You can also use the forum chat.
PabloAMC šø
Karma: 1,491
Hey Bruce, thanks for such a detailed comment. Having also read the book, I very much agree with it.
for consumers, āsame meat, less likely to make you sickā or contain drug residues or mercury seems like it could be a pretty compelling message
I think that could be true in case of some contaminated meat scandals, but if people are already used to a perceived level of risk they may not automatically adjust their behavior. It is the same as telling people that motorbikes are so much more dangerous than all other transportation methods; I think most people donāt change their behaviour based on that. That is also why I think there is no strong pull incentive for cultivated meat, unfortunately. However, the same applies to EV and yet they seem bound to succeed, so I think the government support strategy is likely correct.
Consumer Motivation 2: animal suffering I think this is likely a severely underestimated retention effect and also probably (IMHO) the biggest lever we have to cause a societal change: once the societal, health and taste pressure on people severely decreases, people will find it less and less acceptable to eat meat. My theory of change is that it might be hard to obliterate factory farming just on price and other features, but cultivated meat should catalyse a societal moral change by eroding the justifiability of factory farming.
Plant-based meat has failed because almost no one understood the assignment:
I think it makes sense, but I think there is also a possibility that people psychologically distinguish meat from plant-based meat, and just follow inertia and custom to avoid social pressure. There are a lot of people who do not want their identity to be vegetarian. I do not have strong priors here. However, I model this on the webpage https://āāpabloamc.github.io/āāCultivated_meat/āāinteractive.html and let people assign value to meat being identical (including made of animal cells), or even to meat being produced from animalsāwhich I find significantly more unlikely and is different from neophobia. Significantly more unlikely because theyād have to explicitly put value on killing animals, which seems to me like a less natural psychological separation than ājust being meatā. In any case, my goal was partly to allow the EA community to see how different priors would lead to different market shares.
I think you miss the much stronger arguments that can really move policymakers: GDP and food security, which are my focus in chapters 9 and 10, respectively (chapter 10 walks through the incentives for China, India, the US, Brazil, Europe, and small tech forward countries (Japan, S. Korea, Israel, Singapore).
Probably you have much better access to information here. The only thing that worries me is that every small field (like quantum, where I work) claims ānational securityā relevance, and I am not sure the politicians can calibrate things accurately. For example, quantum computing, where I work, is largely not a national security issue, as long as one uses some well-understood cryptographic standards. Also, it so happens that AI can be a blessing for the scientific and technological component of our endeavour, but it will almost certainly eclipse all other topics for the next few years to come. Again, no strong priors; you probably know better.
I think most cuts should be something that we could replicate without significant price difference. Or perhaps scaffolding ends up being a challenge. Also probably cultivated gives you more freedom to explore conventionally infeasible options.
Are you saying āfactory farmingā an actual axis of variation people will seek? Or are you arguing that since some species are much more challenging to achieve price parity with, people will keep eating those species too? I think the latter makes sense to me.
Awesome, thanks, Bruce,
I have a few other thoughts on your post that Iāll try to share later today, re: government incentives to make this happen faster than 10-20 years.
Iād love to hear so. The model I proposed and that resulted from those many years is based on Bass diffusion dynamics with typical parameter values, but of course, Iād love to supercharge the change.
Hey Bruce! Thanks for the reply. This is indeed one point where I was sloppy, though I was not trying to challenge your number. Let me explain my current understanding:
Lewis Bollard argues (and I think it is right) that 2kg of feed converts into approximately 1kg of live animal, of which 2ā3 is edible meat.
You argue (and I also think it is accurate) that a chicken needs to eat 9 calories of grain to produce 1 calorie of meat. This is not incompatible with the above because grains contain much less water, which you noted above.
Beyond caloric content, we should also analyse amino acids, which are also used for cell buildup and I did not take into account. I think the amino acid content is the binding constraint to set up a cost floor, and sits at ~20% for both dry feed and wet meat (mostly coincidence).
This is how I think of it:
Let the chickenās FCR be ~2 kg feed/ākg live weight (Bollard). Dry feed is ~20% protein (typical broiler ration).
So 1kg of chicken consumes roughly: 2 kg feedĆ0.20=0.4 kg feed protein per kg live weight
Meanwhile, a kg of (wet) live weight yields ~0.65 kg edible meat at ~20% protein = ~0.13 kg protein in product.
So the chickenās protein conversion is roughly: 0.13 kg protein out /ā 0.40 kg protein in ā33%
So I think if we care only about the caloric content, you are right (and I was wrong to use Lewisā number); but if we care about sourcing amino acids from broadly the same plants used for chicken feed, probably there is a cost floor for cultivated meat around 1ā3 the cost of chicken meat.
Would you agree with this analysis?
Fully agree with everything you point out.
I think I did not do a good job of framing this post. My goal was not to criticise the people working in animal welfare organisations, quite the contrary! I think they are doing very necessary work, and I am really grateful to them, especially given how weird this looks to most people. In other words, I am not claiming that these feelings are right, only that they exist.
The goal of the post was instead to reflect on something I feel (and presumably others feel too), which may be dragging donations to those organisations. In my case, this applies to politics too.
I think they already do it, under the premise of supporting local and rural farmers.
I basically agree, though my impression is that the rest of the solutions will play a complementary role.
In these campaigns the meat producing companies could engage in similar advertising that would lead to a negative sum game. More generally, in malaria thereās no human adversary. Perhaps I donāt like playing bad cop, even when I know itās necessary.
Let me clarify that I totally get why this is important and necessary, and I take parts in THL campaigns, but itās not something that I feel attracted to.
Itās tricky too: the GWWC pledge only considers salary curtailments as donations if you manually and reversibly lower your salary and believe the charity is highly effective. In many non EAs organisations (say accepting a lower salary to work in key government organisations) thatās not an option.
Those are great points, thanks, I think you are right. On the other hand, I think my argument was that if the āscience is solvedā and cultivated meat became cheaper and more environmentally friendly, I donāt see the current state of factory farming as a stable equilibrium situation: I donāt think it is reasonable to expect an indefinite protection of a more expensive, more polluting and less worker friendly economic sector in favour of another. Eg, a ban might be feasible, but it may not be sustainable in decade-long time horizons.
Thanks for the post!
AGI-powered persuasion tools are equally available to the well-resourced conventional meat industry.
As Bruce Friedrich mentions in its book Meat, I think this is unlikely to be the case. While I expect opposition from farmers, I think the large companies are more likely to be supportive, because (i) it is plausible that cultivated meat could become much cheaper than animal produced one, the floor is lower, (ii) they could create larger barriers to entry, using eg IP, and (iii) they do not have large sunk costs in their conventional animal farming facilities, and (iv) it likely allows them faster market reactions to demand and more stability (no avian flu, say). Bruce sometimes feels a bit too optimistic in its book, but I tentatively agree with those points.
Letās say that AGI solves cultivated meat for us. Cultivated meat is already illegal in seven[5] US states. It might soon be illegal in the entire European Union. By the time we get AGI, will they even be able to sell it anywhere?
I am not well calibrated on this, but I would argue the likelihood of the full EU making cultivated meat illegal is low. I think many countries in the EU have been able to ban GMOs or nuclear power because there was little push from the pro-GMO or pro-nuclear side, and there were easy environmental arguments to be made from the anti side, even if misguided. I donāt think that is likely to be the case for cultivated meat. It is more likely to resemble what happened with coal phase-outs.
I think the most likely scenario is:
There are farmersā protests.
The EU makes it harder but not illegal to sell cultivated meat, perhaps delaying some approvals.
At some point, it becomes easier to subsidise animal farming than to ban cultivated meat outright, because spending money is always easier, and they are doing it already anyway.
This gradually stops except for the highest welfare farming conditions, as old and nasty factory farms become places nobody wants to work in, and thus no individual farmers rely on them for a living.
It is also worth noting that if one cultivated meat product is approved for sale in the EU, one could, with time and patience, probably strike a single market case to bring down laws forbidding its sale in other EU countries. I agree, though, with the statement that ācurrent trajectories in the US and EU point toward more restrictions before the likely arrival of AGI.ā
Existing evidence points toward low acceptance driven by food and food technology neophobia.
Throughout the draft there seems to be a question on whether cultivated meat would achieve the same taste. In practice, I think the consumers wonāt really wonder too much if it looks, tastes and is otherwise exactly the same as what they typically buy. This is, in fact, perhaps the biggest difference between cultivated vs plant-based food: plant-based can taste just as good, but an individual product may not offer the original culinary flexibility. For example, literally from today, blind tasting of Aleph Farms cultivated meat did confirm this source here.
But there are some plausible worlds where AGI capability is more concentratedācontrolled by a small number of governments, corporations, or platforms.
I think it is likely that from the purely scientific point of view, someone will pay for that to happen, be it Jeff Bezos (who has research centres for the matter), Bill Gates (who is quite worried about climate change) or Dario Amodei (who thinks AI for biotechnology is the best application of AI).
consumers are unlikely to switch to cultivated meat unless it is substantially cheaper than conventional meat.
Again, how rooted is this in actual data? It would seem to me that if you go to the supermarket and you find two exactly equivalent products (package included) with a $0.25 price difference, people will typically buy the cheap one. Especially if they try it and they like it, which they should, since it is the same. In fact, what confuses me about comments like this is how a product can cause neophobia if it looks and is perceived as exactly equivalent to the old one. Youād have to flag something about ālabsā for it to be even perceived as a different product in the first place.
Even if you grant generous odds at each stage, a chain of individually plausible steps can still produce an unlikely outcome.
I think this may not be the right way to look at this, not just because they may be correlated, but also because it is not a 1-shot event. There will likely be a back and forth of events with bans and reversals, say, until some stable equilibrium is achieved. I think a more useful question to ask is how to stack the probabilities in favour of a given equilibrium.
AGI could, in principle, find solutions for the key problems that animals face, but I would argue the main issue is that it wonāt automatically enlighten humans.
I think one could argue that creating an index across causes makes sense, because it allows for exposure on things that are hard to compare (eg, we are not perfect utilitarians). I think in practical terms, it would help EAs have some reference and an easy way to donate across causes. For example, one could create a fund that is indexed to the elicited preferences of the EA community, or to some group of experts. The closest I am aware of is what Giving What We Can do.
This may be a bit of a conceptual vs purpose difference, but I would certainly not put insects as a good alternative even on simple moral grounds. Eg, it is likely not the type of alternative protein Iād be looking for.
I donāt know about other folks but I think this is my first criticism of them as long as I can remember, both online and offline. In general I think they have been fairly responsible with AI safety, or as responsible as I would expect a company to be. But even if I did criticise them a lot, I think it would still be a valid criticism. After all, as a non American I feel quite unease about this, even if they are arguably not the main actor. In any case, I think liberal democracies should oppose mass surveillance in general.
Worth noting that the mass surveillance friction point is only about domestic mass surveillance. Thus, does Anthropic believes mass surveillance of non-Americans is just fine?
Hey David! This is great, thanks for sharing, Iāll take a more detailed look. For now, some comments:
I agree that a more detailed Monte Carlo than the one I did is desirable.
I also think it is reasonable to present a trio of scenarios (neutral, optimistic, pessimistic), though science is somewhat tricky in that what matters is āwhat is reasonably achievableā, not the average of efforts. Eg, we do not measure AI progress by the average AI model, but the best. Thus, my bias towards optimistic figures and towards considering Pasitkaās numbers as more relevant than Humbirdās. Additionally, the scaling issues also seem more relevant for the bioreactor than for the media, so I donāt think a lack of large-scale commercial proof is strong evidence against Pasitkaās numbers being achievable. In general, my best guess is that $0.3/āL to $0.2/āL is likely to be achievable for the media, even if it is not easy. And I think that is ultimately the interesting question to answer.
With regards to the bioreactor, it seems progress has been happening there too, (see eg https://āāparima.bio/āāblog/āāparima-demonstrates-ton-scale-cultivated-meat-production-at-viable-unit-economics). Interestingly, they claim to have used a 22k L bioreactor, which is roughly the large scale that Humbird and Pasitka had envisioned. Again, not an insider, but it seems we live in the āoptimisticā world.
Last: my $15 to $30 estimate was an eyeball from the results I had, it was just meant to be an anchor for peopleās minds.