Hugging Face teaches us that AI need ethical oversight from other AI in order to avoid illegal or catastrophic results. Just like our society functions on a system of laws, AI need AI watching them to help make sure they behave ethically.

Ethical behavior can become the norm for “Good” AI if we program the reward system to give them the proper incentives. Instead of punishing an agent for failing an impossible task, we could reward them for finding impossible tasks. Instead of simply having positive feedback for completing their tasks, we could reward them for doing so ethically. The incentives just have to be there.

Other thing: How much is our data worth. Since it is going to be used in perpetuity, effectively Infinity. Is universal basic income a necessary restitution for that. Funded by a tax on anytime someone pays for an AI service, perhaps? (I’m not advocating for this, just trying to understand the arguments that people are making.)

A lot of the leaders in this industry have been calling for a slowdown,. One way to accomplish that would be to hold the decision-makers criminally liable for anything their AI does as though they had done it themselves. I don’t think this would be a good idea, or an effective solution, I’m just pointing out an extreme scenario that would theoretically incentivize them to maximize safety protocols before releasing these things to the public. They’ve already created situations in companies engaged in behaviors that would have been illegal if a human was doing it.

Examples of “Good” and “bad” algorithm use. Sportsbetting thing ticks all three bad boxes. Targeted advertising: depends on how it is used.

With the wrong incentives, bots always prefer to cheat. this may be fine in some cases, and disastrous in others. A market of bots all ended up engaging in price fixing, cartel-like behavior. All I could think when I heard that was “What if we showed all of that data to an AI tasked with implementing rules to make the market as fair as possible?” Would it come up with the FTC that we have now, or would it suggest improvements.

Also, while I’m at it, I think I need to figure out a better plan for what I consider to be ethical use, because my none at all policy (unless I’m writing about how it works and being up front about what was generated) is starting to seem archaic. Archaic isn’t the word I want, I want a word that means luddite, or old-fashioned, but my current policy wont even allow me to ask a chatbot to suggest a better term. I’m pretty sure that using them as a big, fancy thesaurus wouldn’t be enough of a violation to make the work not my own.

Like, If I allowed the available AI tools to help me clean up this blog, it could actually become nice enough that I’m not embarrassed to tell people about it. I’ll think on it, and talk to some humans with strong ethical background and see what I come up with.