late, a prediction can indicate how certain the system is. That matters when a forecast might inform operational decisions affecting passengers.
Thinking about the wider rail network matters too: a delay is rarely isolated from other trains and connections. The aim is to give operators better information about what could happen next, not to promise that a model can prevent every disruption.
Royal Holloway also says its researchers are working on reducing level-crossing downtime. These are research projects, not a nationwide deployed system; the university’s announcements do not establish that journey times or crossing closures have already improved.
Sources: Royal Holloway: 26 September research update ↗ · Royal Holloway: rail forecasts and confidence ↗
AI scientists hunting for new biology
Another research story worth a closer look this weekend is the AI-assisted biology finding we covered on 24 September. It is not a separate discovery today.
Anthropic researchers tasked AI agents with searching a large collection of DNA sequences for unusual reverse transcriptases, enzymes that copy RNA into DNA. Roughly 950 agents worked for about 21 hours, and one flagged a repeating DNA pattern associated with an enzyme.
Human scientists then investigated the lead in the laboratory. They describe the system as ART, short for array-associated reverse transcriptases. Its arrangement has some features reminiscent of other repeat-array systems, but its biological function remains unknown. There is no evidence here of a new gene-editing tool or treatment.
The work is available as a preprint. The demonstrated achievement is that AI helped researchers identify a lead in a huge dataset and people followed it up experimentally; what the system does still needs to be established.
Sources: Anthropic: AI-assisted research and laboratory follow-up ↗ · Research preprint ↗
People decide what happens next
On the railways, AI might help people anticipate delays and understand uncertainty. In science, it can narrow an enormous search space to something worth testing. Neither is an argument for handing control to AI.
Use machines to find patterns and quantify uncertainty. Let the people with the relevant expertise decide what to do with that information.
No AI for the sake of AI.
Less hype. More real-world AI progress.