AI is not very regulated. For instance, Max Tegmark notes that AI is less regulated than a sandwich shop in the US. This seems kind of wild for an industry where most relevant scientists think there is a serious chance of it destroying humanity. Even for something that occasionally encourages children to kill themselves. Shouldn’t it at least be as regulated as paroxetine?
The kinds of things I hear people say about why it shouldn’t be regulated I think mostly apply to everything—’technology is usually good on net’, ‘the government might not be able to regulate it well’ etc. So why in fact is this so unusually unregulated?
I speculate that the real reason AI isn’t very regulated is that it’s on computers, and regulating the motion of numbers on computers feels unnatural to most people. We have already seen that the world of software and the internet is a much freer place than the physical world. That’s one thing I like about it—I can just set up shop and write blog posts without even having to acquire a blogging permit. People have intuitions that ideas, information, thoughts should be free, and aren’t really real anyway, and can’t really do anything on their own. Are you seriously going to regulate the doing of math? Even wanting to seems kind of absurd—both hopeless and overcontrolling.
But if you do work in a virtual world, you do work in real life. These intuitions of the unreality and passivity of virtual things are wrong, and once the ‘ideas’ and ‘information’ become elaborate calculations that change the real world, their wrongness starts to get intense. We already live in a reality where physical things constantly move as a result of virtual activity. And while we’ve regulated a few virtual areas for this reason (for instance things that involve money or Europe), I’d say we haven’t reckoned with this much yet: the virtual worlds we have created in recent decades remain a wild west, with the help of our intuitions that they should be.
As David has pointed out to me: imagine if companies were breeding new biological entities that were already as capable as humans at many tasks—would the world be responding in remotely the same way? Would that be unregulated? Why is it different on computers?
I think this might change soon—as the impetus for important real world occurrences starts to come from patterns on computers, I think virtual worlds might rapidly start to feel like the proper realm of government control. Ideally we would make that adjustment before we start to pay a lot of costs for the error.
GPT 3 came out in 2020. I think it is the default for six year old technologies to be completely unregulated, and does not require any specific explanation.
This seems like it’s ignoring the huge debates about regulating numbers on computers, i.e., DMCA circumvention software, the “Crypto Wars” and so on.
Yep. The idea that software can be covered by freedom of speech was an important development both in law and in Internet engineering culture. And speech is what chatbots do all day anyway. Regulation of AI necessarily touches on free-speech issues.
Yes. Governments can and do regulate data on the Internet. It’s a controversial issue. See also The Closure of the Internet for a detailed discussion of Internet regulation.
Blogging permits are closer than ever thanks to age verification laws. Over the past decade, those have been implemented around the world despite the obvious risk to freedom of speech. (Tying social media registration to government IDs, credit cards, and facial scans is a big boon to catching dissidents and creates a chilling effect.)
I think we should discount general unwillingness to regulate computers and the Internet.
Policy makers are not forward thinking regarding technology, and humans are very bad at predicting the future. “The fire code is written in blood”.
I think regulation can usually be analyzed as, how did this facilitate business or economic growth. Sandwich regulation makes customers feel safe at miscellaneous restaurants and grocery stores. When it becomes hard to do the AI business without regulation, the industry will call for regulation.
I think a lot of people, including in government, think it is snake oil. they last tried a chatbot > 6 months ago and were unimpressed. it is still the basis of their understanding. to them it is a toy, a curio. regulating it would be like regulating a circus or a science fiction movie. this disconnect is only going to get worse as the rate of improvement increases.