Microsoft had to shut down its Tay AI for becoming a genocidal racist, but now the company has a new confused AI.
The new AI bot, called CaptionBot, looks to help generate automatic, human-like captions for uploaded pictures. CaptionBot is still learning and is often wrong, but users can rate the accuracy of the caption to improve the bot's skill over time. This AI experiment is available on the web and was developed by Microsoft's Cognitive Services team. According to CaptionBot's website, the bot was "created to showcase some of the new capabilities of Microsoft Cognitive Services," including Computer Vision and Natural Language. The bot is made of three APIs: Computer Vision API, Emotion API, and Bing Image API. CaptionBot acknowledges that it will make some mistakes, saying, "I am still learning, so sometimes I get things wrong."
While CaptionBot is still learning, it hopes to describe the content of an image "as well as any human," and perhaps over time it will be able to do so. "Captioning is about taking concrete objects and putting them together in a literal description," said Margaret Mitchell, a Microsoft researcher who is leading the research project. "What I've been calling visual storytelling is about inferring conceptual and abstract ideas from those concrete objects." CaptionBot's responses are somewhat cautious, saying things like, "I am not really confident" and "I think," rather than giving a direct answer. Some of CaptionBot's responses are absurdly wrong, but the AI is just starting out. At least this time, the AI bot has stayed sane after having experienced human interaction.
Photo: © iStock.