Mere weeks after Microsoft had to shut down its Tay chatbot for sending offensive tweets, the company might be in for another AI-related embarrassment – albeit probably not as serious.
This time, the culprit is Captionbot, an AI experiment that tries to recognize what’s on a photo and come up with a text caption. Launched in late March, Captionbot actually does a pretty good job in many examples, but it’s also horridly wrong in others, and when that happens, social media can be unforgiving.
In one example, Captionbot mistakes the First Lady Michelle Obama for a cellphone. In another, it thinks the famous blue/black/white/gold dress is a cat wearing a tie. (Or a suitcase, see image above).
It’s not just that Captionbot is sometimes wrong — it’d be absurd to expect it to always be right in a world where even humans struggle to understand what’s in a CAPTCHA image. But things get awkward when it’s not only wrong, but weird. Check out some examples HERE.
