There's A Lot of Human in Your Artificial Intelligence
NBC has a nice look at the people who make artificial intelligence possible. No, not the programmers or engineers or data scientists. The people who classify the data:
Savreux is part of a hidden army of contract workers who have been doing the behind-the-scenes labor of teaching AI systems how to analyze data so they can generate the kinds of text and images that have wowed the people using newly popular products like ChatGPT. To improve the accuracy of AI, he has labeled photos and made predictions about what text the apps should generate next.
The pay: $15 an hour and up, with no benefits.
The reason large language and picture models and other systems of their ilk can work is twofold. One, they have an enormous amount of data to train on, making their algorithms more effective. And two, they have human beings telling them what is right and what is wrong. Now, classified learning is not the only way to go -- but it is likely the most effective and it certainly seems to be the one that most of these systems rely upon, for good and ill.
Good, because these are still jobs that humans do exceptionally well -- classifying unknown information appropriately based on past experience and conceptual clues. Bad, because these are not considered to be important jobs by the people who create AI systems. The jobs do not pay well, they do not seem to involve significant training, and given the number of people involved and the amount of outsourcing used there are serious privacy concerns.
You have people who are being asked to properly classify material that aren't being trained on tricky questions of fact or interpretation for systems that we know are already being used to spread massive amounts of disinformation. And they potentially have access to people's private data, since that information is collected by these systems if it's entered into them. All of that is supposed to be handled by outsourced people making low wages with uncertain oversight. This is all very similar to how social media companies handled moderation -- outsource, pay the people who did as little as they could, and then when they were traumatized by seeing the worst people had to offer day after day, discard them. You fancy algorithm is very often driven by some form of human misery.
So far, so capitalism. What's the point? When we talk about regulating AI, we must remember that we are not just talking about the algorithms and systems. We are also talking about the people who do the grunt work and the impact their jobs have on our data and their wellbeing. Any regulations that do not protect those people and ensure their work has standards that protect our privacy is going to fall short of what society really needs around AI.