In his 7th (and now final) question on AI, @DAcemogluMIT asks the question that he and I have been debating for years: will AI create as many jobs as it will replace? We actually agree on some important parts. First, while AI may well transform jobs, this process will not be rapid. In my mind, that gives us time to adjust. Second, in the long term, there may be labour reallocations, and they may impact regions differently. Indeed, back in the financial crisis, Daron gave a series of talks worrying about whether government actions may get in the way of a labour reallocation process he thought was necessary there. Third, if it turns out that AI cannot easily automate jobs, then AI will turn out to be "more pro-Worker", and things will be fine. He reiterated this point in his post: "Of course, things may turn out differently. But if they do, it will be because AI becomes more pro-worker. I see this as unlikely, but it is useful to still consider the scenarios that would bring such an outcome. These are: (1) current AI technologies may become turn out pro-worker than I am anticipating; (2) in the future the industry will automatically and by itself change and focus on pro-worker applications; (3) based on AI-driven science and research, new sectors will be created where there will be need for new workers." The first item is one that @avicgoldfarb and I developed in our paper on O-Ring Automation; that when tasks are highly complementary, automating most of them doesn't eliminate a job or even lower worker productivity. This is something @DarioAmodei has also recently argued. When it comes down to it, Daron and I have the same economic model in mind. Where we differ is the probability weights we are putting on uncertain parameters. I tend to be more optimistic. He tends to be less. There is no resolution to this other than hard evidence, and that doesn't exist quite yet. We are at a stage where any glimmer of hard data is playing an undue role in updating people's beliefs here, and we should be more Bayesian about that. There is another aspect where we differ. Daron is of the opinion that governments can move the needle on pro-Worker AI by, say, reforming capital treatment in the tax code and giving people strong property rights over data. However, I think this is all going to require a ton of innovation and adjustment, with its own uncertainties, and so I am sceptical that governments can really help here. Nonetheless, if there are major disruptions and my optimism is misplaced, I think there will be government responses, but not much in that scenario is going to be fun.

Today, I am going to look at @DAcemogluMIT 's third question on AI. It is about the AI investment boom and its consequences. There is a lot to think about with regard to this, but Daron focuses on: "whether the AI boom can continue without leading to a massive increase in inequality." The premise this question is based on is whether an AI boom will continue and earn 10 per cent or more to its investors per annum over the foreseeable future. Daron doubts these returns will be realised (competition, speed of adoption, etc., all constrain it), and I share those doubts. But what if the dreams of investors are realised? Stijn Van Nieuwerburgh has crunched the numbers ( and it is daunting. The revenues required for a 10 per cent return would take up over 10 per cent of national income per annum. (Not because they are the same thing; that is just how this kinda worked out.) Daron worries about where those returns will go. He argues they will go to capital owners, and if that happens, there will be a massive rise in wealth and income inequality. But I don't think you can answer this question without first asking where the revenue will come from. AI investments can create AI, but the revenue still has to come from somewhere. One possibility is that it comes straight out of the labour share of income because AI replaces jobs with machines, and then we get a straight-out redistribution of income. Another possibility is that it comes from higher-value goods and services and, more generally, from higher productivity. Some of that productivity dividend goes to capital owners but, at the same time, it can also go to labour income, and it can also come to people in the form of consumer surplus. How do we delineate between the two in terms of plausibility? Well, for the straight-out redistribution story where the machines replace the people, the 10 per cent comes out of the labour share. The US labour share has been declining and is now about 54% of income. If the 10 per cent comes out of it, that falls to 44%. That sounds bad. But does it make sense? The trillions of dollars in AI investment didn't just pop out of nowhere. I would gather they do not come out of the luxury yacht market either. Instead, it came from other investments -- that is, the part that was producing the existing 46% capital share. Thus, what we have done here is potentially boost the return on capital by some amount. How much? Over the past decade, the NASDAQ has given investors an 18 per cent rate of return. This makes you wonder if Van Nieuwerburgh's 10 per cent return is reasonable. That said, that is 'real money' rather than 'book money.' For the growth scenario, with a 1 per cent boost to growth per annum, I calculate that the labour share would fall to about 47.5% a decade out. So about half the loss that the redistribution scenario has. This is because growth compounds, which softens the impact. But we can go further. Keeping the 10 per cent return (I know it's not the most reasonable ceteris paribus here), if the boost in growth was 2.1% per annum, then the labour share would be unchanged in a decade, allowing the capital owners to get their 10 per cent due. [By the way, the whole 10 per cent number is doing lots. It is the rate of return, and it is also the share of income going to AI investors, but that is just some funny miracle, and no one is really doing anything more than back-of-the-envelope scratches here.) I am not so optimistic that such a growth rate can be achieved, but at the same time, I suspect that the 10 per cent return assumption hinges on that. If it is a lower rate of growth, the premise of Daron's question also doesn't hold. So my answer, therefore, is that there is a good argument that we can do this without the massive increase in inequality. Simply because AI might be a lucrative investment isn't enough to cause us to worry, precisely because we need to answer why it is a lucrative investment in order to provide a more fulsome picture.