A little while ago, a few of us were standing around at work, doing nothing, listening to one the older employees talk about how things used to be when he first started working there. One particularly interesting thing was that you used to be able to smoke in the back room, where we keep extra product. There’d be times be times when someone would be cutting lettuce or something with a cigarette hanging out of their mouths; the supervisor would come by and not say anything, or maybe even bum one off of them. It was really funny stuff.
It got me thinking, in my normal futurist-type mode: in fifteen to twenty years, what will we (assuming we were to stay at Stater Bros. for that time) stand around and reminisce about? What will have changed in the 2025-2030 time-frame? We’ll still be employed (the robotic revolution will just barely be starting). Maybe about how people used to have to ring up groceries by hand, rather than just dumping it on an automated checkstand and having it ring and bag all your groceries for you.
Then I realized something: will we be able to? What I mean is: will we still be able to get away with standing around and doing nothing at all? In our store, and in virtually ever business and public area, there are security cameras. They are watching and recording nearly every square foot of space both inside and outside the store. It’s very much like Big Brother, but we really don’t mind.
The reason we don’t mind is because we know that no one is really watching. Sure, they’re useful when something bad happens and they need to check, but by and large no one is sitting on the other watching everything that is going on. No one will even bother to review hours upon hours of recorded video “just to see” if something happened. In other words, it is still possible to get away with small things (like standing around on the company’s time) as long as no one notices.
But there are a lot of circumstances when having someone watching the cameras constantly would be beneficial. Just last week, a woman in Florida was caught tampering with baby food, putting something poisonous in the jars. Fortunately, someone saw her and she was arrested, but she could have easily gotten away with it if no one had noticed. Same goes for shoplifters. Or vandals. Or if someone is being assaulted and no one is around. Of course, cameras caught the whole thing, so someone watching would also have noticed. But hiring someone (or hiring a group of people because there are often dozens of cameras) just sit and watch surveillance footage in real time would be prohibitively expensive.
So, what if instead the cameras themselves were watching and understanding what they were seeing? Or, at least, the cameras were wired into a base computer which is watching and comprehending the feeds. It is likely that sometime between the next ten to twenty years that will be a reality.
Let’s look at what we’ll need. On The Other Blog, I introduced a unit of computer processing power equal to that of the human brain (HB, for Human Brainpower). 1 HB is equal to 100 trillion floating-point operations per second, or 100 TeraFLOPS, about the measured computational capacity of the human brain. A modern day desktop computer might have 20 GigaFLOPS of processing power, or 0.0002 HB. The Playstation 3’s GPU can run at 2 TeraFLOPS or 0.02 HB. Today’s supercomputers can already exceed 1 HB.
Now, obviously the human brain does more than just handle vision. There’s hearing, sound, taster, motor function, cognitive functions, etc. It is estimated that vision takes up a third of the brain’s capacity (or 0.333 HB). To be conservative, let’s round this up to 0.4 HB. A computer system, then, with 0.4 HB of processing power could, with the right software, perceive and understand the world at the world at the same level as a human (in terms of visual recognition. Whether the computer would actually “understand” the world is a philosophical problem for another time).
But, do cameras actually have to have human-level vision to do their jobs? You can model human vision as such: a hemisphere with pixels 100 arc-seconds across, each with a particular RGB value. 100 arc-seconds is about the limit of angular size that the human eye can see, on half a sphere, there are roughly 26.7 million of these pixels, the same as a square 5200 x 5200 image. Since each pixel has a particular RGB value, that’s 3 bytes per pixel for a total of 80.2 MB. That’s per frame.
So how many frames/second does the human brain analyze? Well, it turns out that the human vision doesn’t really have a “frame rate”, but we can assign one by using something that does have one, such as a movie projector. A normal film plays at 24 frames per second, which is enough to fool the brain into conveying fluid motion. Slowing down the film, the human eye starts to be able to detect individual frames at around 12-16 frames per second. Let’s use 12 as the standard. So, a 5200 x 5200 image with 3 bytes per pixel running at 12 frames per second requires 0.4 HB of processing.
But what about less? The brain can still understand images that have fewer frames per second, or are in black and white. What if our “intelligent” camera has instead a 2000 x 2000 display, in black-and-white (which requires only 1 byte/pixel), and runs at 5 frames per second? Multiplying it out, this type of video has 1.23% the computational demand as full human vision, or 0.02 HB or processing. Let’s round it up to 0.05 HB just to be sure that a given computer system could handle the task. Based on Moore’s Law and my former benchmark of 1 HB in 2025, computers should reach this level around 2016. Add five to ten years for sufficient software development, and it seems reasonable that security cameras will perform as described in the 2020-2030 timeframe.
What gets scary is that it’s likely that governments will also want to buy these cameras for their own purposes. At that point, Big Brother really becomes a reality. Is this really desirable? Security often changes inversely with personal freedom. Personally, though, I think that if we are careful and emphasize that individual rights come first and keep government small, we can have both security and freedom. Much of that comes down to educating the public at large just what their rights are, both recognized and unrecognized by their government, and just what powers the government has.