This estimate assumes that all biological functions in an organism can be replicated with technologies, and that these technologies can reach the same efficiency as the biological functions that reached high efficiency due to evolution and natural selection.
I don’t think this is realistic. Perhaps in isolation you could build systems that efficiently accomplish some of these functions, but in the case of cultured meat they all have to be compatible with/support the growth of animal cells and tissues. This is an enormous handicap. All of the technologies you cite as analogous (solar panels vs plants, cars vs horses, planes vs birds, recombinant vs porcine insulin) represent new approaches that are completely free from the limitations of the biological systems they’ve replaced. I don’t think any of them should be counted as precedents for the type of innovation cultured meat would require.
Yes this is what I meant by “cars are not mechanical horses” in an earlier thread, thanks for putting it in more precise terms.
I think I have two related counterarguments to the OP:
1) the car vs horse and plane vs bird reference classes/analogies seems to me to be moderately strong or even very strong evidence that humans are eventually capable of something that can accomplish what nature does cheaply and well, but only weak evidence for any specific strategy. Maybe plant-based or cultured meat is how we get there, but maybe it’d look entirely different.
2) If anything, the history of human progress in trying to imitate nature is anti-evidence for strategies that look like cultured meat (taking a highly coupled system from nature and then trying to mechanically replicate nature in a very similar way to how nature does it). Notably we do not fly around in ornithopters, most progress in AI doesn’t look like brain scanning and replication, and we still haven’t figured out giant mechs to ride around (cool as they may seem).
The more I think about it, the more I start to believe that cultivated meat is feasible, and that your examples offer some evidence.
So consider the function of flying. You may say that the function of flying with wings cannot be fulfilled with technology, that imitating nature does not work . But your examples refer to humans flying with technologies that use wings. But humans are much heavier than birds. With airplanes, we can fly faster, over longer distances and carry heavier weights, things that biological organisms were never capable of doing. And we are talking about many orders of magnitude faster, further and heavier. Why would the functions necessary for muscle cell growth (e.g. oxygenation, nutrient production, waste removal,...) be any different? Why would these functions never be able to be fulfilled at least as efficiently with technologies than with organs?
Then you offer the example of ornithopters, and I conclude that even imitating nature is possible. The function of flying with wings can be imitated with technology. Those ornithopters are as heavy and large as birds. Some of those ornithopters even use artificial muscles (polymers that can contract like muscles). So the function of muscle contraction can also be replicated. Some ornithopters have machine learning AI such that they can learn to fly, which means the function of brains can be replicated. So the complex combination of functions “learning to fly with wings using muscle contractions” can be replicated. What more evidence do you want that technology can build tools that fulfill complex functions that are fulfilled by biological organisms?
Thanks for the question, had to think a while. About infeasibility of cultivated meat, best counterevidence for me would be seeing a massive disinvestment in cultivated meat R&D, a consensus among researchers openly saying that it is too difficult to make progress.
Another crucial thing that would change my mind, is evidence about the feasibility of plant-based meat, that substitution towards plant-based is faster than I would expect (faster than cultivated meat innovation). This would mean seeing a fast increase in the number of vegans, and especially conservative male meat identifiers switching to plant-based meats.
Ok, thanks for that! Things that will update my mind include:
Social proof: The biggest thing that will change my mind is if Open Phil science team or other EA researchers that I trust generally update towards cost-competitive cultured meat being the most viable and plausible route to reducing factory farming (among many possible options). Of course that might be too late to be useful; I will also update somewhat if the top biotech VCs made major investments into cultured meat, and (to a noticeably lesser extent) if the most prestigious traditional VCs made significant investments in the sector. I will probably also update if top ~10 Metaculus forecasters or other top forecasters a) believe something very different and b) demonstrated to me that they spent considerable time looking into this.
Near term costs that aren’t VC etc subsidized: Humbird’s analysis claims costs of >$200/kg using wild-type animal cells, so if I see credible evidence of <$100/kg in the near future (say next 3 years), I’d consider that a moderate refutation of Humbird’s model, even though his endline numbers were lower. At the very least, I’d want to dig into how they did that. If they did it entirely through metabolic engineering improvements, I’d consider this less of a ding against Humbird than if they did it through other efficiencies.
Amino acid + other input prices: There are a number of things that will update me towards providing/using cheap amino acids is easier than I thought. Eg, Humbird thinks you need very pure amino acids to grow cultured meat. Many cultured meat proponents think you can do it with less. But this is an empirical question: You can just try to make cultured meat using less optimal concentrations of amino acids. There are a number of other statistical tests and empirical arguments that can convince me.
Hygiene: I think I can change my mind if I see either direct or indirect evidence that cheap hygiene at scale is much easier than I think.
Conceptual/analytic framework: On the conceptual level, the thing that will change my mind in terms of whether I’m framing this correctly includes a) if somebody empirically demonstrates that with a more accurate construction of reference classes, things that look like cultured meat (mechanical horses) are actually quite common in the wild. Or b) somebody logically or conceptually convinces me that my current categorizations of this car vs mechanical horses vs selectively bred horses ontology is confused/broken for reasons I don’t currently understand.
Comprehensiveness of search for alternative options: I don’t think of plant-based meat as a large crux, and certainly not vegan identification. I think I’ll be forced to be relatively more bullish on cultured meat if I see that reasonable people already did a comprehensive, first-principles-based search on ways to reduce animal suffering and landed on cultured meat and plant-based meat as the best methods they can find; right now I just think people are way too overindexed on existing ways of doing things (see eg the Decisive book or specifically my notes on “Narrow Framing”).
Not to over-emphasize this part of the debate, but I don’t think future cultured meat systems will be that analogous to how they look in animals. For example, modern bioreactor systems often employ a “sterile boundary” approach to sterility, which is pretty different than using antibodies to attack foreign particles. Depending on what bioreactor system you use, there are lots of things that look pretty different than what happens inside an animal.
The core of our disagreement seems to be here:
I don’t think this is realistic. Perhaps in isolation you could build systems that efficiently accomplish some of these functions, but in the case of cultured meat they all have to be compatible with/support the growth of animal cells and tissues. This is an enormous handicap. All of the technologies you cite as analogous (solar panels vs plants, cars vs horses, planes vs birds, recombinant vs porcine insulin) represent new approaches that are completely free from the limitations of the biological systems they’ve replaced. I don’t think any of them should be counted as precedents for the type of innovation cultured meat would require.
Yes this is what I meant by “cars are not mechanical horses” in an earlier thread, thanks for putting it in more precise terms.
I think I have two related counterarguments to the OP:
1) the car vs horse and plane vs bird reference classes/analogies seems to me to be moderately strong or even very strong evidence that humans are eventually capable of something that can accomplish what nature does cheaply and well, but only weak evidence for any specific strategy. Maybe plant-based or cultured meat is how we get there, but maybe it’d look entirely different.
2) If anything, the history of human progress in trying to imitate nature is anti-evidence for strategies that look like cultured meat (taking a highly coupled system from nature and then trying to mechanically replicate nature in a very similar way to how nature does it). Notably we do not fly around in ornithopters, most progress in AI doesn’t look like brain scanning and replication, and we still haven’t figured out giant mechs to ride around (cool as they may seem).
The more I think about it, the more I start to believe that cultivated meat is feasible, and that your examples offer some evidence.
So consider the function of flying. You may say that the function of flying with wings cannot be fulfilled with technology, that imitating nature does not work . But your examples refer to humans flying with technologies that use wings. But humans are much heavier than birds. With airplanes, we can fly faster, over longer distances and carry heavier weights, things that biological organisms were never capable of doing. And we are talking about many orders of magnitude faster, further and heavier. Why would the functions necessary for muscle cell growth (e.g. oxygenation, nutrient production, waste removal,...) be any different? Why would these functions never be able to be fulfilled at least as efficiently with technologies than with organs?
Then you offer the example of ornithopters, and I conclude that even imitating nature is possible. The function of flying with wings can be imitated with technology. Those ornithopters are as heavy and large as birds. Some of those ornithopters even use artificial muscles (polymers that can contract like muscles). So the function of muscle contraction can also be replicated. Some ornithopters have machine learning AI such that they can learn to fly, which means the function of brains can be replicated. So the complex combination of functions “learning to fly with wings using muscle contractions” can be replicated. What more evidence do you want that technology can build tools that fulfill complex functions that are fulfilled by biological organisms?
What evidence would cause you to change your mind?
Thanks for the question, had to think a while. About infeasibility of cultivated meat, best counterevidence for me would be seeing a massive disinvestment in cultivated meat R&D, a consensus among researchers openly saying that it is too difficult to make progress.
Another crucial thing that would change my mind, is evidence about the feasibility of plant-based meat, that substitution towards plant-based is faster than I would expect (faster than cultivated meat innovation). This would mean seeing a fast increase in the number of vegans, and especially conservative male meat identifiers switching to plant-based meats.
Ok, thanks for that! Things that will update my mind include:
Social proof: The biggest thing that will change my mind is if Open Phil science team or other EA researchers that I trust generally update towards cost-competitive cultured meat being the most viable and plausible route to reducing factory farming (among many possible options). Of course that might be too late to be useful; I will also update somewhat if the top biotech VCs made major investments into cultured meat, and (to a noticeably lesser extent) if the most prestigious traditional VCs made significant investments in the sector. I will probably also update if top ~10 Metaculus forecasters or other top forecasters a) believe something very different and b) demonstrated to me that they spent considerable time looking into this.
Near term costs that aren’t VC etc subsidized: Humbird’s analysis claims costs of >$200/kg using wild-type animal cells, so if I see credible evidence of <$100/kg in the near future (say next 3 years), I’d consider that a moderate refutation of Humbird’s model, even though his endline numbers were lower. At the very least, I’d want to dig into how they did that. If they did it entirely through metabolic engineering improvements, I’d consider this less of a ding against Humbird than if they did it through other efficiencies.
Amino acid + other input prices: There are a number of things that will update me towards providing/using cheap amino acids is easier than I thought. Eg, Humbird thinks you need very pure amino acids to grow cultured meat. Many cultured meat proponents think you can do it with less. But this is an empirical question: You can just try to make cultured meat using less optimal concentrations of amino acids. There are a number of other statistical tests and empirical arguments that can convince me.
Hygiene: I think I can change my mind if I see either direct or indirect evidence that cheap hygiene at scale is much easier than I think.
Conceptual/analytic framework: On the conceptual level, the thing that will change my mind in terms of whether I’m framing this correctly includes a) if somebody empirically demonstrates that with a more accurate construction of reference classes, things that look like cultured meat (mechanical horses) are actually quite common in the wild. Or b) somebody logically or conceptually convinces me that my current categorizations of this car vs mechanical horses vs selectively bred horses ontology is confused/broken for reasons I don’t currently understand.
Comprehensiveness of search for alternative options: I don’t think of plant-based meat as a large crux, and certainly not vegan identification. I think I’ll be forced to be relatively more bullish on cultured meat if I see that reasonable people already did a comprehensive, first-principles-based search on ways to reduce animal suffering and landed on cultured meat and plant-based meat as the best methods they can find; right now I just think people are way too overindexed on existing ways of doing things (see eg the Decisive book or specifically my notes on “Narrow Framing”).
Not to over-emphasize this part of the debate, but I don’t think future cultured meat systems will be that analogous to how they look in animals. For example, modern bioreactor systems often employ a “sterile boundary” approach to sterility, which is pretty different than using antibodies to attack foreign particles. Depending on what bioreactor system you use, there are lots of things that look pretty different than what happens inside an animal.