AFTER TROUBLE WITH its last bot, Microsoft has turned towards a safer experiment: identifying images.
CaptionBot was one of the bots revealed at Microsoft’s developer conference Build and is designed to give any photo you upload a caption. So for two of the preloaded photos, it’s able to identify a snowy mountain or a man on a skateboard.
This is done using natural language so it sounds more like something a person would say.
Much like Facebook’s own bots, this is far from perfect. While it will say if it’s unable to identify something, it will likely struggle in most cases.
For example, it thought two different photos of the iPhone 6 was a keyboard and a remote control, although it has its moments (in one case, it was able to identify prominent tech figures like Apple’s Tim Cook).
But it’s also aware of the problems faced by a previous Microsoft bot. Business Insider tried uploading photos of Hitler to find out it wouldn’t identify them, suggesting that such photos are deliberately blocked.
The block was likely down to Microsoft’s last AI bot, which had to be taken down after it started repeating racist remarks and inflammatory statements. It was brought back a day later but was taken down a second time.
CaptionBot is Microsoft’s latest experiment with artificial intelligence (AI). The company has released other experiments like a bot that guesses your age or says what dog is in an uploaded photo, but recent weeks has seen it take a more serious approach to AI by announcing chatbots with Skype.
To embed this post, copy the code below on your site
have your say