Matthew Stork

79Joined Oct 2021

Comments
9

Agree with your post and want to add one thing. Ultimately this was a failure of the EA ideas more so than the EA community. SBF used EA ideas as a justification for his actions. Very few EAs would condone his amoral stance w.r.t. business ethics, but business ethics isn't really a central part of EA ideas. Ultimately, I think the main failure was EAs failing to adequately condemn naive utilitarianism. 

I think back to the old Scott Alexander post about the rationalist community: Yes, We Have Noticed The Skulls | Slate Star Codex. I think he makes a valid point, that the rationalist community has tried to address the obvious failure modes of rationalism. This is also true of the EA community, in that there has absolutely been some criticism of galaxy brained naive utilitarianism. However, there is a certain defensiveness in Scott's post, an annoyance that people keep bringing up past failure modes even though rationalists try really hard to not fail that way again. I suspect this same defensiveness may have played a role in EA culture. Utilitarianism has always been criticized for the potential that it could be used to justify...well, SBF-style behavior. EAs can argue that we have newer and better formulations of utilitarianism / moral theory that don't run into that problem, and this is true (in theory). However, I do suspect that this topic was undervalued in the EA community, simply because we were super annoyed at critics that keep harping on the risks of naive utilitarianism even though clearly no real EA actually endorses naive utilitarianism. 

I appreciate the response here and want to clarify my argument a bit. I totally understand that currently available SUT isn't sufficient to make cultured meat cost-effective. I'm mostly arguing against the notion that these problems are intractable. To your point about the difficulties with gamma irradiation, it seems likely that there could be a reasonable alternative to gamma irradiation for SUT sterilization. At the moment, pharma companies get by just fine using stainless steel for processes > 2kL, so there isn't much pressure to improve from that angle. If single-use is truly enabling for cultured meat, then that provides an impetus for more investment in improved sterilization technologies.

The purported economic and environmental benefits of SUT are related to the elimination of sterilization steam (because everything is gamma-sterilized before shipping) and the elimination of cleaning chemicals (because the bags are not cleaned for reuse). 

The major cost savings I see for a cultured meat plant would be in a reduction in the requirements for air quality. A fully single use plant with completely aseptic connections can (in-theory) be run aseptically without a clean room. There would just be a small clean room for media and solution prep. Existing pharma plants using SUT tend to still need high quality air as there are some steps in the process that require manual manipulation. I've seen it suggested though that future biopharma processes which use fully integrated and continuous systems can be run  in clean rooms with drastically lower air quality than existing plants.

Combined with 100% manual unpacking, setting, connect/disconnect, and teardown of bags, the single-use idea seemed very much at odds with the fully automatic plant that many propose.

Moving to long duration perfusion (> 30 days) reduces the need for unpacking / teardown. There have been biopharma companies which have demonstrated the ability to run stable perfusion for up to 60 days. For the most part, companies haven't gone longer than that mostly because it's not really necessary.

What is missing to me is an explanation of exactly how your suggestions would prevent a future SBF situation. It's not really clear to me that this is true. The crux of your argument seems to come from this paragraph:

The community was trusting - in this case, much too trusting. And people have said that they trusted the apparent (but illusory) consensus of EAs about FTX. I am one of them. We were all too trusting of someone who, according to several reports, had a history of breaking rules and cheating others, including an acrimonious split that happened early on at Alameda, and evidently more recently frontrunning. But the people who raised flags were evidently ignored, or in other cases feared being pariahs for speaking out more publicly.

Would this have been any different if EA consisted of an archipelago of affiliated groups? If anything, Whistleblowing is easier in a large group since you have a network of folks you can contain to raise the alarm. Without a global EA group, who exactly do the ex-Alameda folks complain to? I guess they could talk to a journalist or something, but "trading firm CEO is kind of an amoral dick" isn't really newsworthy (I'd say that's probably the default assumption).

I also generally disagree that making EA more low trust is a good idea. It's pretty well established that low trust societies have more crime and corruption than high trust societies. In that sense, making EA more low trust seems counterproductive to prevent SBF v2.0. In a low trust society, trust is typically reserved for your immediate community. This has obvious problems though! Making trust community-based (i.e. only trusting people in my immediate EA community) seems worse than making trust idea-based (i.e. trusting anyone that espouses shared EA values). People are more likely to defend bad actors if they consider them to be part of their in-group.

To be honest, I'd recommend the exact opposite course of action: make EA even more high trust. High trust societies succeed by binding members to a common consensus on ethics and morality. EAs need to be clearer about our expectations are with regard to ethics. It was apparently not clear to SBF that being a part of the EA community means adherence to a set of norms outside of naive utilitarian calculus. The EA community should emphatically state our norms and expectations. The corollary to that is that members that break the rules must be called-out and potentially even banished from the group. 

Coming to this pretty late, but I'm curious - does the success of Paxlovid for COVID change your views on this? It took ~21 months from the start of the program to have the drug approved under an EUA. So not as fast as the vaccines, but still relatively fast. Efficacy is pretty amazing at ~90% reduction in severe illness and death (in unvaccinated populations). 

Makes sense, thanks for the context!

Having never purchased a SUB myself, I don't know the exact cost, but $30k for 1000L bag seems ballpark correct for biopharma. That all said, it's just a big 'ol  plastic bag, I would bet that a food-grade version could be at least 10x cheaper.

Looking at the materials (Single-use bioreactor - Wikipedia), these are all very typical polymers that should be quite cheap (in theory at least). You could get away with food grade plastics, which should be significantly cheaper than pharma grade.

SUBs have also only become popular relatively recently. My major biopharma employer only started using SUBs in a significant fashion in the last decade. My point being, I bet there is still room for improvement in manufacturing costs since SUBs are a relatively recent innovation.

Just to hedge a bit, it's a bit hard for me to give you exact numbers here in terms of how much room for improvement there truly is. I'm not a materials scientist and don't have intimate knowledge of SUB construction. It's entirely possible there is some super expensive step required to make those things, that just can't feasibly be reduced.

EDIT: This analysis is probably wrong, see the reply. Leaving post up for context. 

One other comment, related to the Twitter thread that brought me here showing that poultry is 13% efficient at conversion of feed into meat. While this is true, feed costs are a pretty minor cost at least for poultry: NC-Choices-NC-Farm-School-Meat-Chicken-Info-graphic-Breakout.pdf (ncsu.edu)

I'm seeing the cost for conventional feed as $0.89 per lb poultry, compared to total costs of $6.42 per lb, so the feed is only ~14% of the total cost per lb poultry. I suspect that the feed cost is probably far higher for cattle, but that's because cattle are only about 3% efficient in terms of calorie conversion, far less efficient than poultry.

Admittedly, I do agree with your general argument, at least when it comes to poultry. It will be tough to be more than 13% energy efficient for cultured meat. However, the other costs of poultry production are high enough that is still seems reasonable that a cultured meat could still be competitive. 

I work in bioprocessing, and overall wasn't that impressed by the Humbird report. The plant Humbird proposes looks nearly exactly like existing bioprocessing plants for biologic drugs. It shouldn't surprise anyone that trying to make (cheap) food isn't possible using plants that are designed to make extremely high value drugs. Let's take insulin, a relatively cheap biologic drug. One unit of insulin (0.035 mg) retails for about $0.30. Let's assume during the fermentation we get 20% cell weight, 10 g insulin/L, and have a 50% downstream yield. This means that we have 1.4 mg of cell mass per unit of insulin, for a cost of ~$0.20 per mg cell mass, which is $200,000 per kg. A reasonable target for cultured meat might be a retail price of $20 / kg. We are talking a 10,000x difference here, so it's clear to me that a cultured meat plant will need to be drastically different than an insulin plant or an antibody plant. To be fair, I'm not exactly sure what the ideal design would be, just that any plant that looks exactly like a biopharma plant (as Humbird assumes) is not feasible.

Keeping things sterile would probably be the hardest element to scale up for culture meat production in my opinion. Equipment/media sterilization is only sufficient if you run a completely closed process. However, keeping your process completely closed is quite difficult, especially if you are using fixed stainless steel equipment. You are eventually going to have to open up your equipment for cleaning and turnover, and you need literal 100% sterility. Even a tiny bacterial contamination will likely spread to contaminate your whole line, since bacteria grow so much faster than animal cells.

Due to these challenges, basically every biopharma plant uses large sterile suites where the rooms have ISO  7/8 quality air. This is really expensive, and was one of the major expenses in the Humbird report. 

However, where all the existing TEAs fail in my opinion is by ignoring single-use materials. It is potentially possible to have a truly closed system using single-use materials (i.e. plastic flowpaths and bioreactors) that are pre-assembled and pre-sterilized. This would allow you to put this process in a plant with more minimal air purification, which drastically reduces your costs. Single-use bioreactors (SUBs) also have drastically lower capital costs compared to stainless steel, something like 70% lower. The downside of course is that your consumables costs as you need to purchase these plastic bioreactors. As of right now, they also don't make 20k L SUBs, the largest I've seen is 6k L. I imagine larger SUBs would be doable if there was sufficient demand.

Hey all, I've got some thoughts here. I worked in biopharma and spent some time working on a high-cell density perfusion system with CHO cells, so I've got some unique insight here. Right now for biologics production, there is a ton of effort in developing medium-scale perfusion processes run in single-use bioreactors (i.e. plastic bags). None of the analyses focused on this mode of production. There are potentially massive benefits to this approach. 

  1. Single-use equipment solves the sterility issue. Single-use equipment can be pre-assembled and sterilized in place. This allows for plants that don't maintain strict sterility, saving on capital costs and operational costs. Note that this is not standard right now; biopharma plants with single-use equipment still usually maintain high air quality. However, for food-grade applications I bet you could get away without high quality air if you relied on single-use equipment.
  2. There are massively reduced capital costs when using single-use equipment. I've heard that single-use bioreactors (SUBs) can cut CAPEX by ~70% or so, although I imagine these numbers are inflated by companies advertising the sale of SUBs. Still, this puts us much closer to a reasonable cost. Of course, this is offset somewhat by the cost of the bags. However, with a sufficiently long perfusion culture (maybe say 60 days), you can cut the amount of single-use materials required. You also save on cleaning costs, as you can dispose of single-use equipment after use. There is potentially a higher labor cost here, as you will need labor to remove old equipment and install new equipment between runs. This is likely offset somewhat by less labor cost for cleaning.
  3. Humbird suggests using ATF membranes for cell separation in perfusion, which is incredibly expensive (10% of the cost in his optimistic scenario). ATF and other membranes are the standard separation method for biopharma perfusion right now, but there are potentially better methods out there. Stuff like acoustic wave settling or Hydrocyclones could be significantly cheaper (full disclosure, I contributed in a small way to this hydrocyclone work).
  4. Spent media could be purified to recover and recycle any micronutrients. Macronutrient recovery would be harder, but is still potentially feasible. (To be fair, you could do this with any approach, but it's worth nothing)

The main downside is that right now SUBs are limited to ~2,000L. There are bags as large as 6000L out there, although I have no experience with them. However, in general the use of SUBs is relatively new in biopharma, so I expect there is still ample room for scale-up. In any case, this approach would likely require a larger number of smaller perfusion reactors running in parallel. 

I'm generally skeptical of the assertion that this will require "Nobel Prize winning" efforts to fix this. Honestly, sterility seems like the hardest problem of the bunch to me, which is why single-use materials seems like a winner. My gut tells me that it really shouldn't be that hard to get cheap amino acids from soy; this isn't based on a careful analysis mind you, it just really seems like this would be solvable if scaled-up sufficiently.