Obviously agreed on the labour migration opportunity and it's exactly what we're doing with Africa Jobs Fund.
It seems to be an implicit premise in the piece though that export manufacturing won't be possible with AI. I don't think there's a neat direct line from LLMs to the automation of all forms of manufacturing. Certainly automation is growing but tasks that require complex motor skills (like manipulating and sewing fabric to make clothes) ought to still be a viable growth strategy for low-income countries for some time to come.
Also, on the tourism side, just from travelling in Africa, there is certainly some room for productivity gains in hospitality. I don't have numbers to point to, but the gap in output between a worker in a high-end Nairobi cafe and beach restaurant in Mombasa has got to be at least 2x.
Also, fixed endowments are only the limiting factor in a few places (e.g. Rwandan gorillas). But there is a huge amount of natural beauty, wildlife and quality beaches that go un-used across Africa.
So, plenty of headroom to grow tourism across Africa as well! I think the biggest constraint here is often government/infrastructure which make it hard to attract foreign tourists in significant numbers. As shown by the huge premium Rwanda is able to charge to see the gorillas compared to Uganda (and definitely compared to DRC)
Agree on the first point that we're not necessarily headed for full automation, and that garment-style work may be among the last to go within the industry. Two things worry me in the interim. First, the pathway needs to hold for decades to work as a growth escalator. And, second, this pressure doesn't require full automation, just compressed margins / fewer workers per unit of export, which is already the trend.
On tourism, I think we mostly agree; there's a lot of room for lagging countries to catch up, and governance/infrastructure is the binding constraint in most places. My claim is narrower — catch-up gains are a one-time level effect, and can't carry once you're close to the frontier. So tourism can raise incomes meaningfully in specific countries without functioning as the compounding engine manufacturing was.
On export manufacturing, one argument for automation might be that substantial advances in LLMs — until LLMs become very effective researchers themselves — also makes solving robotics easier.
Nice (albeit, somewhat depressing!) piece Deena, thanks for writing it. Obviously agree with the potential of labour migration to promote economic development in poorer countries, but with three big caveats.
Firstly, labour migration opportunities are rarely available for the poorest and most vulnerable people. More needs to be done to open up said opportunities: investing in formal up-skilling programmes, creating an enabling ecosystem, and incentivizing recruitment.
Secondly, as you already note, remittances are used for personal consumption rather than broader economic growth. More needs to be done to make countries of destination financially invest in the countries they are recruiting from, building skilling and labour export systems which serve all markets.
Thirdly, there is a risk that countries of origin promote emigration at the expense of workers' rights (see headlines re Kenyan workers in the GCC...). More needs to be done to learn the lessons from the Philippines and India: how to build an export infrastructure that safeguards diaspora while abroad, and harnesses their contributions when they return.
Thanks Helen — agree with all three points. On the third, the Philippines' POEA/DMW model is imperfect but far ahead of what most African sending countries have. If labor export is going to be a deliberate strategy, that institutional layer is important for workers' protection.
Absolutely. I do worry that their export model has skewed their labour market though... Keen to learn the good lessons from the Philippines and avoid the bad ones!
Good piece. I think the labor migration option is by far the most important and promising prospect. Perhaps not so much for development, but for human progress. The gains are extremely high and I think the positive feedbacks to sending countries aren't fully understood or documented. As another commenter says, there are lots of labor migration cases that are nightmarish (care workers in middle east, construction and logistics in gulf states), there are many more positive examples. I've been tracking the rapid rise of emigration of trained nurses from a number of African countries to fill labor shortages in European countries (especially UK) and it seems like a very positive story.
Thanks Gawain. Agree the sending-country feedbacks are under-documented. Most of the literature is on remittance flows rather than skill/network effects on return. Would be interested in what you're seeing in the nurse data, especially whether training pipelines in origin countries are expanding in response to the UK demand (the "brain gain" story).
In Latin America some of the higher income economies have endowments like copper reserves or a big shipping canal made largely by foreign engineers and workers that can charge tolls on ships passing through from one country en route to another country, neither of which have the canal. Do you think AI boosting growth could raise the value of raw materials, geográphical advantages etc in LIC. East Asian dev path is the coolest but maybe GCC dev path, where foreign tech gives economic value to mineral or other natural resources can raise incomes for LIC.
I worry AI energy demand will make less LNG available for export from the US and conversión to cooking fuel or fertilizers in LMIC. Is there also some hope that AI energy demand will prompt greater investment in solar, wind, batteries, nuclear, geothermal, tidal, synthetic gas energy that may speed the arrival of energy abundance in LMIC?
To have any advantage in Data Centers you need cheap energy and what most people do not think about is CONNECTIVITY. Without diverse, resilient fiber connectivity a data center is a beached whale
Nice piece again Deena. One area that deserves more attention is how firms in developing countries can use AI, particularly LLMs (just LLMs), to strengthen innovation.
In many developing countries, businesses are highly homogeneous. Travel 30 km along a road and you may see the same businesses selling identical products, with very little differentiation. An outcome high birth and death rates of small businesses - literature has well documented these trends). Development programmes have tried to address this through business advisory services and technical assistance, but the problem persists.
This is where AI could make some differences. Businesses need to learn how to use LLMs not simply as a substitute for Google, but as tools for generating ideas, differentiating products, understanding customers and improving business models.
If developing countries fail to take advantage of this soft technological opportunity, they will miss another potential source of growth, just as many developing countries missed earlier opportunities e.g. export-led industrialisation. I don't see any progress in development, if developing countries consistently miss out on emerging growth opportunities years after years.
One thing I would say is that even if manufacturing is mostly automated, that doesn’t mean it doesn’t happen in poor countries. I’d expect fairly significant technological diffusion because the majority of relevant physical inputs don’t exist in countries where the frontier models are built. I’d expect to see robot-run factories managed by American AI models in many LMICs too -- this technical point about location doesn’t mean they reap much of the surplus, but in principle I could see worlds where the technological diffusion mitigates some of the economic harm. I assume you’ve already seen Objection 3 in this article by Tom Davidson: https://www.forethought.org/research/could-one-country-outgrow-the-rest-of-the-world#objection-3-what-about-technological-diffusion-and-stealing
Great piece!
Obviously agreed on the labour migration opportunity and it's exactly what we're doing with Africa Jobs Fund.
It seems to be an implicit premise in the piece though that export manufacturing won't be possible with AI. I don't think there's a neat direct line from LLMs to the automation of all forms of manufacturing. Certainly automation is growing but tasks that require complex motor skills (like manipulating and sewing fabric to make clothes) ought to still be a viable growth strategy for low-income countries for some time to come.
Also, on the tourism side, just from travelling in Africa, there is certainly some room for productivity gains in hospitality. I don't have numbers to point to, but the gap in output between a worker in a high-end Nairobi cafe and beach restaurant in Mombasa has got to be at least 2x.
Also, fixed endowments are only the limiting factor in a few places (e.g. Rwandan gorillas). But there is a huge amount of natural beauty, wildlife and quality beaches that go un-used across Africa.
So, plenty of headroom to grow tourism across Africa as well! I think the biggest constraint here is often government/infrastructure which make it hard to attract foreign tourists in significant numbers. As shown by the huge premium Rwanda is able to charge to see the gorillas compared to Uganda (and definitely compared to DRC)
Thanks Ben!
Agree on the first point that we're not necessarily headed for full automation, and that garment-style work may be among the last to go within the industry. Two things worry me in the interim. First, the pathway needs to hold for decades to work as a growth escalator. And, second, this pressure doesn't require full automation, just compressed margins / fewer workers per unit of export, which is already the trend.
I wrote a bit in my last article about why I think the manufacturing export employment opportunity is likely to decline - https://newsletter.deenamousa.com/p/ai-could-keep-poor-countries-poor.
On tourism, I think we mostly agree; there's a lot of room for lagging countries to catch up, and governance/infrastructure is the binding constraint in most places. My claim is narrower — catch-up gains are a one-time level effect, and can't carry once you're close to the frontier. So tourism can raise incomes meaningfully in specific countries without functioning as the compounding engine manufacturing was.
Yep, think I broadly agree with both of these!
And agreed that the manufacturing path may be getting harder (and narrower), but still think its the best of the available options.
On export manufacturing, one argument for automation might be that substantial advances in LLMs — until LLMs become very effective researchers themselves — also makes solving robotics easier.
Nice (albeit, somewhat depressing!) piece Deena, thanks for writing it. Obviously agree with the potential of labour migration to promote economic development in poorer countries, but with three big caveats.
Firstly, labour migration opportunities are rarely available for the poorest and most vulnerable people. More needs to be done to open up said opportunities: investing in formal up-skilling programmes, creating an enabling ecosystem, and incentivizing recruitment.
Secondly, as you already note, remittances are used for personal consumption rather than broader economic growth. More needs to be done to make countries of destination financially invest in the countries they are recruiting from, building skilling and labour export systems which serve all markets.
Thirdly, there is a risk that countries of origin promote emigration at the expense of workers' rights (see headlines re Kenyan workers in the GCC...). More needs to be done to learn the lessons from the Philippines and India: how to build an export infrastructure that safeguards diaspora while abroad, and harnesses their contributions when they return.
Thanks Helen — agree with all three points. On the third, the Philippines' POEA/DMW model is imperfect but far ahead of what most African sending countries have. If labor export is going to be a deliberate strategy, that institutional layer is important for workers' protection.
Absolutely. I do worry that their export model has skewed their labour market though... Keen to learn the good lessons from the Philippines and avoid the bad ones!
Good piece. I think the labor migration option is by far the most important and promising prospect. Perhaps not so much for development, but for human progress. The gains are extremely high and I think the positive feedbacks to sending countries aren't fully understood or documented. As another commenter says, there are lots of labor migration cases that are nightmarish (care workers in middle east, construction and logistics in gulf states), there are many more positive examples. I've been tracking the rapid rise of emigration of trained nurses from a number of African countries to fill labor shortages in European countries (especially UK) and it seems like a very positive story.
relevant: https://africajobsfund.substack.com/p/the-case-for-international-labour?r=9kmoz&utm_campaign=post-expanded-share&utm_medium=web
Thanks Gawain. Agree the sending-country feedbacks are under-documented. Most of the literature is on remittance flows rather than skill/network effects on return. Would be interested in what you're seeing in the nurse data, especially whether training pipelines in origin countries are expanding in response to the UK demand (the "brain gain" story).
In Latin America some of the higher income economies have endowments like copper reserves or a big shipping canal made largely by foreign engineers and workers that can charge tolls on ships passing through from one country en route to another country, neither of which have the canal. Do you think AI boosting growth could raise the value of raw materials, geográphical advantages etc in LIC. East Asian dev path is the coolest but maybe GCC dev path, where foreign tech gives economic value to mineral or other natural resources can raise incomes for LIC.
I worry AI energy demand will make less LNG available for export from the US and conversión to cooking fuel or fertilizers in LMIC. Is there also some hope that AI energy demand will prompt greater investment in solar, wind, batteries, nuclear, geothermal, tidal, synthetic gas energy that may speed the arrival of energy abundance in LMIC?
To have any advantage in Data Centers you need cheap energy and what most people do not think about is CONNECTIVITY. Without diverse, resilient fiber connectivity a data center is a beached whale
Nice piece again Deena. One area that deserves more attention is how firms in developing countries can use AI, particularly LLMs (just LLMs), to strengthen innovation.
In many developing countries, businesses are highly homogeneous. Travel 30 km along a road and you may see the same businesses selling identical products, with very little differentiation. An outcome high birth and death rates of small businesses - literature has well documented these trends). Development programmes have tried to address this through business advisory services and technical assistance, but the problem persists.
This is where AI could make some differences. Businesses need to learn how to use LLMs not simply as a substitute for Google, but as tools for generating ideas, differentiating products, understanding customers and improving business models.
If developing countries fail to take advantage of this soft technological opportunity, they will miss another potential source of growth, just as many developing countries missed earlier opportunities e.g. export-led industrialisation. I don't see any progress in development, if developing countries consistently miss out on emerging growth opportunities years after years.
Amazing piece!
One thing I would say is that even if manufacturing is mostly automated, that doesn’t mean it doesn’t happen in poor countries. I’d expect fairly significant technological diffusion because the majority of relevant physical inputs don’t exist in countries where the frontier models are built. I’d expect to see robot-run factories managed by American AI models in many LMICs too -- this technical point about location doesn’t mean they reap much of the surplus, but in principle I could see worlds where the technological diffusion mitigates some of the economic harm. I assume you’ve already seen Objection 3 in this article by Tom Davidson: https://www.forethought.org/research/could-one-country-outgrow-the-rest-of-the-world#objection-3-what-about-technological-diffusion-and-stealing