Excerpts from an interview with prolific author and technology freedom advocate Cory Doctorow.
I don’t like the term artificial intelligence. It is neither artificial nor is it intelligent. I don’t even really like the term machine learning. But calling it ‘statistical inference’ lacks a certain je ne sais quoi. So, we’ll call it machine learning, which is best understood as allowing for automated judgement at a scale that human beings couldn’t attain. So, if you want to identify everything that is face-shaped in a crowd by looking through a database of all the faces that you know about, a state’s ability to conduct that would be constrained by how many people they had. The former East Germany had one in 60 people working in some capacity for the intelligence services, but they couldn’t have come close to current scales of surveillance.
But that brings up a couple of important problems. The first is that it might work, and the second is that it might not. If it does work, it’s an intelligence capacity beyond the dreams of any dictator in history. The easier it is for a government to prevent any opposition, the less it has to pay attention to governing well to stop opposition from forming in the first place. The cheaper it is to build prisons, the fewer hospitals, roads, and schools you need to build and the less you have to govern well and the more you can govern in the interests of the powerful. And so, when it works, it’s bad.
And when it fails, it’s bad because it is by definition operating at a scale that’s too fast to have a human in the loop. If you have millions of judgements being made every second that no human could ever hope to supervise, and if there’s only a small amount of error, say it’s 1%. Well, 1% of a million is 10,000 errors a second.