I first wrote about facial recognition back in 2017, shortly after starting this blogue. (stderr) Then, (stderr) I discovered to my complete lack of surprise, that there is a lack of proper paranoia about the topic, and – I suppose that’s justified because climate change is likely to wipe us out, while having no more personal privacy, won’t. I guess we’re arguing about “insult to injury” which is a great way of spicing up your injuries. Add it to heirarchical storage and, whups, things look bad. (stderr) It seems like just the other day I was laughing at people who were implementing poorly-planned “data lakes” but it seems that, miracle, AI actually has stepped in to help with the problem.

So, one thing you need to understand about all those heat-spewing data centers: they are being built that way because it’s the lazy way to do it. That’s another issue, entirely, but it all comes down to being able to reliably select a bunch of stuff from a database. Back when I used to teach classes on that kind of stuff, I remember telling my students that the key was to find an “oracle” – a source of corroberation or correlation that is either incredibly accurate, or predictably accurate. (If your oracle is only predictably accurate, you simply assemble a panel of them into a voting architecture). There are still tons of ways of jamming the system’s input but the designers of the system have placed themselves across the vital pathways of modern civilization: you cannot use your credit card unless your face matches, you cannot board a plane unless your face matches. Add to that about a bazillion parking garages and basically anything that gets paid with a credit card, and you can correlate the payment information to a quick photograph. As I predicted, capitalists rapidly mooted any data privacy protections by simply offering lookup as a “cloud” service: an organization that can’t populate and train its own AI to swim in its own data lake, can just go to one of many companies that sell that information, no questions asked. “Who is this?” I guess, is the only question.

self-portrait 2017

When you hear the ICE goons brag about “we have a little database, too” what they are talking about is a massive cache of identity information. Here is how it goes: a bunch of face images are collected at a protest. Now, someone at an FBI-run “data fusion center” uploads the pictures and scrubs them for rapid identity. Basically, anyone who has already been to a protest and been photographed, will just be confirmed as still looking more or less the same. That starts a thread building in Palantir, looking for clusters of images that match repeatedly in different circumstances – aha, that guy who appears with Marcus frequently is his buddy, [redacted] who drives a Subaru and lives in the outskirts of DC. The presence of the Subaru is confirmed by tollbooth cameras. The parking lot cameras confirm that Marcus’ Toyota is also in the area. We now know, also, that drones are in use to monitor citizens – drones that can generate identity-capable images from miles away; that all goes into the data lake. The data lake becomes a cache – it’s all the people that the system has ever looked up, and all the places and times they have appeared. Mike and Tom, who have not appeared at a protest, do not appear in the data lake, either. Back in the day, when I was teaching people how to build these systems, I always explained it as a pyramid. The more layers you have, the more distilled and refined the data become, so you don’t have to search the whole data lake, or go to someone’s facial identification database, you’ve already got the distilled data. My guess is that, seriously, the system engineers who are building most of the systems out there never took any of my classes, so they don’t know about Bloom Filters or K-D trees or any of the really impressive techniques that you encounter past the heat-death of CS-101. I remember when people used to say “these systems are not real-time enough to be a threat.” Ha, ha, ha, they know who you are within less than a second, and the data returned includes a link to a complete dossier.

What’s really interesting, to me, about a lot of this stuff, is that it illustrates a paradigm in intelligence that I have discussed before – namely, the “retro-scope effect.” All this stuff can predict who may show up at the next protest, at best but not who’s going to throw a petrol jar. Yet. The AIs would certainly be able to scrub through a list of all the dossiers present at a given protests, and search for indicators in their private social media messages and texts that they are planning on bringing a weapon. How do we defeat this? We don’t. That battle is lost. The next battle is to make effectuating the information useless. How do we do that? Well, we need to get some people inside the system, I’m afraid. It wouldn’t be hard – get a job working at Palantir or FBI, NSA, or CIA. Sure, they’re going to do background checks, so that just means you can’t be someone they already have a dossier on. The East German Stasi had massive files on everyone and agents provocateurs out in the population, as well. Did any of it work? Nope. It snagged a few individuals, but probably ones that were being stupid. Let me use an example based on current events: you should not applaud it as a great break-through if your security process identifies Chud the Chud Whatever as a “possible trouble-maker.” Especially now, as he has made some trouble! There’s a slightly subtle point there, which is that if you ask an AI “Who is most likely to cause trouble at a rally?” you’re going to have to weed out a few answers that are 100%. OK, where do you stop weeding?

[I asked GPT to generate a succession of increasingly less likely to be recognized images, based on what it knows of what parameters the current face recognition software looks for]

The way to fix all this crap is to just wait for the state to topple from its own incompetence. It all boils down to variations of the force protection problem: the police will insist on being able to beat protesters without recrimination, which means that now you need to peel off some of the military to protect the police, and then how do you protect the military? The insurgent’s problem is to attack the roots of the supply chain and the roots of personnel management. For databases, you need to feed them false data. Some hero of the revolution has to get a job at one of the fusion centers and start back-feeding photos of congresspeople into the watch-lists. You maybe don’t need to be at a fusion center; based on my experience of how bad government security around computers typically is, a janitor in a DHS office or a fusion center could pretty easily access the system. Here’s another hint: airport consoles are totally ignored during non-office hours; sneak around and add keystroke loggers to the DHS scanning stations. I have investigated this likelihood several times at several airports. But, honestly, it’d be more fun to get a job working for DHS, get paid $50,000/year and get free training on the software, medical plan, vacation, etc. The current generation of kids are going to have to learn to be stainless steel rats, creeping in the walls of the system and giving it a bellyful of chicken with a little poop stirred in.

[what is really scary is the realization that moderately overweight bald white guys aged around 60, look very much the same when bald. Think about that: it tells you something about the weight our own algorithm puts on hair.
I asked GPT to make me an updated version of my early effort with tape, except designed to error out the recognizers.]

I guess my view of these things is leaning toward expecting the current generation to be prepared to fight for its life. And the only thing I think I have learned, from years of thought and study, is that the way to defeat systems is by becoming part of the system. If you want to, consider the republicans’ efforts at taking over the government to be a case study in exactly what I am talking about. Just remember that they are interested in destroying institutions, where as in the future our project will be to drive them insane.