Hey Gabe! Emanuele here 👋. Do you have any sense for what AI regulation could (or should) look like? For example, I’ve asked Duck.ai this very question and it didn’t mention anything about unemployment, which as you mentioned is likely to be the hottest topic, and it also seems the hardest to regulate (balancing the Great Race 🚣 vs job security 👨💼).
I'm not sure yet, but I want to explore more and so thanks for the push to do so. At a high level, each of the listed harms--jobs, privacy, inaccurate information, loss of human connection, the environment, and bias--could ultimately have a regulatory component in addition to safety/security, but I agree jobs so far hasn't gotten much attention in terms of regulatory response.
I think the potential for disruption on job is large enough that it warrants thinking about it early, i.e. now. As for what to do exactly, I plan to dig into that more and come up with some concrete ideas, but for starters, the recommendations from the 1960s report referenced above aren't bad: "income guarantees, relocation assistance, federal unemployment benefits, education subsidies, and government jobs."
People will need assistance and we should encourage retraining. Trouble is assistance is expensive and retraining generally hasn't worked in the past. However, I'm a little more hopeful here for a few reasons: (1) AI promises a lot of economic growth and so there should be extra money; (2) I don't think we've ever done retraining quite right, for example it shouldn't be government retraining programs but probably tuition subsidies for vocational programs as part of unemployment benefits; (3) there is an opporunity for using AI for retraining though that is more speculative and not sure how that would work yet exactly.
Hah, I put it in the caption, but that’s easy to miss. It wasn’t particularly good at this task or I wasn’t particularly good at doing it, as it took many iterations to get this and then I gave up making it better from here because every time I tried it got worse in some other dimension.
I think protests have become more ineffective over time. We started to see that in 2011 with the Arab Spring. People were saying it was Twitter that enabled those, but I would argue on the other side that extreme transparency from social media makes it much more difficult for leaders to continue a movement once it starts. It is too easy to tear it down by millions of paper cuts, and it fizzles out.
We've seen that over and over again the last 15 years.
I haven’t seen anything specifically about the effectiveness of protests over time, but I think they haven't been that effective overall, even historically. I think the best I've seen is the quote I inserted above, which was from a 2018 study looking back 30 years, but it doesn't look they studied the time variable.
That said, my argument here is, essentially, that while they are usually ineffective, if the ingredients are just right, then they can be effective, including in the present, and those ingredients might end up being just right for the case of AI and jobs: truly bipartisan (wheras most protests are partisan), directly affecting a significant amount of the population (wheras most protests are indirect), and affecting most demographic groups (wheras most protests are more narrow). The other ingredient is time though, whether a large effect happens over a shorter period of time like a few years vs. diffused over several decades.
Corroborating your feelings about how protests in America are largely ineffective, I am reminded of Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens, a study which found that average American citizens have very little influence on US policy between 1981 and 2002. It instead found that only economic elites and interest groups had significant influence, with economic elites having the most. Since protests are one of the ways average citizens attempt to influence policy in America, the study implies that protests are mostly ineffective.
Because I think it is likely that economic elites will oppose the reforms needed to help Americans deal with AI related job losses, I don't think anti-AI protests will be successful. Moreover, I believe that AI will reduce the amount of jobs in the economy since I don't think an AI powered economy will grow fast enough to compensate for job losses.
My personal worst case scenario is that a very large amount of job losses occur and most Americans become destitute, leading to a protest movement that turns violent. This violent rebellion is put down through the use of autonomous drones. After the failed revolt, economic elites start believing that the rest of the population is a unnecessary liability and act accordingly.
I have a similar fear and have been searching for the right historical analogy. I think the closest thing I've come up with so far (which I plan to write up soon) is the yellow-vest protests in France from a few years ago (https://en.wikipedia.org/wiki/Yellow_vests_protests), which did turn violent, though not to the degree of the Luddite movement.
Hey Gabe! Emanuele here 👋. Do you have any sense for what AI regulation could (or should) look like? For example, I’ve asked Duck.ai this very question and it didn’t mention anything about unemployment, which as you mentioned is likely to be the hottest topic, and it also seems the hardest to regulate (balancing the Great Race 🚣 vs job security 👨💼).
I'm not sure yet, but I want to explore more and so thanks for the push to do so. At a high level, each of the listed harms--jobs, privacy, inaccurate information, loss of human connection, the environment, and bias--could ultimately have a regulatory component in addition to safety/security, but I agree jobs so far hasn't gotten much attention in terms of regulatory response.
I think the potential for disruption on job is large enough that it warrants thinking about it early, i.e. now. As for what to do exactly, I plan to dig into that more and come up with some concrete ideas, but for starters, the recommendations from the 1960s report referenced above aren't bad: "income guarantees, relocation assistance, federal unemployment benefits, education subsidies, and government jobs."
People will need assistance and we should encourage retraining. Trouble is assistance is expensive and retraining generally hasn't worked in the past. However, I'm a little more hopeful here for a few reasons: (1) AI promises a lot of economic growth and so there should be extra money; (2) I don't think we've ever done retraining quite right, for example it shouldn't be government retraining programs but probably tuition subsidies for vocational programs as part of unemployment benefits; (3) there is an opporunity for using AI for retraining though that is more speculative and not sure how that would work yet exactly.
It is not lost on us that the image strongly appears to be AI generated.
Hah, I put it in the caption, but that’s easy to miss. It wasn’t particularly good at this task or I wasn’t particularly good at doing it, as it took many iterations to get this and then I gave up making it better from here because every time I tried it got worse in some other dimension.
I think protests have become more ineffective over time. We started to see that in 2011 with the Arab Spring. People were saying it was Twitter that enabled those, but I would argue on the other side that extreme transparency from social media makes it much more difficult for leaders to continue a movement once it starts. It is too easy to tear it down by millions of paper cuts, and it fizzles out.
We've seen that over and over again the last 15 years.
I haven’t seen anything specifically about the effectiveness of protests over time, but I think they haven't been that effective overall, even historically. I think the best I've seen is the quote I inserted above, which was from a 2018 study looking back 30 years, but it doesn't look they studied the time variable.
That said, my argument here is, essentially, that while they are usually ineffective, if the ingredients are just right, then they can be effective, including in the present, and those ingredients might end up being just right for the case of AI and jobs: truly bipartisan (wheras most protests are partisan), directly affecting a significant amount of the population (wheras most protests are indirect), and affecting most demographic groups (wheras most protests are more narrow). The other ingredient is time though, whether a large effect happens over a shorter period of time like a few years vs. diffused over several decades.
Corroborating your feelings about how protests in America are largely ineffective, I am reminded of Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens, a study which found that average American citizens have very little influence on US policy between 1981 and 2002. It instead found that only economic elites and interest groups had significant influence, with economic elites having the most. Since protests are one of the ways average citizens attempt to influence policy in America, the study implies that protests are mostly ineffective.
Because I think it is likely that economic elites will oppose the reforms needed to help Americans deal with AI related job losses, I don't think anti-AI protests will be successful. Moreover, I believe that AI will reduce the amount of jobs in the economy since I don't think an AI powered economy will grow fast enough to compensate for job losses.
My personal worst case scenario is that a very large amount of job losses occur and most Americans become destitute, leading to a protest movement that turns violent. This violent rebellion is put down through the use of autonomous drones. After the failed revolt, economic elites start believing that the rest of the population is a unnecessary liability and act accordingly.
Link to "Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens": https://www.cambridge.org/core/journals/perspectives-on-politics/article/testing-theoriesof-american-politics-elites-interest-groups-and-averagecitizens/62327F513959D0A304D4893B382B992B
I have a similar fear and have been searching for the right historical analogy. I think the closest thing I've come up with so far (which I plan to write up soon) is the yellow-vest protests in France from a few years ago (https://en.wikipedia.org/wiki/Yellow_vests_protests), which did turn violent, though not to the degree of the Luddite movement.