Readers like you keep news free for everyone.

More than 5,000 readers have already pitched in to keep free access to The Journal.

For the price of one cup of coffee each week you can help keep paywalls away.

Support us today
Not now
Sunday 26 March 2023 Dublin: 7°C
imageidentify If it doesn't have context, some of its guesses can be a little off.
# machine learning
This site wants to try identify any image you show it
It’s a new project from Wolfram Alpha and it’s quite good at recognising general objects.

AS FAR AS online resources go, Wolfram Alpha is one of the more useful ones out there. Describing itself as a ‘computational knowledge engine’, it can understand questions, context and is able to answer complex equations.

Now its creator Stephen Wolfram has a new project called the Language Image Identification Project, allowing anyone to take a picture and see what the service thinks it is.

It works similarly to reverse image searching on search engines, but this one goes a step further and provides additional data about its guess.

It takes a few seconds to provide an answer – the site keeps the pictures sent in and uses them for future reference so keep that in mind – but it’s the result of years of work from Wolfram and his team. A service that looks beyond pixels and tries to identify the general object.

“It won’t always get it right, but most of the time I think it does remarkably well,” wrote Wolfram on his blog. “And to me what’s particularly fascinating is that when it does get something wrong, the mistakes it makes mostly seem remarkably human.”

You’re not going to get exact descriptions from it, but anything that’s general will be recognised. If it requires context or trends (a new product for example), then it’s not going to be as accurate.

For example, a picture of an iPhone provided us the answer ‘remote control’ which is wrong but you can see where the confusion comes from. Other objects brought similar results like a typewriter (‘computer keyboard), a PS4 (‘device’ and ‘electric battery’) and a Ferrari (‘coupe’).

iPhone remote quinton quinton

To try and train the system, it was fed “a few tens of millions ” images for it to learn, with a number of different or unusual images thrown in for good measure.

unexpected-images-tested-with-imageidentify Stephen Wolfram Blog Stephen Wolfram Blog

While it’s easy to poke fun at this, the implications behind it are far greater than it might imply.

By implementing machine learning, you can tell if it got its guess wrong and explain what it is instead. The system will take that on board and will not only use the same image for future guesses, but it’ll cross-reference it with similar images so it gets the next guess right.

Wolfram says that it will continue to train and develop it and that it’s probably best to view this as how far artificial intelligence has come.

Think of it as a celebration of where artificial intelligence has reached. Think of it as an intellectual recreation that helps build intuition for what artificial intelligence is like. But don’t forget the part that I think is most exciting: it’s also practical technology, that you can use here and now in the Wolfram Language, and deploy wherever you want.

Read: How do you want your Windows 10? Cos there’s going to be seven different versions… >

Read:  This camera drone flies itself and will follow you wherever you go* >

Your Voice
Readers Comments