How can policy keep up with AI advances?
80000hours.org
80000hours.org
It forced people to think more clearly, and it made the models used sound to normal people like boring applications of numerical optimization, which is pretty accurate. It's not something a normal person should feel any sense of excitement about, any more than they should care about a new type of satellite dish or whatever -- you have to be a pretty odd duck to find it intrinsically interesting.
Sadly, the decisions around AI will likely be based on those pieces of fiction. The average person has spent more time watching movies about AI rather than reading actual research material on it, and will vote accordingly. We should really refrain from using "AI" as a term until artificial general intelligence is actually upon us. Using it to refer to GANs or Deep Convolutional Networks is just feeding the news beast.
In addition to changing the terminology though, I think it's also important to change the fiction as well. Project Hieroglyph[1] is one of my favourite efforts in the realm of fiction - it aims to combine scientific papers with fictional stories that represent those advances in a clear and accurate way. If people are going to make decisions based on fiction, we should at least keep the fiction accountable to being realistic and accurate.
There are very basic functions that the legislative power is still missing. Still, they want to regulate on super advanced topics up to the most specific details.
That's why you should minimize top-level regulation (e.g. EU, US level). And make regulation as local as possible.
Localizing regulation allows for faster changes and better adapted regulations to what people want. It also allows for A/B testing regulation, before rolling it out to the masses.
Take health care as an example. The US has a lot of red and blue states. Most people in red states don't want public health care, most people in blue states want public health care.
On the other hand the more people you have in a health care system the better the risk pooling. So you don't want it to be too local (e.g. district level).
So for effective regulation the federal level is wrong. The federal level should neither make a public health system nor interfere with lower level decisions to have one. But states might be a good level to start with.
That way you could have public health care in blue states quickly. The red states would stay out of it. This would also allow for better A/B Testing. Texas might see their neighbors in California faring better or worse now and then either Texas or California may change their mind about it.
This would even make it possible that the public insurer in Vermont is maybe more efficient than the one in California. Now if you allow people in California to buy Vermont public insurance, you could even create competition that keeps the public system more healthy.
It would make getting a majority easier. Because the more local you get, the easier it is to make decisions.
And if you say this is too much load on local regulators. My proposal would be that the federal level provides a default, and states can override that default. Cities/Counties can override state defaults, etc.
This way you always get the most accurate. Most democratic decision.
This would also limit the effectiveness of lobbying, because the price of bribing more and more local politicians would become ineffective in many more cases.
Nothing is more democratic and stable than a system where bad agents only need to convince one podunk town which can locally supercede federal legislation to legalize kickbacks and completely break everything.
The more local you go with your bribery, the more money you need to spend on individuals and the less influence you buy.
Right now it's perfect for bribery, where you only need to bribe a select few and you can destroy a whole country.
And those lawyers and governors haven't gotten it wrong, they get those bribes they want and don't need to really answer to their voters. Perfect. For them.
There are plenty of fraud reporting (not small scale stuff, we're talking commissions, etc.) that describe how municipalities are functionally captured. It is done today, but on a scale far smaller than what your plan allows. Also, since the stakes are smaller, the oversight is smaller, which means there's more, but less impactful fraud at that level. With your plan, we'd have more, but now potentially even more impactful fraud.
You seem to believe you need to bribe ALL municipal-level politicians to get what you want. That's not the case. One jurisdiction is enough for most ends. As for price, you've just created competition for corruption among every county. That drives prices down, not up.
Now that the attack surface is huge and the prices have plummeted, what do you think happens next?
This is one counterargument, but there's a ton more. In short, every co-ordination problem now has no venue to be solved unless it falls under your umbrella of 'big, unchanging things', and sadly a lot of co-ordination problems just plain aren't that simple.
As a bit of a double whammy, co-ordination problems underlie a large portion (if not the overwhelming majority) of our issues in social systems. Removing the government's ability to act on them is fatal to the government's ability to govern.
Were you trying to say that we are going to be in trouble with the plebs if we don't manage to create an all powerful AI for them to treat like God?