Right now, video is the main input that computers have for understanding the physical world. Video takes in the trees and cars and people and turns them into data that software can analyze.
When there are cameras everywhere, there will be analytics everywhere. Vast new realms of life will be tracked and measured and optimized in the way that the web is now.
I don't worry about everyone livestreaming their neighbors' backyards—who wants to watch that?—but I do worry about applying statistical analysis to anything anyone points a camera at. Let me explain why.
Look at CamioCam, an app that debuted earlier this year that takes any smartphone or tablet and turns it into an always-on camera that uploads clips of anything interesting that happens in its field of view. These clips are then analyzed and can be searched. And it does all this for free.
"In the grandiose vision, we think this could be a Google for a real world," CamioCam founder Carter Maslan said when his company launched. "It could be ranging from when were elephants at a watering hole in Kenya, or when were the last five mail deliveries at your house."
I tried out CamioCam this morning. I set it up pointed out onto my street and let it start recording clips. We have a lot of pedestrian, bicycle, and car traffic outside, so there was plenty to see. I noticed in one of the earliest clips, a dogwalker was wearing a red beanie. So I wondered, if this is really Google for the real world, could I search for red hat, and have it show up?
Sure enough, the red-hatted-man clip showed up (along with one false positive). And CamioCam is even kind enough to highlight the man. And this is only the cheapest, easiest version of this technology.
The company 3VR offers a video analytics package that counts foot traffic, does rough demographics analysis on shoppers in stores, and identifies how long people dwell in certain spots. Another package lets security personnel identify faces and search for them across other video footage.
IBM sells software that lets camera owners search for people based on characteristics like hairline, clothing, and skin tone.
And if you want to see the leading edge of this kind of thing, IARPA, which is the advanced projects skunkworks for the spy agencies like the NSA and CIA, has been working on searching through massive volumes of YouTube videos in a project named ALADDIN. An IARPA research bot may very well have already spidered through something you've uploaded. In the test cases, they were looking for innocuous things like "parking a vehicle" or "playing fetch," but if they can find those things across YouTube, how much harder is "lighting a joint" or "drinking a beer while underage"? The IARPA funding, which was doled out to commercial and university researchers, has already generated more than 150 papers indexed by Google Scholar.
All these projects are going to make searching and analyzing the physical world, via video, much, much easier in the next couple of years.
The repercussions of this technical change are only beginning to be felt. And the problems are not exclusively, or even primarily, about NSA-style surveillance.
The standards of online data collection and analysis—the culture of COLLECT ALL THE DATA—is spreading to the physical world through these technologies. In online courses, all kinds of data is collected about student performance. If a student doesn't watch a lecture, the teacher (and the administration) know that. But in the physical world, the same standard doesn't hold. Not all professors take attendance in lectures.
But, as the Harvard faculty discovered a few weeks ago, using the tools of video analytics, professors don't have to call roll for heads to be counted. The school administration secretly took attendance in a small set of classes. University officials hadn't told anyone they were doing so, and no one asked students if they'd been to class. Instead, they set up a camera trained on the lecture hall, took the video, and used software to measure how many students were in seats. Harry Lewis, former dean of Harvard College and a computer science professor, spoke up at a faculty meeting on the issue. "Just because technology can be used to answer a question doesn’t mean that it should be," he said.
The Harvard administrator in charge of the project, Peter K. Bol, defended the project, saying, "I can report that every single person I met with thought the data was interesting and potentially useful."
And this is the crux of the problem: of course analyzing the physical world with our software tools will yield interesting insights. These are powerful techniques that can do things no human eye or brain can.
So maybe companies like CamioCam or 3VR will help you catch the dog that's shitting on your lawn or help colleges force more kids to attend lectures. But do we want to apply the logic of web tracking and measurement to the physical world? As Harvard's experience shows, there is no consensus that this kind of surveillance is a good thing, no matter its benefits.