Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
I think all the bullet point critiques above are flaws in a reasonable model even *given* fully autonomous self-replicating factories? I do in fact believe in the self-replicating factories dream, but I think that's roughly compatible with my last line above: - self-replicating factories are possible, but progress towards them goes in two phases: 1. Waymo-like dynamics in robotics improvement, eg asymptotic progression towards sufficient reliability, such that the first mostly automated factory construction happens a long time before the first fully automated construction. this seems like it should be the default hypothesis given everything we know about robotics, but also immediately prevents sigma >1 in CES by definition. 2. constructing the entire industrial supply chain at once is not possible, requires huge amounts of novel R&D, and there are tons of bottlenecks to improving your process. I don't believe in the "hypergrowth via simple manufacturing processes" idea at all. in practice, you'd continually expand the "envelope of automation" deeper in supply chains, like an onion, over the course of years. but then this is exactly how automated R&D is progressing on the software-only side, it's just a continuation of that rather than a bend in the curve. - then economically, self-replicating goods see rapid price compression causing a deflationary environment that counters the effect of the capex to make them. - there would probably a time-bounded GDP bump (above my modal 5% from just generic AI agents in enterprise), but not large (since counteracted by the price compression), and non-automated goods and services (positional goods, relational services) would increase in % of the economy by definition. - increasingly, models of the industrial explosion like in AI 2040 seem to avoid highlighting the effect I describe on GDP because it doesn't line up with narratives about "AI becoming all of the economy" if an industrial explosion. instead, they show strong raw output growth, which obviously does happen. - but there is an unclear link between raw output growth, how much the actor captures the economic value provided, and how much this translates into power. the last one particularly is usually just assumed I think. - in practice I think a significant part of the output of the industrial explosion for many highly valuable current manufactured goods would be competitive for some time in cost with historical manufacturers, due to proprietary process knowledge for incumbents (Corning is a good example). Of course we should still have some probability on crazy scenarios, but the sheer weight of historical economic understanding of what happens when a good gets automated to me straightforwardly gives the above heuristics.
@herbiebradley @Jsevillamol i really don't get your econ views Herbie - do you just not believe we can create fully autonomous self-replicating factories?
He challenges claims of immediate hypergrowth from fully autonomous systems.
I think all the bullet point critiques above are flaws in a reasonable model even *given* fully autonomous self-replicating factories? I do in fact believe in the self-replicating factories dream, but I think that's roughly compatible with my last line above: - self-replicating factories are possible, but progress towards them goes in two phases: 1. Waymo-like dynamics in robotics improvement, eg asymptotic progression towards sufficient reliability, such that the first mostly automated factory construction happens a long time before the first fully automated construction. this seems like it should be the default hypothesis given everything we know about robotics, but also immediately prevents sigma >1 in CES by definition. 2. constructing the entire industrial supply chain at once is not possible, requires huge amounts of novel R&D, and there are tons of bottlenecks to improving your process. I don't believe in the "hypergrowth via simple manufacturing processes" idea at all. in practice, you'd continually expand the "envelope of automation" deeper in supply chains, like an onion, over the course of years. but then this is exactly how automated R&D is progressing on the software-only side, it's just a continuation of that rather than a bend in the curve. - then economically, self-replicating goods see rapid price compression causing a deflationary environment that counters the effect of the capex to make them. - there would probably a time-bounded GDP bump (above my modal 5% from just generic AI agents in enterprise), but not large (since counteracted by the price compression), and non-automated goods and services (positional goods, relational services) would increase in % of the economy by definition. - increasingly, models of the industrial explosion like in AI 2040 seem to avoid highlighting the effect I describe on GDP because it doesn't line up with narratives about "AI becoming all of the economy" if an industrial explosion. instead, they show strong raw output growth, which obviously does happen. - but there is an unclear link between raw output growth, how much the actor captures the economic value provided, and how much this translates into power. the last one particularly is usually just assumed I think. - in practice I think a significant part of the output of the industrial explosion for many highly valuable current manufactured goods would be competitive for some time in cost with historical manufacturers, due to proprietary process knowledge for incumbents (Corning is a good example). Of course we should still have some probability on crazy scenarios, but the sheer weight of historical economic understanding of what happens when a good gets automated to me straightforwardly gives the above heuristics.
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.