I digested a paper from Nobel winner Acemoglu and a couple of others with the help of notebookLM and learned a few things. 1. A.I is a bet on replacing labour with capital – when capital replaces labour, it worsens inequality. Rich get richer, poor get poorer. 2. The cost of capital compared to the cost of labour is a significant driver of automation. Part of why A.I investments make sense is because taxes make labour more expensive in the U.S. 3. Previous automations lowered costs and improved productivity, leading to more money chasing fewer goods & services, thus stimulating growth in the economy to meet that extra demand. Lots of A.I lowers costs only marginally (likely within the margin of the difference between taxes on labour and capital) and isn't really improving productivity. So no wage growth, no new demand to cater for, and no new goods and services. 4. A.I growth is effectively being subsidised by (the U.S) govt. A.I is largely cost-cutting and not value-creating, with a minimal productivity effect and low replacement effect. A.I will thus lead to stagnating growth, and I believe over time governments will realize this and tax A.I so as to promote replacement jobs and avoid deadweight A.I. Jobs provide a significant social benefit that data centres and frontier models don't. 5. With A.I, jobs are moving from being defined as a whole block into a number of tasks. Tasks will continue to be taken away, especially where judgement isn't required. Perhaps modular (informal) judgement jobs is where we end up, and govts that can will provide more welfare payments to the poor. 7. In informal economies like Nigeria with low wages and low productivity, A.I will not replace many jobs. The limited formal jobs we have will hardly change by much. What A.I will do to informal jobs is yet to be seen. Maybe replacement opportunities abound there? Our govt's A.I strategy has to be at the implementation layer of our informal economy and then work backwards. 8. What does this mean for the future of work? There will be clear winners and losers from A.I. To be a winner, break down your job into tasks. Understand where human judgement is needed and get super skilled in that area. Usually it would be around relationships and tacit knowledge that's difficult to codify. Or areas where responsibility is needed. I'm still learning about the economics of A.I. Please share your thoughts, learnings and reference articles.