This is mind-boggling. But perhaps not surprising. It’s another very large leap in AI/robotics/super-computing.

At Google, Tobias Weyand and his team have trained a deep-learning machine to work out the location of almost any photo using only the photo; only the pixels it contains.

They fed the neural network–which they’ve named PlaNet–126 million images along with their accompanying Exif location data. Based upon that info, when fed a new image, PlaNet analyses visual clues (buildings, topography, vegetation, weather, etc) to determine the most probable location.

“PlaNet is able to localize 3.6% of the images at street-level accuracy and 10.1% at city-level accuracy,” say Weyand and co. What’s more, the machine determines the country of origin in a further 28.4% of the photos and the continent in 48% of them.

So that’s 3 in every 100 PlaNet puts right at the front door. 1 in 10, it knows the city, and in over a quarter the neural network correctly identifies the country. Impressive.

What’s the future looking like? For the first time, I’m really fretting over the strength of the global human community to make the right–and ever-recurring–decisions on the ethics and control of AI. After a computer beat the world’s best Go player this month, the making of AlphaGo (another Google project) warned against hasty application of AI technologies. Stephen Hawking has been erring caution for decades. AI is still is a long way from applying learning from one system to another so we needn’t worry about self-driving cars learning to control markets or journalism robots changing careers and taking the reigns of power.

But who’s to say that in a few decades, you won’t be able to connect the automated detection of news events in, say, social media, to trigger the dispatch of drones pre-programmed to make photos (behind police tape/above breaking stories/inside toxic environs)? Then the images are sent images to the systems of robot journalists which in turn publish a story in a template. Possible? Maybe. Fine-tune the identifying capacity of PlaNet and you’ve accurate captioning info enough to furnish a news agency … and dispose of the photojournalist!?!

Such a scenario would take care of breaking news, but I’ll still wager on humans, not robots, to fashion the long-form documentary projects. Hell, by then, documentary photography stories might be one of the few things left connecting us!

Thanks Robert Gumpert for the tip off