I agree a longer investigation of bottlenecks from raw materials would be useful.
But our estimate of robot doubling times does implicitly account for resource costs. We estimate how long it takes $1 of physical capital to create $1 of value-add. That value-add could come in the form of mining new resources or in the form of turning inputs (of raw materials or physical capital) into more useful outputs. The calc is saying: take the world’s stock of physical capital; imagine they’re used exclusively for mining materials and building more factories; how long until they can mine the materials and build physical capital to essentially double that initial stock.
Yes raw material prices will rise as stocks diminish, as has happened historically. But also, as has happened historically, we’ll find new sources of raw materials and we’ll innovate to do without the scarce resources. By extrapolating historical trends we’re assuming that these competing forces will play as they have done so before. I understand you think we should add a pessimistic adjustment here, implicitly assuming that the rising prices will win out over innovation more than has happened historically. But why?
Normal timescales for building new mines using mature, already well-established technologies is over a decade for exploration to feasibility, 1.8 years of construction planning and getting environmental permits (unlike a robot factory, which you may built almost whenever you like, you have to build a mine on the actual minerals) and 2.6 of construction and production set up: https://www.statista.com/statistics/1297832/global-average-lead-times-for-mineral-resources-from-discovery-to-production This is the timescales on which commodity cycles boom and bust.
I do not assume that “innovation may lose”, I actually do not consider new technologies for mining resources whatsoever in my reasoning, only the application of existing technologies to the deposits formerly uneconomical to extract. I try to explain that all these processes are just much slower than the timescales you are discussing.
How much is construction and production delayed by red tape / burucreacy? (This post assumes no human bottlenecks to simplify)
And how much faster could it go if ppl/robots worked day and night and spent 10x as much?
Btw, it takes 3 years to build a mine and that can’t be delayed, that doesn’t mean the doubling time is limited at 3 years. You could double your robots every year and then, each 3 year cycle, 8x your mines. You’d just need to spend a larger fraction of your robot labor on it, which is doable bc mines cost a lot less than other manufacturing physical capital does today.
Re high cost of capital, yes you’re right about interest rates and that does increase the cost. (This is explicitly considered in the “how long for a robot to pay for its own construction calculation, and implicit in the “how long for physical capital to make more physical capital calc”.) Though you also now make physical capital that lasts less long, saving costs. You can afford it bc you have AGI so have massive revenues and investment and bc you’re just making the physical capital yourself with your existing physical capital
I agree a longer investigation of bottlenecks from raw materials would be useful.
But our estimate of robot doubling times does implicitly account for resource costs. We estimate how long it takes $1 of physical capital to create $1 of value-add. That value-add could come in the form of mining new resources or in the form of turning inputs (of raw materials or physical capital) into more useful outputs. The calc is saying: take the world’s stock of physical capital; imagine they’re used exclusively for mining materials and building more factories; how long until they can mine the materials and build physical capital to essentially double that initial stock.
Yes raw material prices will rise as stocks diminish, as has happened historically. But also, as has happened historically, we’ll find new sources of raw materials and we’ll innovate to do without the scarce resources. By extrapolating historical trends we’re assuming that these competing forces will play as they have done so before. I understand you think we should add a pessimistic adjustment here, implicitly assuming that the rising prices will win out over innovation more than has happened historically. But why?
Normal timescales for building new mines using mature, already well-established technologies is over a decade for exploration to feasibility, 1.8 years of construction planning and getting environmental permits (unlike a robot factory, which you may built almost whenever you like, you have to build a mine on the actual minerals) and 2.6 of construction and production set up: https://www.statista.com/statistics/1297832/global-average-lead-times-for-mineral-resources-from-discovery-to-production This is the timescales on which commodity cycles boom and bust.
I do not assume that “innovation may lose”, I actually do not consider new technologies for mining resources whatsoever in my reasoning, only the application of existing technologies to the deposits formerly uneconomical to extract. I try to explain that all these processes are just much slower than the timescales you are discussing.
And you seem to have missed my argument about capital: if the interest rates skyrocketed due to transformative AI (see, e. g., https://www.lesswrong.com/posts/k6rkFMM2x5gqJyfmJ/on-ai-and-interest-rates), how would you finance all these mines?
Thanks
How much is construction and production delayed by red tape / burucreacy? (This post assumes no human bottlenecks to simplify)
And how much faster could it go if ppl/robots worked day and night and spent 10x as much?
Btw, it takes 3 years to build a mine and that can’t be delayed, that doesn’t mean the doubling time is limited at 3 years. You could double your robots every year and then, each 3 year cycle, 8x your mines. You’d just need to spend a larger fraction of your robot labor on it, which is doable bc mines cost a lot less than other manufacturing physical capital does today.
Re high cost of capital, yes you’re right about interest rates and that does increase the cost. (This is explicitly considered in the “how long for a robot to pay for its own construction calculation, and implicit in the “how long for physical capital to make more physical capital calc”.) Though you also now make physical capital that lasts less long, saving costs. You can afford it bc you have AGI so have massive revenues and investment and bc you’re just making the physical capital yourself with your existing physical capital