On “End high-skilled immigration programs”: The thing about big-brained stuff like this is it very rarely works. Consider:
What is p(doom|immigration restrictions)-p(doom|status quo immigration)? To that end: might immigration be useful in AI Safety research as well?
What is E[utility from AI doom]-E[utility from not AI doom]? This also probably gets into all sorts of infinite ethics/pascal’s mugging issues.
How likely are you to actually change immigration laws like this?
What is the non-AI-related utility of immigration, before AI doom or assuming AI doom never comes?
What other externalities might exist from trying to get involved in immigration politics?
After doing all these calculations you will almost assuredly get a value less than intervening in politics to tackle AI Safety a different way.
The other stuff seems more reasonable but if you’re going to restrict immigrants’ ability to work on AI you might as well restrict natives’ ability to work on AI as well. I doubt that the former is much easier than the latter.
“The other stuff seems more reasonable but if you’re going to restrict immigrants’ ability to work on AI you might as well restrict natives’ ability to work on AI as well. I doubt that the former is much easier than the latter.”
This part of your comment I disagree on. There are specific provisions in US law to protect domestic physicians, immigrants on H1B visas have way fewer rights and are more dependent on their employers than citizen employees, and certain federal jobs or contractor positions are limited to citizens/permanent residents. I think this isn’t outlandish, but certainly not hard.
The end of high-skilled immigration won’t happen, I agree. Even when RW populists actually win national elections, they don’t do this.
If AI doomers think the expected harms of AI are too low to justify even temporary tweaks to US immigration policy, that suggests the risk of AI killing us all isn’t that high.
Focusing on immigration isn’t a clear win in that it would require the expenditure of political capital/lobbying resources and potentially burn a lot of credibility among the Democrats.
But I think the deeper issue is that this doesn’t seem like a good way of identifying the truth. I guess maybe you could make the argument that if the doom worldview suggests that we should make immigration changes and people with that worldview irrationally reject it, then maybe we can be a bit more skeptical of their reasoning abilities in general.
However, given basically any group, you can find one thing they are irrational about and then try to use this to discredit them. So this isn’t a very reliable method of reasoning.
When I was writing this- and when I think about AI Risk in general- as someone without a ML background, I tend to fall back on looking for non-technical heuristics like interest rates/market caps of hardware companies. So I am influenced perhaps more than a more technical person would be by these kind of meta or revealed preferences arguments.
I think Democrats (and left-wingers in other countres) could embrace increasing high-skilled immigration in ways that steer talent away from AI. In the US. H1-B visas could be changed to not permit work on AI or types of AI, and federal funding of science could steer people away from AI. So I think there is a path for Democrats to use immigration to reduce AI risk. The right could potentially use all three tactics, I think.
I guess my perspective is that all that these revealed preferences show is that people prefer to maintain their social status (benefit accrues to them personally) rather than support an unpopular change that is extremely unlikely to happen and where their support is extremely unlikely to make a difference (benefits are distributed).
So even if I accept this method of finding truth, it actually shows less than it might appear at first glance.
What about increasing immigration to the US across the board, especially low-skill migrants and high-skill migrants outside the tech sector? Development economist Lant Pritchett has written (in 2017 and in 2023) that migration restrictions incentivize the creation of more labor-saving technologies than would be necessary in a fully open world:
There is no global scarcity of people who would like to be long-haul truck drivers in the United States, where the median wage for such work is $23 per hour. In the developing world, truck drivers make around $4 per hour. Yet firms cannot recruit workers from abroad even at the higher wage because of restrictions on immigration, so business leaders in the United States are impelled to choose machines over people and eradicate jobs through the use of technology. (“People Over Robots”, 2023)
I spent some time last summer looking into the “other countries” idea: if we’d like to slow down both Chinese AI timelines without speeding up US timelines, what if we tried to get countries that aren’t the US (or the UK, since DeepMind is there) to accept more STEM talent from China? TLDR:
There are very few countries at the intersection of “has enough of an AI industry and general hospitality to Chinese immigrants (e.g., low xenophobia, widely spoken languages) that they’d be interested in moving” + “doesn’t have so much of an AI industry that this would defeat the purpose” + “has sufficiently tractable immigration policy that they might actually do it.” GovAI did some survey work on this. Canada, Switzerland, France, Australia, and Singapore looked like plausible candidates (in order of how much time I spent looking into them).
Because this policy might also draw migrants who would have instead moved to the US, the question of whether this is a good idea hinges in part on the question of whether overall timelines or the West-China gap is more important (as is often the case). I think recently consensus has moved in the “timelines” direction (in light of very fast Western progress and the export controls and domestic regulation likely slowing things down in China), making it look more appealing.
Happy to share my (now somewhat outdated) draft privately if people are curious.
I am an immigrant (F1) in the US (and I would like to think of myself as ‘high-skilled’ although others may disagree!). So, I am clearly biased.
It all sounds good in theory but it would be hard to assess what counts as AI-related work and what doesn’t. Even just a simple regression can be called “Machine Learning”. Economists have to do regressions basically everyday. In that case, they would have it in their resume and then someone in the VISA office would still reject them because they would have to assume ‘Regression means AI’ and your plan to let the “Oxford-educated economist” into the US would still be foiled. Also skilled economists could just work at tech firms and help them make money which they could then invest in developing AI further.
PS:- I downvoted this post. As I said, I am biased. But I saw very little effort from the author to actually provide references/evidence to substantiate some of the implied causalities. For instance, the author says things like “This policy is unattractive if you think”. IMHO a better post in the “Effective Altrusim” forum (which is based on ‘evidence and reasoning’ and not just what people like to think) would be link to some empirical work that actually shows firms don’t simply just shift if you restrict high-skilled immigration. Here is a study that claims the contrary that indeed immigration restrictions on high-skilled workers simply moves operations offshore: Glennon, B. (2023). How do restrictions on high-skilled immigration affect offshoring? Evidence from the H-1B program. Management Science. https://economics.nd.edu/assets/361162/bglennon_immigration_offshoring.pdf
My impression is that immigration policy is unusually difficult to effect given how much of a hot-button issue it is in the US (ironic, given your forum handle). So while the scale may be large, I’m skeptical of the tractability.
On OpenPhil’s behavior, yeah, if they’re making it much easier for AI labs to hire talent abroad, then they’re doing a mistake, but that path from all-cause increases in high skill immigration to AI capabilities increases has enough noise that the effects here may be diffuse enough to ignore. There’s also the case that AI safety benefits a bunch from high skill immigration as well. Indeed, many people I regard as particularly impactful in the field have been kicked out of Berkeley due to visa issues.
Potential area of fruitful cause prioritization research if you have the time and motive to do the math! The particular difficulty I see is there may be no advocacy groups who’d like greater high skill immigration in all areas other than AI research, except safety research. So if the cost of starting such a group or giving extant groups ultimatums is too high, then OpenPhil will be faced with a tricky decision.
On “End high-skilled immigration programs”: The thing about big-brained stuff like this is it very rarely works. Consider:
What is p(doom|immigration restrictions)-p(doom|status quo immigration)? To that end: might immigration be useful in AI Safety research as well?
What is E[utility from AI doom]-E[utility from not AI doom]? This also probably gets into all sorts of infinite ethics/pascal’s mugging issues.
How likely are you to actually change immigration laws like this?
What is the non-AI-related utility of immigration, before AI doom or assuming AI doom never comes?
What other externalities might exist from trying to get involved in immigration politics?
After doing all these calculations you will almost assuredly get a value less than intervening in politics to tackle AI Safety a different way.
The other stuff seems more reasonable but if you’re going to restrict immigrants’ ability to work on AI you might as well restrict natives’ ability to work on AI as well. I doubt that the former is much easier than the latter.
“The other stuff seems more reasonable but if you’re going to restrict immigrants’ ability to work on AI you might as well restrict natives’ ability to work on AI as well. I doubt that the former is much easier than the latter.”
This part of your comment I disagree on. There are specific provisions in US law to protect domestic physicians, immigrants on H1B visas have way fewer rights and are more dependent on their employers than citizen employees, and certain federal jobs or contractor positions are limited to citizens/permanent residents. I think this isn’t outlandish, but certainly not hard.
The end of high-skilled immigration won’t happen, I agree. Even when RW populists actually win national elections, they don’t do this.
I agree with this up until:
Focusing on immigration isn’t a clear win in that it would require the expenditure of political capital/lobbying resources and potentially burn a lot of credibility among the Democrats.
But I think the deeper issue is that this doesn’t seem like a good way of identifying the truth. I guess maybe you could make the argument that if the doom worldview suggests that we should make immigration changes and people with that worldview irrationally reject it, then maybe we can be a bit more skeptical of their reasoning abilities in general.
However, given basically any group, you can find one thing they are irrational about and then try to use this to discredit them. So this isn’t a very reliable method of reasoning.
Thanks for the feedback.
When I was writing this- and when I think about AI Risk in general- as someone without a ML background, I tend to fall back on looking for non-technical heuristics like interest rates/market caps of hardware companies. So I am influenced perhaps more than a more technical person would be by these kind of meta or revealed preferences arguments.
I think Democrats (and left-wingers in other countres) could embrace increasing high-skilled immigration in ways that steer talent away from AI. In the US. H1-B visas could be changed to not permit work on AI or types of AI, and federal funding of science could steer people away from AI. So I think there is a path for Democrats to use immigration to reduce AI risk. The right could potentially use all three tactics, I think.
I guess my perspective is that all that these revealed preferences show is that people prefer to maintain their social status (benefit accrues to them personally) rather than support an unpopular change that is extremely unlikely to happen and where their support is extremely unlikely to make a difference (benefits are distributed).
So even if I accept this method of finding truth, it actually shows less than it might appear at first glance.
What about increasing immigration to the US across the board, especially low-skill migrants and high-skill migrants outside the tech sector? Development economist Lant Pritchett has written (in 2017 and in 2023) that migration restrictions incentivize the creation of more labor-saving technologies than would be necessary in a fully open world:
I spent some time last summer looking into the “other countries” idea: if we’d like to slow down both Chinese AI timelines without speeding up US timelines, what if we tried to get countries that aren’t the US (or the UK, since DeepMind is there) to accept more STEM talent from China? TLDR:
There are very few countries at the intersection of “has enough of an AI industry and general hospitality to Chinese immigrants (e.g., low xenophobia, widely spoken languages) that they’d be interested in moving” + “doesn’t have so much of an AI industry that this would defeat the purpose” + “has sufficiently tractable immigration policy that they might actually do it.” GovAI did some survey work on this. Canada, Switzerland, France, Australia, and Singapore looked like plausible candidates (in order of how much time I spent looking into them).
Because this policy might also draw migrants who would have instead moved to the US, the question of whether this is a good idea hinges in part on the question of whether overall timelines or the West-China gap is more important (as is often the case). I think recently consensus has moved in the “timelines” direction (in light of very fast Western progress and the export controls and domestic regulation likely slowing things down in China), making it look more appealing.
Happy to share my (now somewhat outdated) draft privately if people are curious.
I am an immigrant (F1) in the US (and I would like to think of myself as ‘high-skilled’ although others may disagree!). So, I am clearly biased.
It all sounds good in theory but it would be hard to assess what counts as AI-related work and what doesn’t. Even just a simple regression can be called “Machine Learning”. Economists have to do regressions basically everyday. In that case, they would have it in their resume and then someone in the VISA office would still reject them because they would have to assume ‘Regression means AI’ and your plan to let the “Oxford-educated economist” into the US would still be foiled. Also skilled economists could just work at tech firms and help them make money which they could then invest in developing AI further.
PS:- I downvoted this post. As I said, I am biased. But I saw very little effort from the author to actually provide references/evidence to substantiate some of the implied causalities. For instance, the author says things like “This policy is unattractive if you think”. IMHO a better post in the “Effective Altrusim” forum (which is based on ‘evidence and reasoning’ and not just what people like to think) would be link to some empirical work that actually shows firms don’t simply just shift if you restrict high-skilled immigration. Here is a study that claims the contrary that indeed immigration restrictions on high-skilled workers simply moves operations offshore: Glennon, B. (2023). How do restrictions on high-skilled immigration affect offshoring? Evidence from the H-1B program. Management Science. https://economics.nd.edu/assets/361162/bglennon_immigration_offshoring.pdf
My impression is that immigration policy is unusually difficult to effect given how much of a hot-button issue it is in the US (ironic, given your forum handle). So while the scale may be large, I’m skeptical of the tractability.
On OpenPhil’s behavior, yeah, if they’re making it much easier for AI labs to hire talent abroad, then they’re doing a mistake, but that path from all-cause increases in high skill immigration to AI capabilities increases has enough noise that the effects here may be diffuse enough to ignore. There’s also the case that AI safety benefits a bunch from high skill immigration as well. Indeed, many people I regard as particularly impactful in the field have been kicked out of Berkeley due to visa issues.
Potential area of fruitful cause prioritization research if you have the time and motive to do the math! The particular difficulty I see is there may be no advocacy groups who’d like greater high skill immigration in all areas other than AI research, except safety research. So if the cost of starting such a group or giving extant groups ultimatums is too high, then OpenPhil will be faced with a tricky decision.