Artificial intelligence continues to grow rapidly, but it’s still not perfect. Look no further than Google’s I/O keynote from this week where it showed off its AI model Gemini giving a terrible answer to a struggling photographer.
In the demo video, a photographer runs into trouble when his analog camera lever stops working correctly. Naturally, he asks Gemini what to do. He opens Google Lens and prompts, “Why is the lever not moving all the way?” A few suggestions come up, which vary in usefulness. Astonishingly, of all these options, Gemini chooses the worst of them all, a detail noticed by The Verge.
In the video, the AI assistant highlights the suggestion to “open the back door and gently remove the film.” Of course, exposing the roll to light would destroy the film and any pictures already taken. It’s also clear from the video seen in the Google Lens frame that this photographer is outside or, at the very least, in a lit space.
Other suggestions, like “nudge the shutter a bit to reset it,” even in rare cases, as it says, seem like easy ways to damage a camera.
What’s likely happening in this scenario is the film is spent. The lever manually moves the film forward on analog cameras after taking a photo, but once the roll is spent, there’s nothing left to move. It’s a common scenario, as it happens every time a photographer finishes a roll of film.
Again, this is in Google’s own keynote demo video. It’s a completely unforced error. Someone had to capture this demo of Gemini in action, edit it into a video, and then decide, yes, this is how to showcase this tool.
The entire I/O keynote focused primarily on how Gemini can make everyday tasks easier and how intuitive the AI model is. In another example, Gemini turns a text-based study guide into a chat between two AI voices, allowing users to ask questions. It even goes so far as to insert a metaphor outside the source text. Seeing less-than-stellar advice (again, in Google’s own demo) doesn’t inspire confidence in the quality of those breakdowns.
In short, trust but verify. And maybe don’t rely on AI for camera troubleshooting advice.
Image credits: Header photo licensed via Depositphotos.