Artifical intelligence is not actually intelligent, and OpenAI is not particularly open about how they operate.
The 1986 movie, Short Circuit, teaches us at least three things about the era:
- People in power did not yet understand that having a white actor play an Indian was a very bad idea.
- They did know that machines would be able to “learn” much more quickly than a person could teach.
- Steve Guttenberg really was everywhere.
The capacity of modern AI language models to absorb input is vastly beyond what Johnny 5 could ever have dreamed of. However, AI is not sentient. It does not, can not, know anything. So it absorbs all of the racist, sexist, immoral, horrible things right alongside the best of our species. And has no way to know which is which.
Companies have learned (after a few failed attempts) that they need to at least try not to replicate the worst of humanity at machine scale.
The answer to that dilemma, so far, has been paying underpaid, under-protected people to be human filth filters, reading and watching horrendous material so that the rest of us can experience the nice Disneyfied “magic” of generative AI. Just don’t think too hard about how it works. Quick, send Nick Cave another “Nick Cave” song.
This is not a new story, just the latest reboot. Some of Facebook’s outsourced content moderation teams have at least received a settlement for their own trauma. While some people are trying to teach the machines how to act like a decent person, Amazon (among others) is trying to turn people into robots.
In every case, the secret of their success is in hiding the reality of how things are done. As long as we don’t have to see it happen, we don’t have to think about what it means. Pay no attention to those behind the curtain, because it might stop you clicking that ‘buy’ button.