Thoughts for September 4, 2022
Good afternoon. Today’s topics are marine plastics, energy shortages, video games and artificial intelligence, and income inequality.
Marine Plastics
This week I’ve been tasked with developing a more thorough approach to the issue of plastic pollution in the oceans. My first step in a problem like this is to get a sense of the scale of the issue. That helps us determine how much of a priority it should be, as well as what kind of measures to deal with it would be appropriate. As with most topics, I find the evidence base to be disappointingly thin, but here goes.
A paper by Beaumont et al. is a major lynchpin. Citing other sources, they estimate (as of 2011) that the cumulative plastic pollution is 75-150 million tons. As of 2010, they estimate that the annual rate of pollution was 4.8-12.7 million tons.
Fishing gear comprises a significant portion of that, including both gear that is lost accidentally and intentionally discarded at sea. But how significant is also unclear. The World Wildlife Fund isn’t the best source of information on this topic as we’ll see later, but here they state that 0.5-1 million tons of fishing gear are discarded each year. I’ve seen those numbers in other places, and although it seems to be little more than a back-of-the-napkin calculation, it’s the best we have. There’s a commonly cited figures of 640,000 tons per year from fishing gear; this paper goes into the murky origins of that number and why it is at best false precision.
You may have also seen a factoid that fishing gear comprises half of ocean plastic pollution. As far as I can tell, that comes from this paper, which did some sampling in the Great Pacific Garbage Patch (an area in the deep Pacific where much plastic has accumulated) and found that 46% of their samples were fishing gear. But that’s just the GPGP, not the ocean as a whole, and as noted above, it seems that the figure for the whole ocean is much less. Many of the same authors put out another paper just a few days ago that the figure might actually be 75-86%.
One thing I like to do is find rough monetized estimates of the damages from environmental problems, so that we can see, at least on an order-of-magnitude basis, how severe they are and how they compare to other problems. For example, mainstream estimates of the social cost of carbon generally fall between $50 and $100/ton, and world emissions are about 40 billion tons CO2-equivalent, so that works out to around $2-4 trillion per year. How does plastic pollution compare?
Going back to Beaumont et al., their estimates of a “social cost of marine plastic” are based on ecosystem services. They start with Costanza et al.’s figure that the ecosystem service value provided by the open ocean and coastal habitats amount to $49.7 trillion per year. They estimate that 1-5% of that value has been lost due to marine plastic, for a cost of ~$0.5-2.5 trillion per year. Dividing by the 75-150 million tons of cumulative plastic pollution, that works out to $3300-$33,000/ton/year of marine plastic.
To translate that into lifetime costs, this report by Dalberg, commissioned by the WWF, assumes that a ton of plastic emissions will have constant costs indefinitely into the future and applies a 2% discount rate. With a CPI adjustment, they estimate annual damages from plastic pollution to be $3.1 trillion and the social cost of marine pollution to be ~$281,000.
There are numerous reasons to be doubtful about this figure. A non-exhaustive list:
It is not at all clear how reliable the ecosystem service figures of Costanza et al. are. That would be a good subject for another day.
Where does the 1-5% loss of Beaumont et al. come from? “In light of this evidence, it is considered reasonable to postulate a 1–5% reduction in marine ecosystem service delivery as a result of the stock of marine plastic in the oceans in 2011.” It does not appear to be any more than a guess.
Is a 2% discount rate appropriate? It seems rather low.
The indefinite lifetime of plastic assumes that none will be mineralized over time, or that the cost is unaffected by breakdown. It also assumes that there will never be any clean-up.
There is no distinction between the location of emissions, type of resin, or the shape/size of plastic pieces.
Since the WWF report builds off the Beaumont et al. estimate, this estimate is, as far as I can find, the only independent estimate of the social cost of marine plastic (if there is another that I am unaware of, I would be grateful to be informed of it).
A gut check is in order. The WWF report puts the annual cost of plastic pollution at $3.1 trillion, which would be similar to annual global warming costs at a social cost of carbon of $75/ton. This does not seem to me to be plausible.
The WWF has a long-standing anti-plastic campaign, so it would be tempting to conclude that their research is fudged to produce the highest plausible numbers they could. But in truth, I have little idea of what the numbers should be. So, for the remainder of this post, I will go with the WWF’s figure.
There are a couple more things we can do to try to translate this cost into understandable terms. First, let’s look at wildcatch fishing. As of 2015, Our World in Data reports that there were 93.74 million tons of wild-caught seafood (yes, I am aware that aquaculture is responsible for some fishing gear loss as well, but for this exercise, I am going to attribute to wild catch only). We’ll use the upper estimate of 1 million tons per year of fishing gear lost. That works out to about $3000/ton, or $1.36 per pound of wild-caught seafood. As of today, market prices at the Fisherman’s Market in New Bedford, MA generally run around $6-20/pound, though more for some premium products.
Now let’s look at plastic straws. Plastic straw bans have already become a poster child for symbolic and innumerate environmental actions, but I’ll pile on a bit more. A plastic straw weighs 0.4 grams, so if I were to take a straw directly to Cannon Beach and chuck it into the ocean, it would do about 11 cents of damage. According to Our World in Data, the United States produces 37.83 million tons of plastic waste per year, of which less than 1%, or 267,469 tons, is mismanaged (litter, etc.). If waste is mismanaged, there is a 0.35% chance it ends up in the ocean. That means that the social cost of a typical straw is around 0.000032 cents. If I were to live to age 100 (59 more years) and use three plastic straws per day for the remainder of my life, and dispose of them as is typical for plastic waste in the United States, my future lifetime damages will be around 21 cents.
To get a little more serious about solutions, a point that the OWID article makes clearly is that mismanagement of waste, rather than the sheer amount of plastic, is by far the more important factor is how much pollution a country generates. Another important, though less controllable, factor is how much coastal population a country has. The obvious solution is to develop better waste management systems, while banning single-use plastic or plastic in general will accomplish little at high cost.
As for fishing gear in particular, the EIA (Environmental Investigation Agency, not the Energy Information Administration) outlines some provisions of more stringent international agreements about fishing gear loss. This article argues that stronger enforcement of existing agreements, particularly the London Protocol and the International Convention for the Prevention of Pollution from Ships, would be more appropriate in the near term.
In summary, marine plastics are a serious issue, but one that is plagued by bad data and poorly-crafted solutions. Much better solutions are possible.
Energy Shortages
Back in the mid-to-late 2000s, there was the peak oil movement, which argued that the world was near limits to oil extract, and it would soon decline for irreversible physical reasons, leading to severe economic dislocation. The predictions associated with the movement were wrong, oil production continued to increase in the 2010s, and prices fell. The movement mostly disappeared, and insofar as it is remembered at all, it is remembered mostly as a cautionary tale about limits-to-growth ideology.
But there is still a die-hard core of the movement. One of them is Gail Tverberg, who wrote this piece arguing that current energy problems, which have their proximate cause in disruptions in Eastern Europe, are really symptoms of developing chronic energy shortages.
It is clear that there are no genuine physical shortages on the availability of energy. There is no fundamental technical reason why it would be impossible to greatly increase energy consumption, even without any breakthrough technologies. It would be well within the realm of possibility to build massive numbers of nuclear reactors and solar farms, and use them to power electrolysis that would create synthetics fuels for processes for which direct use of electricity in unsuitable. And there is no obvious reason to expect that breakthroughs, such as in fusion or space-based solar power, would be impossible.
But the picture becomes more murky when one considers economic, political, social, and cultural challenges. A major reason that limits-to-growth assertions fail is that they don’t account for economic feedbacks. If there is a credible threat of shortage of a commodity, futures prices rise, which create incentives to increase production or use the commodity more efficiently, which in eventually negates the price rise. I had assumed and argued that a similar mechanism should hold for political constraints. If, for example, permitting of new mines is too stringent and results in a mineral price rise, this should result in a public backlash which causes permitting to be less stringent. Now I am less confident that such a mechanism holds.
Video Games and Artificial Intelligence
I found this post from last month interesting, arguing the importance of video games in artificial intelligence research. A large portion of work in AI goes to video games (and other types of games); some recent examples include OpenAI’s work on playing Atari and other retro games straight from output pixels and DeepMind’s mastery of Starcraft II via AlphaStar. There was also the recent victory of AlphaGo in the game of Go, a game that had previously been regarded as very difficult for an AI to master, and going back a ways, Deep Blue’s 1997 victory over Gary Kasparov in chess.
While video games may seem frivolous, the post argues that gaming is an ideal platform for developing AI techniques, and it is quite likely AI-complete, meaning that the techniques required for mastery of gameplay are equivalent to those required to develop artificial general intelligence. Beyond gameplay, gaming provides an interesting platform for generative content, as for instance illustrated here (the author of the blog post is also one of the authors of the paper). I’ve tried my hand at generative gaming content as well; the results, I must admit, fell far short of expectations.
There is connection on the hardware side as well. Similar matrix multiplication operations that are needed for 3D graphics are also needed for back-propagation of deep learning models, and thus custom TPUs (tensor processing units) for deep learning evolved from GPUs (graphics processing units).
Togelius’ post must have been aimed at the older generation. For people my age and younger, I don’t think any convincing is required that video game playing and asset generation are serious endeavors. Gamification breaks down the barriers between video games and “serious” endeavors as well.
Income Inequality
Scott Lincicome of the Cato Institute wrote this piece last year on income inequality. He argues—with plenty of data—that much popular understanding about income inequality is based on faulty data.
First, much of our understanding of increasing inequality is based on pre-tax statistics. Transfer payments to lower income brackets have generally been getting more generous over time, so when post-tax figures are considered, most of the apparent increase in inequality disappears. It can be debated whether the pre-tax or post-tax figures are more relevant; Lincicome clearly gravitates toward the latter and argues that even if we think that pre-tax inequality is more relevant, that problem by definition cannot be solved with transfer payments or moving toward a more progressive tax system.
He also debunks the pay-productivity gap, something I’ve touched on in the past, and the idea that the income share of labor has been decreasing at the expense of income share of capital.
There is also a widespread perception (see e.g. Pew Research) that the United States has much higher inequality than other wealthy countries. That’s not a point that Lincicome addresses in his post, but given the issues around pre- and post-tax inequality and other statistical issues, I wonder how true this is as well.