Å®ÉúСÊÓƵ

Step aside, Picasso. This AI can – but they won’t always be recognisable.

The algorithm, created by Tao Xu at Microsoft Research and his colleagues, works similarly to other AIs that generate art or pictures. It is trained on a database of photographs and descriptions, so it matches certain words to particular colours, textures, and shapes. And like other AIs, it is not always spot-on and the images don’t always make sense.

Cristóbal Valenzuela at New York University made a website to demonstrate the model, where you can type in anything you want and it will build a picture for you. You can see examples in the pictures above.

It’s pretty good at scenic views. If you write a description of an animal, it isn’t too far off – it gets the texture right, at least. But the AI doesn’t do so well with humans, mixing and scrambling their features like Pablo Picasso’s famous portraits. On the upside, it does seem to be able to tell you how to get to Sesame Street.