Thoughts for January 22, 2023
Good evening. This week’s topics are the technological singularity, the economics of density, polycrisis, and gas stoves.
Technological Singularity
I have written a lot here about the “Great Stagnation”, the idea that scientific and technological progress are slowing down. There is much evidence to suggest that such a condition reigns today, and that insofar as progress is continuing in the world today, it is primarily in the form of convergence toward an increasingly static frontier. But here I want to look at an extreme opposite view.
Like most concepts, the history of the singularity, in some sort, is older than one might realize. According to this review, Marquis de Concorcet in Sketch for a Historical Picture of the Progress of the Human Spirit (published posthumously in 1795) discussed (though not by name) the possibility of a singularity in the development of knowledge. The work is one of the most important expositions of the Enlightenment conception of progress. Stanislav Ulam, in an essay on the life of John von Neumann, coined the term “singularity”.
In contemporary usage, the main principle of the singularity is that a sufficiently advanced artificial intelligence system will be capable, with minimal immediate human supervision, of designing an even more advanced system. This second system could design a third, even more advanced system, and so forth. Because human supervision is not needed, this process could proceed very rapidly, and in a short period of time, such an advanced AI system could be available so as to fundamentally transform society. So great would be the transformation that pre-singularity understanding of the dynamics of society would no longer be applicable; the term is meant to be analogous to the singularity of a black hole, where established physical principles break down. This conception of the singularity is, to my knowledge, attributable to I.J. Good in the 1960s.
Vernor Vinge deserves much credit for popularizing the idea, especially in this 1993 essay. In a short span, Vinge addresses intelligence augmentation (the idea that humans and computers work together in a post-singularity environment), technological unemployment, AI safety, and antiaging. Vinge does not address with specificity how the singularity might be achieved or lay out an AI research program, which is probably for the best. In all, the essay has aged well given its fast-moving subject matter, despite the abstract.
Within thirty years [by 2023], we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.
In 2005, Ray Kurzweil wrote The Singularity is Near, which documented trends in computation and artificial intelligence and forecast that the technological singularity would occur in 2045. We are almost half way from the book’s publication date to the forecast date. Kurzweil is the most prominent current popularizer of the singularity idea, and he was a major influence on my own early thought on the subject, but I have since moved on.
The main problem (among many) that I have with this concept is that the discussion of intelligence is very fuzzy. Moravec’s Paradox (named for Hans Moravec), for instance, observes that certain high level tasks are much easier for artificial intelligence systems than low-level sensorimotor skills. For instance, it turned out to be easier to design Deep Blue, which could play chess better than any human, than to design reliable facial recognition, while grandmaster-level chess is impossible for most people and facial recognition is trivially easy. Thus machine intelligence—if intelligence is the right word, which I think not—is not commensurate with human intelligence, and concepts such as “superhuman intelligence” are incoherent.
Now would not be the time to attempt to describe in detail the pathways to a singularity-inducing artificial general intelligence, but one particularly interesting approach is Eric Drexler’s 2019 Reframing Superintelligence: Comprehensive AI Services as General Intelligence. The piece moves away from the common agent-like conception of AGI, and it suggests a greater emphasis on infrastructural aspects of AI than is usually appreciated. But Drexler’s model might suggest a greater decomposability of AI than is possible in actuality.
If you intend to follow one and only one link in this section, I suggest Vernor Vinge’s essay. It is well-written, short, packs in many concepts, and highly influential.
Economics of Density
The concept of agglomeration economies is well established in urban economics. It is usually framed as a scaling relation, in which if a city size doubles, the per-capita GDP (all else being equal), increases by X%, where X is typically between 5 and 10. This paper is a good example, and it finds among U.S. cities one of the stronger scaling relationships (11% for a city size doubling). These results suggest that by liberalizing zoning or improving transportation, the United States and most countries could create a great deal of wealth. But how do we translate this into something useful for individual situations?
There was a study a few years about a waterfront redevelopment project in Eugene, Oregon (a project that is ongoing), which I will use as an example. These numbers will be very rough-and-ready, meant to explore concepts rather than carry any precision.
The external agglomeration benefit of an additional person in a city (read: commutershed) is equal to the GDP of the city times the agglomeration elasticity divided by the population. The Eugene metro area has a GDP of around $19 billion and a population of around 383,000, so if we use an agglomeration elasticity of 0.05, we get that the external agglomeration benefit of adding one person is around $2500/year. The project envisions 360-375 new households; if we take 360 households and assume two people her household (no point looking up precise figures because this is meant to be a quick sketch), that works out to around $1.7 million per year, or $35 million of net present value if a 5% discount rate is used. This compares to a project cost of just under $139 million. Since agglomeration is not discussed in the report, that could be decisive in determining whether the project is a good investment.
There are lots of questions. First, despite the fact that the literature usually refers to agglomeration as an externality, here I am trying to treat it like a line item in a cost-benefit analysis, in the way one might do with CO2 emissions, and it is not clear to me that this is appropriate. It is also not clear whether the residents of these proposed developments would really be new to the region, or if this is displacing development elsewhere in the city, in which case there might not be any agglomeration benefit. There are some undesirable externalities related to agglomeration, such as congestion, pollution, and crime, which I am not attempting to account for here.
However, the idea of using agglomeration economies in cost-benefit calculations is not unheard of. There are many examples, such as this one, of using it with transportation projects. Here’s an example from New Zealand that uses agglomeration as part of the cost-benefit analysis of a residential upzoning proposal. In this example, the agglomeration benefit is about five times as great as the consumer welfare benefit of lower housing costs and is decisive in making the case of net benefit (there are, however, many aspects of the report that confuse me, and I have low confidence in the numbers).
Polycrisis
“Polycrisis” is a buzzword that, over the last year, has suddenly appeared much more often. This is the idea that
There are several major crises occurring in the world today,
They are interrelated and interact, and thus cannot be understood in isolation.
The immediate prompt for this section is the recent Global Risks Report 2023, coming out of this month’s meeting of the World Economic Forum in Davos. The crises are cost of living, climate change, ecosystem decline, social polarization, “economic warfare” (also a broad heading), pandemic risk, and debt, among others.
I find these reports, which I have been reading for the last few years, to be a useful window into the current state of mainstream elite thinking. But they are dull, forgettable, and not very insightful. It has that indescribable NGO-speak style of writing which manages to use a lot of words to convey not much information. But polycrisis is the hook that gives some coherence to these ideas.
There is a tendency to view the challenges of the present as unique and unprecedented, which helps justify the moniker of “crisis”, but how helpful that framing is can be debated. In a trivial sense, any set of circumstances is both unique if viewed with sufficient specificity, and precedented if viewed with sufficient generality. The impulse to group these together into a “polycrisis” can also be justified in some sense—any two subjects can be related if one tries hard enough to do so—but whether this is helpful in understanding or solving the problems is another matter. Here I don’t see it as helpful. I wrote the following (“some” was meant to be “something”) on Twitter a few days ago.
The way one frames a problem can subtly guide a reader to a desired solution, as done in this CNBC article, “The world’s in a ‘polycrisis’ — and these countries want to quash it by looking beyond GDP”. The title alone triggers a gag reflex. Sometimes an article is badly titled and worth reading, but here the article is even worse. I cannot summarize it. It is little more than a collection of degrowth-adjacent talking points that looks like it could have been written by ChatGPT.
Perhaps I shouldn’t be so hard on the WEF report and the polycrisis concept in general. There are a cluster of problems that I write about frequently, including poor economic performance, especially in wealthier countries; declining birth rates; democratic backsliding; and regulatory sclerosis. I often write about these problems as a whole under headings such as “stagnation”. But seeing others do the same thing makes me think that this is a bad habit that should be rethought.
Gas Stoves
I was thinking about writing more extensively about the kerfuffle around gas stove regulation, but this piece entitled Bad Stove Science Is a Roadblock on the Path to Electrification by Alex Trembath discusses the issue better than I would have, so I’ll link this instead.
A few things should be kept in mind:
The proposal would ban gas stoves in new construction. It would not affect stoves in already-built structures.
The rule would come from the Consumer Product Safety Commission and be based on health impacts.
No rule is finalized at this point.
Numbers are key. I have no problem believing that gas stoves generate indoor air pollution that harms human health. But how extensive is the harm? Put that in monetary terms so we can compare with the cost of the regulation and decide if it is worth it. Maybe it is, but I doubt it.