Pinterest and Facebook are the unlikely pioneers of artificial intelligence-led photo recognition lately, but Google and Microsoft are not far behind.
Artificial intelligence (AI) covers a whole swathe of areas, including anything from robotics to gaming, but its basis rests solely on big data, and how it analyses such.
When it comes to today’s leading areas of AI research, we are quickly drawn to image recognition. Why? Well, it’s the best example yet of computers catching up with humans – though they’re a fair bit off just yet.
Social media, you might be surprised to read, is actually driving a lot of today’s image recognition developments. That’s largely because social media itself is entirely structured through big data, with each operator quickly becoming experts in this field.
We feed this monster
Take Facebook, for example. Through its own algorithms, it orchestrates a system that promotes the right ads or services, at the right time, in the right place, to the right users.
You help the company do this, of course, through things like ‘liking’ band pages, but pretty much everything you do within the Facebook service (and beyond) is fed into the company’s burgeoning AI database.
This, lately, has shifted more towards images. Well, lets face it, social media has, lately, shifted more towards images.
Google was the first tech giant to show its hand in this regard, playing its part in two groundbreaking papers this time last year that showed us, for perhaps the first time, software that could ‘read’ images.
Bidding to caption a series of images, the two pieces of research used machine-learning to try and largely fail at describing simple settings. But, in terms of machine learning, it is the failing that eventually leads to succeeding.
A neverending story
The company didn’t stop there, obviously, using its Photos app to learn even more about how we take images, what we take pictures of and, largely, why.
Since Google opened up Photos back in May, allowing people to back up unlimited hi-res images and video, for free, 100m people have started using the service.
It is mass use like this that leads to improvements, just think of Microsoft’s instantly viral ‘How old am I?’ software last May that let users upload a selfie, before the software guessed at the person’s age.
Through basic corrections by users who took the program in good spirits, Microsoft achieved more in a few days’ research than most companies could garner in a year.
The latest step
Now, though, Facebook and Pinterest are driving ahead of the rest. The latter has, for example, just announced a new feature called Guided Search.
This lets you hone in on one element of an image, and search for other images that look like that. So, if you have an image of the interior of a shoe shop, you can choose just one pair at the back right and it can look for similar pictures.
It’s a bit like taking a screen grab of something and dropping it in Google Image search, I guess, but it’s all housed in Pinterest’s service. It is computer software reaching the image recognition levels of an infant, or dog perhaps.
You can view it in action, here.
This may not sound impressive but, in comparison to what went before, it’s pretty cool. Yet even Pinterest’s work is no match for Facebook right now.
Facebook the leader
At the start of the month, Facebook revealed its latest AI project, with a new app that can answer verbal questions about what appears in pictures.
Looking at a number of photos, Yann LeCun, director of Facebook’s artificial intelligence research group, showed the app answering several questions about what was going on, what animals were involved, what they were doing, what colour were they etc.
One example was a dog with a toy in its mouth, with the app correctly answering a question about what game the dog was playing: frisbee.
Now, for a little bit more invasive fun (well, it is Facebook), the social media giant’s latest toy is ‘Photo Magic’, which trawls through your phone’s photo roll to find other Facebook users’ faces. If it finds one, it prompts you to upload the image and tag them in it.
As Forbes recently explained, photos and videos were, up until quite recently, thought of as collections of random inputs of data. Now, though, they are not. They can be measured, sorted, discovered, reported, tagged, shared, cropped and more.
The future of AI is being shaped by what you see today. Or, rather, what the software can see.