A Thousand Words

Image recognition has long been a major challenge in computing. There was in fact an old saw that a computer could beat the world chess champion, but was unable to enter a room and recognize the chess board.

When you type into a computer, as I’m doing now, you create a connected system. My hands are the actuators linked to a computer (my brain) and the keyboard is the sensor linked to another computer (the laptop I’m tapping away on).

Computers can easily read my words back to me, and if this one is put in speech mode (secretary, please read out the memo I just dictated) then the connected system works the other way: the computer instructs its actuator (the microphone) to read this sentence, my ears are the sensors, and my brain processes the signal.

Computer now easily recognize text, and can fix your spelling and to some extent your grammar—they don’t deal particularly well with common sense. If you write ‘I just broke my heel’ as ‘I just broke my meal’, your computer will be perfectly happy.

In addition, the machine won’t understand that you’re probably talking about your shoe, and not a fracture. You can take this a couple of steps further. If you log on to your computer, the machine knows whether you’re male or female, and might wonder why a guy is wearing heels, or whether to summon an ambulance.

The same applies to images: if you post a picture somewhere, some clever assistant like Cortana or Siri (where do they get those names) might comment: that’s a really cute puppy—but they don’t.

The nine-eleven attacks (or seven-eleven in Trumpspeak) changed air travel and banking regulations forever, and marked the real beginning of government Big Data. From Harvard to Haifa, a quiet revolution was underway in image recognition.

It began with the installation of cameras and fingerprinting at airports, and continued voluntarily, as millions posted their faces, and those of their erstwhile friends, on Facebook, together with descriptions of lifestyle, locations, times, and habits—in short, a data harvester’s dream.

Algorithm using fast Fourier transform (FFT) to deblur text.

Algorithm using fast Fourier transform (FFT) to deblur text.

All that was needed was for the computer to recognize the chessboard. And that has come of age at last. You’ve seen the signs—iris recognition software, phone cameras that detect and recognize faces, software to remove red eye, recolor images, and perform sophisticated manipulation on color, hue, light, grain, and many other properties.

The example above is from code developed by Vladimir Yuzhikov, a Russian programmer focused, if you excuse the pun, on image processing—the software, released in 2012, can sharpen any image: buildings, text, or a face.

And this week a couple of articles appeared in the British press about an app called FindFace. It uses a Russian social media site, Vkontakte, and can match a face to a photo with a reliability of 70%.

The two newspapers involved, Telegraph and Guardian, are on opposite sides of the UK political spectrum—just read their views on Brexit. On this topic, though, they are agreed: it’s scary.

Findface uses a high performance algorithm and marries it to Big Data on Vkontakte.

FindFace uses a high performance algorithm and marries it to Big Data on Vkontakte.

The twin keys to success for FindFace are the combination of databases like Vkontakte, and of course FaceBook, with a very fast algorithm that can process a billion records in one second.

I imagine the algorithm works by zooming in on particular facial metrics that remain unchanged, and then refines the results. But it couldn’t do its job unless a site like Vkontakte existed, where two hundred million former Soviets come to air their dreams.

I don’t believe Mr. Sugarmountain cynically considered that Facebook would be the greatest police state in the world, and a voluntary one to boot. I really think it started as an easy way for people to share experiences, thoughts, and dreams—a way of cutting through the social awkwardness of pimply adolescents.

That was then, but he machine is now in relentless motion, and even if you don’t indulge, you’re on it through third parties.

When I google any name in Facebook, I see endless users—and about 70% have a good, clear photo—the kind of thing FindFace loves.

The app already has five hundred thousand users, and has handled three million searches. What are folks looking for? The details of a woman who took your fancy on the subway, girls that look like Angelina Jolie, whether a good-looking guy is married…

The list goes on, but it quickly gets more sinister, as details of a potential robbery victim emerge—wow, that’s an expensive car. Or someone who might be an assassin’s target.

Some of these Big Data aspects are part of my book Atmos Fear, but the world has evolved in the last three years, and the tools to control us are far more sophisticated today.

Of course, seeing the public use this app means that its covert usage has been around for a while. The company candidly states that if the FSB were to get in touch, they would definitely consider their offer.

But the FSB, like the CIA and the Mossad, have been at it for years.

The India Road, Atmos Fear, and Clear Eyes. QR links for smartphones and tablets.

The India Road, Atmos Fear, and Clear Eyes. QR links for smartphones and tablets.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: