AI Fails to Rate Wheelchair Access at Historic Sites, Study Finds

When you plan a visit to a historic site, you might check online if it's wheelchair accessible. But what if that information is wrong or missing? A new study looked at whether AI could help assess accessibility at heritage sites in Spain and Italy. They found that AI, while fast, isn't reliable enough to make the call on its own.

The researchers tested an AI system that was supposed to evaluate accessibility across five parks. It only managed to score 66% of the required checks. More importantly, it couldn't assess things like how staff interact with visitors with disabilities or the actual physical experience of navigating the site. Instead, the AI just repeated what the sites said about themselves online, which might not be accurate.

So what does this mean for you? If you're planning a trip and rely on AI tools to tell you if a place is accessible, you could be misled. The study suggests that AI should be used to gather information, but human experts and simple statistical models should make the final judgment. That way, you get a more reliable picture of what to expect.

The researchers propose a team approach: AI collects data, and then interpretable statistical models (think simple math that shows its work) check how confident we can be in that data. This combination could lead to a trustworthy accessibility atlas for heritage sites. For now, if you need accessibility info, it's best to double-check with the site directly or use specialized resources rather than relying solely on AI.