TLDR

As part of our ongoing Pivotal Questions initiative, The Unjournal is framing a set of specific questions to help assess the economic viability of cultured meat (CM) and inform funding choices. We’re seeking feedback on our proposed operationalizations for these questions, integrating these as predictions in our Metaculus forecasting community space, and prioritizing research for evaluation.

Project Context

As noted in our introductory post, this project aims to elicit and refine high-value “Pivotal Questions” from global-impact-focused organizations, curate and evaluate relevant research on these, and inform key funding and policy decisions. Our proposed process:^[1]

1. Elicit ‘target questions’ from impact-focused organizations
2. Select the most useful target questions, get feedback, and collaboratively refine them to make them more precise, tractable, and decision-relevant
3. Elicit stakeholder and expert beliefs over these questions, including through (Metaculus/Minitaculus) prediction markets
4. Source and prioritize research informing the target questions
5. Commission expert evaluations of this research, focused on how the research informs the target questions, and eliciting beliefs over these
6. Get feedback from research authors and the target organization(s), and updated beliefs about the target questions
7. Prepare a ‘synthesis report’, quantifying our ~consensus updated beliefs about the questions, and resolving the Minitaculus markets
8. Complete and publish the ‘target question evaluation packages’

Based on conversations with a range of organizations (see list here), we sketched 19 potential pivotal questions (see here). We’ve prioritized four of these for our pilot, broadly labeled:

1. “How do plant-based products substitute for animal products (re their welfare footprint)?”; suggested by Animal Charity Evaluators
2. DALY/WELLBY interconvertibility; suggested by Founders Pledge
3. Measuring Well-being/WELLBY reliability; suggested by Founders Pledge
4. Cell-cultured meat cost and price; internally suggested (but linked to stated priorities of several organizations; see “Decision-Relevance” below)

We’re currently in steps 2-3 for the cultured meat question – gathering feedback and beginning to post questions for predictions. At the same time, we’re sourcing research and commissioning one preliminary evaluation package (beginning steps 4-5). We’ve identified an initial narrow set of operationalized questions surrounding the production and cost of cultured meat, which we expect to provide high value-of-information. This post, as well as our new Minitaculus page, is an open invitation for others interested in this question to share their thoughts on these operationalizations, suggest related research for us to consider, and participate in forecasting and ‘belief elicitation’.

We’re also sharing this to illustrate and get feedback on our process for the Pivotal Questions project in general. We plan to share more details on the other three pilot questions soon.

Background on Cultured Meat

Despite the serious consequences of animal agriculture, global demand for meat is projected to increase substantially. Cultured meat, in theory, seems like the ideal solution to this quandary. It could drastically reduce the animal welfare (as well as environmental) cost of meat production without requiring major shifts in consumer behavior. However, the current cost of producing cultured meat and the technical challenges of scaling have cast doubt on whether the product could ever be cost-competitive enough with conventional meat to allow mass adoption.

Existing Research

Previous research on the viability of cultured meat has emphasized that significant cost barriers and biotechnical limits would need to be overcome for CM to become even somewhat competitive with conventional meat. David Humbird’s 2020 techno-economic analysis (TEA), commissioned by Open Philanthropy, is one of the most well-known and influential reports on the topic. Humbird concludes that CM is highly unlikely to reach price parity with conventional meat, largely due to issues like the cost of cell media, significant capital expenditure, and biophysical limits on cell density in larger bioreactors. This aligns with a previous TEA by Risner et al (2020), which modelled that the cost per kilogram of cultured meat would be in the thousands of dollars, except in an idealized best-case-scenario in which all possible technical barriers are solved.

Building from these TEAs, Neil Dullaghan and Linch Zhang from Rethink Priorities posted two reports, one comparing these analyses and one conducting a forecasting exercise, to provide further clarity on CM’s potential. Their synthesis generally supported Humbird’s conclusion that CM is unlikely to reach substantial production volumes.

However, some newer reports have challenged the limits cited in these TEAs. For instance, a 2024 paper by Pasitka et al demonstrates a continuous production process that could bring the price of cultivated chicken to $6.2/lb, which is comparable to organic chicken. Both GFI and Lever VC, a venture capital firm investing in food technology, claim that cell media expenses have fallen significantly over the course of just a few years. Furthermore, some journalistic sources have reported progress in reducing media costs, bioreactor investments, and other major cost drivers. These might represent innovations that could bring cultured meat to market on a large scale, but since many of these sources have a financial interest in CM’s success, their claims merit extra scrutiny.

Formulating the pivotal question

The viability of cell-cultured meat seems to be important for a great deal of funding and attention in the effective animal welfare space. We came to this specific question, with a focus on cost, particularly because of the attention and apparent impact of the Humbird TEA and the Dullaghan and Zhang posts. We see substantial research on this, and some signs that this research has affected funding. Given the apparent lack of consensus on CM’s potential to replace conventional meat, we see an opportunity for expert evaluation to guide future investments.

We worked in a Google doc^[2] to formulate, consider, and operationalize the pivotal question, largely following our overall draft template.^[3] Below, we clarify the thought process outlined in these documents.  

Path to Impact/Goals/Model

To understand the value of addressing a question, we need to know what decisions it might inform, and how. We can frame our goals in terms of the outcomes we care about, and model the determinants of those outcomes. We took a step back to consider “if we had an omniscient understanding of this issue, what would we really like to know to most improve animal welfare? Considering this “god’s-eye view” helped us formulate our goal-focused questions below:

Goal-focused questions

 “What is the expected-value (and probability distribution) of the impact on animal welfare^[4] from funding the development of cell-cultured meat?”

This essentially states the direct criterion for an ~effective altruist’s decision to choose whether to allocate resources to this (see “decision relevance” below).
 

We could refine this question further to specifically consider:

(i) marginal (aka incremental) funding,

(ii) very high levels of funding, or

(iii) the impact relative to the best alternative animal welfare interventions.^[5]

So why not use these questions? They bring in a range of other considerations, e.g., consumer acceptance, population growth, animal welfare legislation, etc. These are all important, but we don’t want to spread too thin; we plan to leverage a fairly narrow set of expertise and (mainly TEA) research. We envision a modular approach to constructing actionable insights; developing building-blocks of knowledge while keeping the overall model and goals in sight.^[6] 

To put this in context, for most of our pivotal questions, we’re aiming to refine our understanding (i.e., concentrate our calibrated beliefs) over ~one part of an overall value model, taking other parameters as given.

Here, this model might be something like (arrows: causal impact)

Value model

1. Investment in cultured meat (CM) research & dev. → innovation and capacity-building → Cost of CM production (in each year)^[7]

2. [Background] Taste implications of CM and PBA → Share of CM in hybrid (plant + CM products)

3. [Background] Cost of CM production (by type & quality), [Other background factors] → [caveat^[8]] … availability, store prices of CM

4. [Background] (Consumer price), [quality, availability] of all CMs and hybrids →  (consumption of CMs/hybrids) → consumption of animal products
5. [Background]: Consumption of each animal product → animal welfare (footprint)

Footnote: definitions and notation.^[9] 

With this full understanding of the relationship, we could target the level of investment that yields the best animal welfare outcomes. But building that level of understanding is beyond any one project’s scope; we instead focus on clarifying one aspect of this model. Informing each part of the model will improve our predictions about ‘how investment in CM will impact animal welfare’, and improve our funding choices.

Motivated by the existing research and discussion, we decided to target a PQ focusing on element (1) of the above model, and to refine it as follows.

Initial formulation: “What is cell-cultured meat likely to cost over the next years and decades?... And how will this vary as a (causal) function of the level of investments made?”

Why ‘cost’?

As the above ‘Value Model’ suggests, the impact of investing in cultured meat depends on a range of factors, including consumer acceptance, legislative hurdles, and welfare weights. We think our pivotal questions and expert research evaluation work will add the most value with a narrower focus. We have chosen to focus on the production side, mainly because of what seems to be a strong, influential, and unresolved body of TEA research regarding CM, drawing on real-world practical data and evidence.^[10] This approach aligns well with our “quantified operationalized questions” approach.

Dullaghan & Zhang’s forecasting exercise focused on production volume as a measure of CM’s potential influence. Their goal, to inform “whether philanthropic dollars can be used to nudge cultured meat trajectories in a way that makes a significant positive impact on farmed animals,” is quite similar to ours. While quantity is a more direct indicator of success, the realized quantity of CM will also depend on a complex interplay among consumer preferences, demand elasticity, global population, and more. This could distract evaluators from focusing on the issues where the strong cost-focused TEA research base offers the greatest value.  

Cost vs price: Although previous forecasting and discussion put some emphasis on consumer price, there are reasons to think it might be less informative about the outcomes we care about, and less useful as a prediction target. Prices and quantities are jointly determined as part of an equilibrium of supply, demand, and other functions.^[11] For example, a low price could reflect a large supply and mass adoption, or it could result from a monopoly with a cost curve that starts fairly low, but with highly price-sensitive demand and a low output.

In contrast, the TEA work seems well-suited to make inferences about cost. Our chosen formulation, ~“average cost at an efficient scale” represents the minimum viable price. Along with marginal cost and other factors, average cost helps inform predictions for price and volume. As such, we consider cost a key building block that can then inform further studies and predictions of CM outcomes.

Causal impact? (footnote)^[12]

Decision-relevance

Is this question useful? When eliciting target questions from organizations, we also ask about the decisions these questions will inform. This helps us clarify the actionable value of our work, hone in on the most relevant part of the value model, and formulate the pivotal question.

While we don’t yet have a partner organization for this PQ, we sketched some possible decisions that we expect will be relevant to potential CM funders, and how we expect our project might inform those choices. We maintain an animal-welfare-focused perspective for this project; although cultured meat has significant environmental and food security implications, these seem to be of a lesser magnitude (rough intuition in footnote).^[13] Most of the EA organizations that fund CM development seem to justify this primarily based on animal welfare concerns (e.g., Open Philanthropy, EA Funds). 

1. Decision: “How much should we invest in the development of CM”?

   Value of Information (VOI) – How might the “cost of CM” question inform this?

   If CM has significant potential to become cost-competitive with conventional meat, increased funding for R&D could accelerate breakthroughs and hasten consumer adoption. Otherwise, funding might be more impactful going towards other animal welfare interventions. Furthermore, if cultured meat is already very unlikely (or very likely) to become economically viable, the impact of further funding may be quite low compared to the impact of funding other animal welfare interventions.^[14] 

2. Decision: “Which R&D areas should we  fund?”

   While the overall “cost of CM” forecast may not directly inform this, the modeling and TEA work is likely to provide insights. Understanding the key cost drivers and potential blockers could help guide R&D and other interventions to be more fruitful. E.g., if the most tractable factor that substantially drives costs is whether we can use food-grade versus pharmaceutical grade bioreactors, bioengineering research could focus on ways to make the former safer, and legal/policy teams could strategize how to convince regulators to allow this use.

3. Decision: “How should we allocate additional animal welfare funding?”

   The probability of the mass adoption of cultured meat is likely to affect the value of other animal welfare interventions. For instance, if CM is fairly likely to reach price-parity (or cost-parity) given achievable levels of investment, there might be less value in developing plant-based meat alternatives. Furthermore, if CM is expected to completely replace animal-derived meat soon, lobbying for animal welfare reforms could be even less relevant, as these reforms would only matter in the short term.

   Although we will consider all three paths, we will emphasize the first path (‘how much to invest’). This is most directly tied to the question we initially formulated. It is foundational, high-value, and quantifiable.

Operationalizing this Question

We believe that we’re converging on a clear, high-impact question that addresses key decisions surrounding the funding of cultured meat and animal welfare policy in general. We next try to narrow down the scope of this question, to make it more fruitful for evaluators and forecasters to address.

Proposed operationalization

What will be the average production cost (per edible kg) of cultured chicken meat in the given years, expressed in value units, across all large-scale plants in the world?”

More precisely defining the italicized terms and explaining our choices:^[15] 

• Average production cost (AC): This represents the capital investment (annualized charge) and CM-related operating costs of a given firm divided by its annual output of edible CM (in kg). AC is a useful measure because it captures the two most significant, well-studied, and predictable expenses of CM production (bioreactor investment and media cost) while also approximating the minimum sustainable price of CM.

  • Economic theory: Under ‘best-case’ conditions,^[16] firms operate at the minimum point of their average total cost curves^[17] and prices converge to this level, yielding zero economic profits. This is the ~lowest sustainable price in a free market.^[18]

  • On the other hand, if firms have market power (e.g, with barriers to entry), they will set higher prices, absent regulation.^[19] AC still represents an informative lower bound.^[20]

• Cultured chicken meat: Although most TEAs focus on bovine or other mammalian cells, chicken has a greater relevance to animal welfare considerations, both in the number of individuals and the average quality of life.

• Years: We are targeting 2031, 2036, and 2051 as short, medium, and long-term timelines, respectively.^[21] This keeps us in line with Rethink Priorities’ previous forecasting work, allowing us to better compare predictions.

• Value units: In inflation-adjusted (CPI) 2025 US dollars.

  • Alt.: price relative to animal-based chicken (see footnote)^[22]

     

We propose two further cost measures to help us consider additional scenarios and policy questions:

1. Marginal Cost at Scale. While AC incorporates the capital expenditure involved in getting facilities up and running, marginal cost solely captures the cost of producing each additional kilogram of product. While these marginal costs are unlikely to reflect the sustainable market price of CM, they are important for modeling the sensitivity to market power and the nature of competition, and to considering the marginal impact of subsidies.

2. Average Cost Function. Careful forecasters are likely to model the relationship between costs plant capacity, and global production (e.g., a  power law scaling).^[23] Explicitly stating this would help us consider how robust the AC forecast is to alternative assumptions, and to extrapolate to different scenarios.

Specific cost elements

The following sub-questions represent some of the most significant cost drivers in the production of cultured meat. Nearly every TEA cites cell media as the most costly component of CM, which is why so many of these questions explore the likelihood that workarounds and cost reductions will be developed.

1. By 2036, what percent of commercial cultured meat will be produced using food-grade cell media (as opposed to pharmaceutical-grade)?

2. What will be the average cost of recombinant growth factors per gram^[24] (of input) in commercial cultured meat production in 2036?

3. What will be the cost of cell media per kg of cultured meat in 2036?
4. Will most cultured meat (by volume) be produced with plant hydrolysates in 2036?
5. Conditional on  “most CM is produced with plant hydrolysates in 2036?”, what will be the average price of 1 kg of cultured meat?
6. Conditional on  “most CM is NOT produced with plant hydrolysates in 2036?”, what will be the average price of 1 kg of cultured meat?
7. What will be the average quantity of cell media, in liters, necessary to produce 1kg of cultured meat in 2036?
8. What will be the highest cell density of cultivated meat produced in a 20,000 liter bioreactor by 2036?
9. What share of cultured meat companies (those with capex over XXX amount) will design and build their own bioreactors by 2036? (This is motivated by the assumption, reported here, that doing so will be significantly less expensive than purchasing off-the-shelf pharmaceutical bioreactors)

10. (Possibly) What is the cost of capital/’required rate of return’ in this context^[25]

Additional Cruxes and Outcomes

These questions take a big-picture approach to forecasting CM

• What will the cumulative investment in CM be by the aforementioned dates?
• What will the average global production cost/kg be in [the Medium run], conditional on low, medium, or high cumulative investment in CM by [Medium run]
• How many metric tons of cultured meat on its own or as an input into a hybrid product will be sold in the [Medium-run year] and [Longer-run year]
• Adapted from Rethink Priorities Forecasting post, “What is the probability that impartial effective animal advocacy evaluators in 2036 will view having spent $10M over the previous 10 years on cultured meat as better than a dollar-weighted average of Open Philanthropy’s existing farmed animal welfare donations”?

Next Stages

As described above (Project Context), as we converge on a headline pivotal question, we aim to elicit beliefs from stakeholders, experts, and the wisdom of the crowds, curate research, have this research evaluated by experts, update beliefs, and synthesize this work and its implications. We’re pushing forward on several of these steps. Some quick updates…

Metaculus questions

Objectively verifiable questions/Opinionated resolution

We are posting new questions and curating existing questions in our community space, provided by Metaculus; these spaces offer more flexibility for opinionated resolution, resembling the Manifold’s approach.

E.g., the community question posted here: “What will be the average production cost (per edible kg) of cultured chicken meat at the end of the following years…, across all large-scale plants in the world”?

These will be resolved by the question curator (David Reinstein, for now). We sketch our plan to use LLM’s as a ~less biased and low-cost source of evidence for this (see Ozzie’s more extensive discussion of this here.) However, we are considering adjusting this to use a more explicit set of criteria/resolution criteria, including likely data sources. This would allow us to get the questions approved for the general Metaculus feed, for more exposure.

“Predict the expert judgment” questions

As part of the belief elicitation, we will also ask our expert team and evaluators, as well as key stakeholders, to weigh in and give their probability distribution forecasts on the Metaculus platform at several points in time. This will enable further objectively-resolved questions such as “What will be the midpoint of the aggregated belief distribution of The Unjournal’s evaluators and experts team over the question ‘What will be the average production cost’...” This can be applied to questions that can never be definitively and objectively resolved (e.g., ‘what is the true animal welfare impact…’)  It can also be used to resolve questions far in advance of the actual event being forecasted. E.g., we could resolve “(beliefs about) average production cost in 2051” at the end of 2026.

We plan to mathematically aggregate the stated belief distributions both for this and for our synthesis report. We’re leaning towards a linear weighting scheme^[26] with equal weights; but considering performance-based weighting for future work^[27]

This is similar to what game theorists call a ‘beauty contest’, where participants are rewarded for stating ‘the most popular value’ or as close as possible to the average participant’s statement. It also resembles the ‘reciprocal scoring’ used by Karger et al (2021).  However, here we are asking participants to predict the ~average value of the smaller group of experts.  If participants trust that the experts will have a chance of correctly incorporating all potentially available information and insight, this might be an incentive-compatible way to elicit the ‘wisdom of the crowds’.

Tournament

To encourage participation, we plan to sponsor a Metaculus tournament (with monetary prizes) for forecasting the above questions. We plan to announce details soon.

Update 5 Aug 2025: Due to budget constraints, we will not be sponsoring a Metaculus tournament at this point in time, but we are seeking further funding to enable this soon. 

Curating literature (TEAs and more)

We’ve done some review and curation of the relevant research, relying on our familiarity with this work as well as leveraging LLM tools like elicit.com.

The research we are considering passing on to evaluators includes the aforementioned Rethink Priorities work as well as TEAs, syntheses, and associated analysis (see footnote)^[28].

Expert evaluation of PQ and specific research

We’ve adjusted our initial proposal to give our evaluators a greater role in assessing and comparing the literature as a whole, rather than only evaluating a single paper/project.  

We’re broadening the scope and increasing the compensation. We will ask them to (approximately):

1. State your initial beliefs about the PQs and underlying factors
2. Consider a set of  TEAs/papers, ranking and discussing their credibility and relevance
3. Evaluate one TEA carefully, largely following the Unjournal applied stream guidelines but focusing on the implications for the Pivotal Questions.
4. State and justify your (revised) beliefs about the PQs in light of this evaluation and your work above.

Credit and Acknowledgements

David Reinstein co-wrote this post with Ashley Meader. We've benefited from discussions and updates from @Linch Zhang, @Neil_Dullaghan🔹,  @Jacob_Peacock , Jacob Schmeiss, @Trevor Woolley,  Zach Groff, Anca Hanea, Bob Kubinec, Christian Williams of Metaculus, Elliot Swartz of GFI, @Ozzie Gooen, David Manhein, Julian Jamison, Zuzana Sperlova, Vince Mak, anonymous potential evaluators, and members of The Unjournal team, among others.

1. ^{^}

    The parts in italics were outlined in our plan but not included in the original post.

    

2. ^{^}

    We are cleaning up this Gdoc and removing sensitive information (names of potential evaluators etc.), and aim to share and link it here.

    

3. ^{^}

    Some aspects of that template were less relevant here,  in part because we proposed this question, not a stakeholder.

    

^{^}

 “On animal welfare” – we could also include human welfare through channels like health and global warming. However, we expect the impact on non-human animals to be dominant here, at least if one puts moderately small welfare weights on chickens, pigs, etc.

 

^{^}

 E.g., “What is the expected ratio of the effectiveness of CM funding relative to a ‘dollar-weighted average of Open Philanthropy’s existing farmed animal welfare donations’”. (Partly from D&Z’s forecasting study).
 

^{^}

 If we could get an informativeanswer to a ‘goal-focused’ question – e.g., the expected animal welfare value of subsidizing cell-cultured meat (relative to what we think is the best alternative in this space, perhaps corporate AW welfare campaigns) – this might be sufficient for the funding or policy decision we're looking to make now. But this answer may be a function of several sub-questions. Even if we can't answer all of those sub-questions now, each may (1) be informative (causing beliefs updates) and (2) be a building block towards future insight. Sub-questions like (a) the predicted cost of cultured meat for different subsidy levels over time, (b) the expected substitution impact on animal products for each price, (c) the benefit/costs of corporate AW campaigns, etc. Having a more precise estimate of (a) could substantially concentrate and inform our beliefs about the goal-focused question. We can continue to build this as we learn more about (b) and (c). And (a) may still be useful if our underlying model adjusts, e.g., if other AW intervention alternatives rise to the fore, if  our timeline changes, etc. (a) may also inform related questions surrounding environmental impacts, impacts of transformative AI, etc.

 

^{^}

  We might consider a cost curve as a function of both scale and experience, both globally and at the plant level.  Research discoveries could shift this downwards, allowing lower-cost production for each level of scale and experience. General investment might also build scale and capacity, and subsidize initial production to enable learning-by-doing gains.

 

^{^}

 This is a slightly incorrect simplification. Considering consumers and producers acting to maximize their profit or utility…,   Price does not ‘causally determine’ quantity; quantity and price will be determined simultaneously by the combination of production functions (cost curves), market structure, consumer preferences (demand), and any taxes or regulations.

 

^{^}

Variables in bold are ‘levers’ potentially within the control of animal welfare advocates, funders, and CM producers.

Variables in parentheses may depend on factors outside of this control, like supermarket markups, shelf placement, etc.

‘Background’: Aspects of the functional relationship that are outside the bounds of the current investigation. The current PQ is more focused on the cost of CM/hybrids as a function of investment and time.
 

^{^}

  In contrast, while there is some research on attitudes towards these products and willingness to purchase them, this largely relies on hypothetical survey data about a product people are not familiar with.

 

^{^}

 More precisely, an outcome of consumer preferences, production functions, constraints, market structure, and government intervention.

 

^{^}

As suggested above, just knowing how much people think the cost of producing cultured meat is likely to decline does not tell us that investing in its development will make much of a difference. Ideally, we want to know how much each dollar we spend will speed this up relative to the counterfactual. But this is extremely challenging, bringing in  a range of additional hard-to-measure considerations about the impact of research and development. While our operationalized questions mainly focus on cost predictions, we do introduce a conditional cost question depending on investment. This is not entirely causal, but plausibly gets close to it.

 

^{^}

 Even if climate change is considered as high a priority as farmed animal welfare, livestock farming contributes only about 15% of human greenhouse gas emissions, and even if cultured meat replaces this, it may continue to contribute a substantial share of these emissions (e.g., figure 1 in the synthesis here). Cultured meat is also technologically complicated and seems unlikely to provide a useful alternative food supply in the event of massive societal disruption or nuclear devastation.

 

^{^}

 Admittedly, this “humped-shape relationship” is a based on a loose argument; some of the “additional” questions we pose below get more specifically at the difference in outcomes with low versus high levels of investment.

 

^{^}

 While we try to be precise, in making predictions and stating beliefs for these questions, we will advise evaluators and stakeholders to aim at the most meaningful and useful interpretations and not dig into the wording for loopholes.

 

^{^}

Including free entry and exit of firms and convex cost curves.
 

^{^}

 And, with convex cost, this minimum occurs where the average cost and marginal cost curves intersect, thus marginal cost pricing. 
 

^{^}

 This doesn’t hold if we have diminishing marginal cost everywhere, i.e., ‘global returns to scale’. But in this context, given the constraints of tank sizes and plant sizes, I don’t expect increase returns to scale to be substantial after a reasonably large size.

 

^{^}

 Governments could also choose to set price ceilings, which could be sustainable and welfare-increasing as long as they are set above this average cost (at scale).  
 

^{^}

 Otherwise the ‘markups’ are seen to determined relative to marginal cost, depending on the number of firms, on the nature of their competition (e.g., price-setting vs. quantity setting), and on the demand elasticity.

 

^{^}

We also include end of 2026 for for some of the forecasting work, to have a faster feedback loop, e.g., for Metaculus predictions. However, we expect this very short run indicator to be less informative overall.

 

^{^}

 Alternatively, we might express this relative to the average global cost of the comparable meat product, say average chicken meat (whole chicken minus bones etc.). This comparative metric might be more useful if we anticipate large changes in the structure of the economy.
 

^{^}

As we are also asking the evaluators to consider the average plant capacity and global production, see below, plugging these  into their cost curve will yield the predicted average cost. Global production volume may matter, e.g.,  because of nonexcludable learning-by-doing effects.

 

^{^}

 This is formulated as grams per input to focus on the cost of the raw material itself without requiring consideration of different cell lines' metabolic rates. 
 

^{^}

 This is important overall because if we have a model of the cost of production year by year and the reasonable markup levels, this tells us how much investment will be supported by the future expected payoff stream.  How could we best operationalize this? IIRC D&Z asked something about the ‘number of years to recover the investment’. The required rate of return (RoR) will depend on how investors perceive the (market-correlated) riskiness of this project, as well as on the prevailing interest rates at the time.

 

^{^}

 With log/multiplicative aggregation as a robustness check.

 

^{^}

  We’re considering  what sort of belief aggregation to commit to; ideally the chosen approach should be posted along with the Metaculus question. Note most questions take on continuous values, so people will specify entire belief distributions.
 

Deciding between things like the following – leaning towards option A for now.   I give some loose justifications and considerations below

A. Linear opinion pool — averaging assigned probabilities of each value,  weighting all equally

• Recommended by Anca Hanea
• Simple to calculate
• Represents Bayesian updating if each forecaster observes a noisy measurement of the true value, plus a normally distributed error-term.  (And presumably, they should then report a normally distributed belief.)
• Weights everyone ‘the same’, less sensitive to overconfidence


B.  Logarithmic opinion pool — multiplying assigned probabilities of each value (with normalization), weighting all equally

• Metaculus uses this (which could also help with the interface and implementation)
• Coherent with their scoring rule (not sure why I should care about this)
• ~Simple to justify — it’s Bayesian updating if all forecasters are ‘independent’.  But I’m struggling with getting an intution for what ‘independence’ means in this context. And this doesn’t seem so plausible as they will all be looking at the same information and updates.


C. Something in between linear and log like a ‘power mean’

• Both groups probably  have reasonable justifications for what they do, so the ‘truth’ is probably intermediate
  • … but how to choose a reasonable power to raise this to? Something ad-hoc like ½?
     

D. Any of the above, but weighted by previous performance (perhaps weights based on their previous cumulative log likelihood score)

• But not all forecasters may have defined scores
• And they don’t have the same experience … some may have forecast easier things
   

E. Kick the can down the road and say something like

 The “consensus” will be stated as a probability distribution aggregating the stated beliefs and uncertainty of all participants in a way deemed to be most suitable by David Reinstein, consulting with experts in these methods such as Anca Hanea. (The median of this belief distribution will be used to resolve the relevant Metaculus questions)

 

^{^}

 E.g.,

“Techno-economic modeling and assessment of cultivated meat: Impact of production bioreactor scale” by Negulescu, et al (2022).  [?2023].

“Scale-up economics for cultured meat” by David Humbird (2021).

“TEA of Cultivated Meat” by CE Delft (2021), commissioned by GFI

“Technological prospecting: the case of cultured meat” by Fernandes et al (2022).

 “Empirical economic analysis shows cost-effective continuous manufacturing of cultivated chicken using animal-free medium”

“Bringing cultured meat to market: Technical, socio-political, and regulatory challenges in cellular agriculture”

GFI has also mentioned some upcoming work they believe will make a substantial impact, but this work is unlikely to be released before “mid to late September”.  

Show all footnotes