Interesting links for Jan 24 2023
Lab-grown meat; timeline for an AI to build its own tools; elites & experts; countries' prospects, given aging populations
Here’s a grab bag of links that I found to be interesting.
Can stem cell meat save the planet?
Most meat eaters will readily acknowledge, as they prepare to tuck into a porterhouse, that factory farming is cruel, and that it should be eliminated. And, if you point out the incongruity of this position given the porterhouse placed in front of them, most will readily, if somewhat uneasily, acknowledge the apparent hypocrisy. People like their meat. They also do not like factory farming.
People are mainly not intellectually consistent. They maintain their inconsistency even when it is pointed out to them.
But what if meat—chicken or lamb or steak or even fish—could be grown in a lab, and not as part of an animal? Here’s some detail:
Stem cells are taken from animals through a biopsy and are then frozen in liquid nitrogen to preserve them for several years. To produce meat, the cells are multiplied in a bioreactor. The technology isn’t quite yet ready for mass production, but theoretically, a single biopsy would be sufficient to produce hundreds of tons of meat.
The American startup Eat Just, based in Silicon Valley, is currently in the process of opening a laboratory in Singapore. The company’s focus is on producing chicken meat, which it plans to introduce to supermarkets in the coming years.
The attraction is obvious: if you can grow meat in a bio-reactor, all the concerns about factory farming and animal cruelty disappear. This has a certain intuitive appeal, and if Eat Just can scale up its production and reduce production costs, it has a chance at reducing people’s demand for factory farmed meat.
However, there is quite a bit of skepticism about whether this technology can ever be reliably scaled. And, if it can’t be scaled, then its production costs can’t decline to the point where factory farming is no longer viable. Here’s a long article about scaling issues with lab grown meat:
Their future doesn’t look good. Humbird worked off the assumption that the industry would grow to produce 100 kilotons per year worldwide—roughly the amount of plant-based “meat” produced in 2020. He found that even given those economies of scale, which would lower input and material costs to prices that don’t exist today, a facility producing roughly 6.8 kilotons of cultured meat per year would fail to create a cost-competitive product. Using large, 20,000 L reactors would result in a production cost of about $17 per pound of meat, according to the analysis. Relying on smaller, more medium-efficient perfusion reactors would be even pricier, resulting in a final cost of over $23 per pound.
Based on Humbird’s analysis of cell biology, process design, input expenses, capital costs, economies of scale, and other factors, these figures represent the lowest prices companies can expect. And if $17 per pound doesn’t sound too high, consider this: The final product would be a single-cell slurry, a mix of 30 percent animal cells and 70 percent water, suitable only for ground-meat-style products like burgers and nuggets. With markups being what they are, a $17 pound of ground cultivated meat at the factory quickly becomes $40 at the grocery store—or a $100 quarter-pounder at a restaurant. Anything resembling a steak would require additional production processes, introduce new engineering challenges, and ultimately contribute additional expense.
More research on scalability-related issues is available here. While I remain optimistic that this is a solvable problem, I also think that it’s important to acknowledge the very real problems that cultivate meat producers have to overcome in order to have a viable product that can be sold to the masses at a mass market price. If cultivated meat remains a luxury item, then its effect on factory farming and animal cruelty will be marginal.
The anonymous Twitter account notes this interesting prediction on Metaculus:
The forecast is here. The question was originally posted on December 30, 2016, and opinion has recently converged on 2028. The question reads, in part:
When will AI systems become sophisticated enough that they can build, to some specification, a system that can itself do sophisticated programming?
This question will resolve on the date when an AI system exists that could (if it chose to!) successfully comply with the request "build me a general-purpose programming system that can write from scratch a deep-learning system capable of transcribing human speech."
If this does not occur between January 1, 2018 to January 1, 2100, this question will resolve as Ambiguous.
Of course, “sophisticated programming” is fairly ambiguous. What counts as sophisticated programming, as opposed to unsophisticated programming? Is a compiler autonomously doing “sophisticated programming” when it translates code from one language to another?
Are elites displacing experts?
Robin Hanson writes:
Experts are people who are good at and know about particular things. They may be trained in them, or show a track record of accomplishment. We tend to defer to other experts in similar tasks to judge who is expert at something. We organize experts so that they can focus on the things at which they are best, and coordinate with other experts on related tasks. Expert talk tends to be precise, practical, concrete, and narrowly focused on particular tasks. Experts more engage detailed arguments and admit when they are wrong.
Elites are generally impressive people appropriate for leadership roles, who are accepted as such by both the masses and a wide range of other elites. Elites know less about particular technical tasks, and more about navigating social communities. Many kinds of impressiveness contribute to their eliteness, including wealth, beauty, intelligence, personality, connections, breadth, style, polish, and taste. Elite talk tends to be more artistic, stylish, and aspirational, and less precise, detailed, or logical.
I think this is useful framing. One consequence of this is that experts frequently are over-confident about their knowledge of things adjacent to their area of expertise. While elites execute on ideas, experts debate ideas.
Let me make this more concrete: I recently wrote about Bryan Caplan’s skepticism of ChatGPT. Caplan is an expert at teaching economics. And base rates appear frequently in the study of economics. It’s not too much of a stretch, then, for him to assert that paying attention to base rates is a good heuristic.
He argues from this observation that the base rate for over-hyped technology is something like 95%. Therefore, according to his heuristic, ChatGPT is more likely than not to be over-hyped. To those who think that ChatGPT is over-hyped, ignoring it or expressing skepticism about it seems like the more rational play.
I obviously disagree with Caplan’s view on this. I pay attention to all the entrepreneurial activity happening around ChatGPT and generative AIs more generally. Hundreds of tools are being built to allow people to use this technology to be more productive and efficient. My heuristic is: Pay attention to the elites who build and do, not to the experts who debate.
How much does aging really hurt a country?
Noah Smith writes:
If countries are smart, they’ll replenish their young populations with immigration. But even domestic political reluctance can be overcome, immigration is only a temporary stopgap; fertility is already fairly low everywhere except Africa, where it’s falling at an accelerating pace. In 1990 the average African woman was expected to have 6 children in her lifetime; today that’s down to 4, and the drop is accelerating.
In other words, it’s not just China. Every single country on Earth is either having to deal with population aging, or soon will have to deal with it.
So we need to ask: How much does this really matter? There’s a school of thought that says aging is no biggie — population decline may reduce total GDP, but per capita GDP could stay the same, since the denominator will be shrinking as well. Yes, there will be more retirees, but we’ll just replace their labor with robots. And of course, having fewer human beings allows us to grow living standards more without destroying the natural world and putting strains on scarce resources.
Those who expect aging to make China decline into irrelevance are almost certain to be disappointed. Bert Hofman has a good post explaining how China can compensate for the effects of aging — having people work longer, improve education, use more automation, allocating capital better, and so on. The example of Japan, whose population has been falling since 2008, is instructive.
And on the topic of Japan, he writes, in a different post:
The near-universal practice of seniority-based promotion, combined with low startup rates and population aging, has led to an ossified class of corporate executives and managers who would rather preside comfortably over declining little empires than embrace new technologies and business models and take new risks. That in turn has caused Japanese companies to fall behind foreign rivals as they miss technological revolution after revolution — microprocessors, smartphones, semiconductor foundries, battery-powered cars, etc.
His point seems to be that aging populations are only one variable which determines a country’s future. Countries are too multi-variate to ascribe their rise or decline to any one variable. Aging populations are a concern, to be sure, but other variables affect a country’s prospects as well. Be wary of ascribing too much causation to a single variable.