We're not really confident, but Microsoft's new CaptionBot seems a bit ropey

As Microsoft made clear at the recent Build conference, it's going big on bots (including a number of efforts for Skype), and has just released a new bot onto the web which will take any image and write a caption for it.

The imaginatively named CaptionBot uses AI smarts (image recognition accuracy and 'ultra-deep learning' is something Redmond has been working on for some time now) to analyse the image and attempt to understand the content, then tries to produce an appropriate one-line wonder to encapsulate what's going on in the picture.

All you have to do is head here and click to upload a photo from your PC. (Incidentally, for those concerned about privacy aspects, Microsoft does say it keeps the images uploaded – all in the name of learning and helping to improve CaptionBot – but no personal information is taken or stored).

Or alternatively, if you don't want to use a photo from your hard drive, you can slap in a URL pointing to an image.

And then, in Microsoft's words, CaptionBot will "describe it as well as any human". A bold boast indeed – and of course one that needs to be tested. So we fired some images at the bot to see what captions Redmond's AI would come back with.

Here are the results in all their glory:

CaptionBot Partridge

A-ha! It's Sam Waterston. Or at least 52% of him.

CaptionBot field

We think CaptionBot may just have come up with a brilliant new invention here – the kite-brella! We're off down the patent office right now…

CaptionBot Mortal Kombat

Either that, or a bloke having a piece of wood smashed over his head. The two are easily confused. (And don't you recognise Mortal Kombat when you see it, CaptionBot?)

CaptionBot frisbee

A frisbee? Huh? An invisible frisbee maybe. There's not even anything circular in the image…

CaptionBot group shot

Well, you're not alone in that, CaptionBot.