Idiotic Bigotry Machine

There's a lot of discussion in tech at the moment about artificial intelligence becoming racist. The argument you will often hear is along the lines of "racist programmers = racist AI", which has some validity to it, but is way over-simplified—and in my opinion—masking a bigger problem. But before I get to that, I'd like to mention two other problems that are acting as contributing factors.

1) Blind faith in systems you don't understand. The people who market this kind of technology obviously want to hype it up like it's completely infallible, and the people who buy, and operate it, often don't have a fuckin' clue how it works—or for that matter—how, when, or why it doesn't work. It's just a magical black-box solution, and if it seems to work perfectly 90 or 95 percent of the time, why wouldn't you just assume that it works perfectly 100 percent of the time? Automotive-technicians are the worst. I predict we'll see a time when your car can be on fire, but if the diagnostic readout from the computer says everything is fine, the technician will stand there and tell you, in all seriousness, that your car is definitely NOT on fire.

2) The tech-press are a mob of sycophantic arse-lickers. Coverage around the advancement of self-driving cars has been a fine example of this. I can't tell you how many articles I've seen and heard, where the "journalist" has bluntly stated that "self-driving cars WILL dramatically reduce traffic accidents". Absolute certainty. No ifs or buts about it. And this is despite the fact that there are entire podcasts—ENTIRE WEBSITES—that cover nothing but the real-world problems caused by software bugs. Somehow these cunts have miraculously divined that self-driving cars are going to be the one field of technology where code will be completely and utterly error free. Oh, and keep in mind that all of their information is coming from internal testing done by the people who are at some point going to be selling this stuff.

I don't know if it's coming through, but this is one of those topics that absolutely grinds my gears.

Anyway, on to the subject at hand …

At this point in time, the term "artificial intelligence" is utterly meaningless. As soon as the phrase started to gain traction in the media, there was a mad scramble to label anything and everything as "AI". 99% of it is no more fucking intelligent than an air-conditioner with a thermostat in it. "Oh look, it knows when to turn itself on and off, how intelligent". What a lot of the technology is about is really really advanced pattern-recognition (and yes, we can talk about how much of human intelligence is based off of pattern-recognition if you like). You take a big heap of data and feed it into the system, the system sifts through it, searching for patterns, and then it uses statistical averages to try and determine an optimal choice for a specific scenario. And here's where we run into the issue of the racist AI.

Let's just say you feed a whole bunch of demographic information into one of these things. It's going to notice that different races have different averages when it comes to income, education, employment, criminality, incarceration, health, nutrition, lifestyle, and everything else. And it's not just race either; these things are going to be judging people against all of their demographical averages—height, weight, age, gender, sexual-orientation. So, why is this potentially a bad thing? Let's just say you have one of these systems set up for recommending career paths to people. It chomps on the data and sees that amongst teachers, the single biggest trend is the number of white women from middle-income families working in the profession. This then becomes the most heavily-weighted factor when recommending teaching as a career path. Over time, this skews the numbers even further, which creates even greater justification for sending even more middle-class white women into teaching. It's probably not the greatest example, but it should give you some idea of how these things can devolve into self-reinforcing feedback loops. By making recommendations based on averages, you often end up incrementally pushing things even further towards those averages.

As the technology progresses, and we integrate more and more of this crap into our daily lives, there's no telling how much of what we do is going to be influenced by these algorithmic recommendations. From jobs to relationships, right down to what we eat, wear, and watch on telly. And of course nobody's ever going to question the answers the magic box spits out. Why would you? After all, it's always been right in the past, hasn't it?

Oh, and that's not even touching on what happens when people start intentionally biasing the AI to push a specific product or agenda. Fun times ahead.