- Bio min wage for humans is ~2,400 kcal, or ~10,000 kJ (10J) - *That* has to be taken as chemical energy, so there's also a minimum transmission factor from whatever source is getting *turned into* chemical energy - Almost all calories: that means the sun. Most efficient food might be something like sugarcane, which is still a brutal 2% of solar radiation. Standard commercial silicon photovoltaic panels are more like 15-22%, and that's just going to go up. Claude says: > **Sugarcane** - With its C4 photosynthetic pathway, sugarcane converts approximately 2% of solar radiation into biomass, which is among the highest efficiencies for terrestrial crops. The resulting simple sugars are almost 100% digestible by humans. I've got a Gemini Deep Research report running on this... > The photosynthetic efficiency in agricultural crops, where the focus is on the conversion of solar energy into edible yield such as seeds, fruits, and tubers, is generally low. Most crop plants store only about 0.25% to 0.5% of the total sunlight they receive in the product. Sugarcane is an exception, with reported biomass yields storing as much as 3.5% of total solar energy received annually, although the efficiency for sucrose production is lower. A study comparing plants and photovoltaic cells found that the solar energy conversion efficiency for crop plants is around 1%. However, there is significant potential to improve this natural system through bioengineering and genetic modifications aimed at enhancing various stages of photosynthesis. For instance, the Realizing Increased Photosynthetic Efficiency (RIPE) project is an international effort focused on developing more productive crops by improving photosynthesis. Theoretical calculations suggest that the maximum conversion efficiency of solar energy to biomass is 4.6% for C3 photosynthesis and 6% for C4 photosynthesis under specific optimal conditions. Notably, modern soybean varieties have achieved near-maximum efficiency in light interception and partitioning carbon to seeds, suggesting that future improvements might focus on enhancing the efficiency of converting absorbed light energy into biomass. I think Carl Shulman might have had some working on this as well. So, where does that bring us? Well, that's really the absolute floor on wages for a human - [[Thermoeconomics]] anyone? And that's the *marginal cost*! In today's world, there's a huge upfront investment - all the energy it takes to get a human to age ~18... So it might make more sense to consider the average BMW required over a lifetime - which would bump it up to maybe even something like >2x 10J. What that means is that if there's even a chance that AI reduce *all* production functions to below that level, we need to watch out. I think a ~10x gap - *today* - in *fuel* efficiency (let alone *usage*) is not enough of a moat to be comfortable. For instance: some people just invoke 'comparative advantage' and leave it there. That makes sense in a world where you expect many or indeed all tasks to pay over the BMW - which, if you were only looking at today's world, might be your intuition. In that world, it's OK for someone to have no absolute advantage; they can still productively work to their comparative advantage, and it'll be worth trading with people who are worse at everything. Opportunity cost is what's key here: yes, you could make widget B yourself, but widget A is worth more and you can produce more of *those* in the same time, so some mix + trade will probably be better. (I vaguely remember reading *somewhere* that specialisation is not strictly a requirement for 'extent of the market' to be beneficial; probably a Mike Munger episode of EconTalk. Should look at this again.) But historically this is a *huge outlier* - subsistence farming has been the default mode of production for a lot longer, and hunting-gathering before that seems... a little more complicated (and much more Malthusian). But... that bites here as well, right? Like maybe the Nth robot is still cheaper to build & run in the lithium mines than the Nth human, but if it has a *higher opportunity cost*, it would still make sense to "employ" humans in the mines. - I guess that only makes sense if robots are in some sense scarce - But that *could* be the case: surely lithium is going to be more scarce than self-replicating cells fuelled by the sun or, indeed, 'just thin air': ![[Pasted image 20250514141914.png]] (Image description for search: Solein is protein made without plants, animals, sun / photosynthesis) So, OK, maybe there's room for *organic life* of some description to some extent. Then we get into the question of - how well-engineered are humans? How well did evolution do? I think there's a lot of answers here that bode *very badly* for any goal akin to "current humans / their direct descendants are flourishing". For instance: - Genetically engineer a variety of more optimised (i.e. specialised) organic workers. You probably keep no humans, just our genetic information - you have advanced enough GenEng to resurrect homo sapiens wholesale. If you *did* keep any as a diversify-your-resiliency-portfolio play, it probably wouldn't be many - the same genetic engineering would probably help keep required pop size & costs down. - Humans are actually pretty well-optimised for some things, and you keep humans around doing those things in anything down to the worst living conditions that will ever exist in the universe. Ignoring humans: bad end. Hoover up all resources. Humans die through indifference. So we're kind of back where we started but with less elegance and more epicycles; what matters is production functions, yes, but also then utility functions - and where these come together, relative prices. We need AI production to be expensive enough, and utility from other things to be high enough to produce *opportunity costs* that are high enough, to cover the BMW. That still seems pretty unlikely, but it takes more to ?debunk. Like - if we say a Dyson sphere gets us to ~100% solar efficiency, then genetically perfect sugarcane would have to 50x to get us on par for efficiency of fuel. (Modulo storage, but probably batteries beat cupboards on this one - especially if we need freezers! And electricity probably wins on transmission too, but this might be more complicated.) *If* we were on par with fuel, then it's going to come down to usage. [[Comparing Actuator and Human Muscle Efficiency]] So, like with all things evolutionary, humans are a particular set of trade-offs. But I think that still at best means human arms working in some factories, not whole humans - and probably not even that, as the true comparison is (whole human) to (human arm, supporting infra) to (whole specialised robot), and the first two have a lot of "dead weight" for any *particular* task that the latter does not (again, we were optimised for some level of generality). ALSO: we've veered heavily into the physical realm here. I think there's fully no hope in the cognitive realm. AI are currently an umpteenth (should quantify this...) of the BMW, and I would say on many tasks are currently on par or better than some fraction of knowledge workers. Between approximately perfect scalability of effort, faster serial thinking speed, lower energy usage, and ongoing improvement in all these things - there won't be cognitive jobs paying above BMW *for reasons of gains from trade*. There could be other reasons, though I'm yet to see any that are all three of compelling, plausible, and deep. [[➡️ Seedbox]] because I think there's a neat framework in here, probably. # So - Absolute advantage is actually insufficient to reason from - But comparative advantage is also definitely insufficient # Related [[Reject market theodicy, embrace agnosticism]]