When AI sounds certain, blind users have to trust it – even when it is guessing

Artificial intelligence (AI) can help blind and visually impaired people make sense of the world around them – while also exposing where it falls short. The AI systems solve many tasks correctly, but they lack a built-in stop button when the evidence is too thin. In other words, they cannot reliably distinguish between knowledge and qualified guesswork – and often respond without signalling uncertainty, new research shows.