Assorted Topics for July 6, 2024
After the series on food loss and waste over the last three weeks, I would like to address several shorter topics today.
How do we Measure Food Waste?
OK, evidently we’re not quite done with food waste. Last week, I stated,
In 2015, the United Nations adopted the Sustainable Development Goals, a sprawling set of goals and metrics toward, well, sustainability and development. One of them, Target 12.3, aims to reduce food loss and waste by half from 2015 to 2030. We are not on track to achieve the goal.
It should be “… reduce per capita food loss and waste by half…”. We are still not on track.
I considered whether food waste is generally higher in high-income countries. This is something that feels like it should be true, and it has been common wisdom for a long time. However, research from the United Nations Environment Programme and the Food and Agriculture Organization of the United Nations finds that food waste is generally the same across national income levels.
To support the view that waste is higher in high-income countries, there is this paper that I mentioned last week and an older paper with a similar methodology. These papers use an energy balance strategy, where they start with food available to consumers, as measured from the FAO’s Food Balances, and subtract an estimate of human caloric intake, which is based on weight gain/loss and physical activity, such as done here. Here is another study showing generally higher waste in high-income countries, though I admit that I don’t follow their methodology.
The first problem is that, when one subtracts two quantities with noise, the result is extremely noisy. Look at the plot on the left here, from van den Bos Verma et al., which shows national per capita income and estimated food waste for various countries.
Negative values are obviously wrong. The Barrera and Hertel study sweeps the issue under the rug by rounding the negative values up to zero. Subtracting two values that mostly cancel each other out severely compounds noise, but it also magnifies possible bias. And so the big question is, is there anything about the methodology that would cause a spurious income/waste relationship to appear?
Here are some of the measurement and definitional issues that come into play.
How accurate are the Food Balance numbers, and are there income biases?
How accurate is the model of caloric consumption?
Food grown for one’s own personal consumption is often not counted in the Food Balance numbers. This issue is more pronounced in low-income countries.
The studies try to account for conversion in food processing. For example, it takes one kilogram of wheat to make about 0.8 kg of pasta. How accurately is this done?
Does food waste include inedible (e.g. fish bones, banana peels) parts?
Does food waste include food that is composted?
Does food waste include scraps that are fed to animals?
On the last three questions, it appears to me that the two clusters of studies differ in their answers, but I am unsure. I am inclined to put more stock on the UNEP/FAO analysis, and conclude that there is not an income/waste relationship, but the data in this analysis is also rather spotty. If any readers have insights into international measurements of food waste, I could be grateful to hear them.
Youth Unemployment
Several times, I have mentioned the Reading History Sideways fallacy, which is also the title of a book by Arland Thornton. The fallacy is to assume a universal development pattern that can be represented by income-per-capita, and so low-income countries today show us what high-income countries looked like in the past, and high-income countries today show us what low-income countries will look like in the future. The fallacy can lead us astray, for example, when we try to project fertility trends or environmental impacts via Kuznets curves. I have been looking for a good illustration of principle. Youth unemployment is not perfect, but it helps.
Youth unemployment shows an obvious increasing trend from 1991 to 2020, rising from 10% to 17%, and a fall since 2020. It remains to be seen how durable this fall will be.
If we read history sideways, then it stands to reason that since countries get wealthier over time, then wealthy countries must have have higher youth unemployment than poorer countries. However, from the World Bank again, there is no obvious income/unemployment relationship.
As one might expect, youth unemployment is another area that is fraught with measurement and definitional issues. India, for instance, might have one of the highest youth unemployment rates in the world of 45.4% (ages 20-24 for these figures), but the Indian government claims that youth unemployment (ages 15-29 this time) for 2022 was 5%.
Youth unemployment is another topic on my length to-do list.
Automated Coding
Automated coding, which we can defined as the use of an artificial intelligence system to generate code based on natural language specifications, has been rapid advancing in the last few years.
The foregoing shows the progression of scores from various models on the benchmark set HumanEval. Introduced in 2021, HumanEval is a set of 164 programming problems with natural language specifications and unit tests. The same paper introduced Codex, a model that solves (that is, generates code from the specifications that passes the unit tests) 28.8% of the problems. As of this writing, the state of the art is a model called Language Model Debugger that solves 98.2% of the problems, a rate so high that HumanEval is losing its utility for evaluation.
A few weeks ago, I mentioned SWE-agent, which moves to a more advanced benchmark set SWE-bench. Since that post, SWE-agent’s 12.47% pass rate has already been beaten by Factory’s Code Droid, with a 19.27% pass rate.
I can attest from personal experience that tools such as GitHub’s Copilot are widely used, and Copilot’s performance has noticeably improved since it was introduced in 2021. Given that the technology is still rapidly evolving, it is grossly premature to speculate on the impact that automated coding might have on the profession of software engineering, let alone on the economy as a whole. But if even long-time deep learning skeptics such as myself are changing their minds, then it seems that the impact will be significant.
Should Joe Biden Step Aside?
Since last week’s presidential debate, when Joe Biden’s apparent senior moments were on clear display, speculation is at a fevered pitch as to whether Biden should step aside for the 2024 Democratic nomination in favor of another candidate. I won’t render an opinion on that subject, but I do want to offer some historical perspective.
First off, I should admit that I have not actually watched the debate. I find it inexplicable that a debate was held this early, and I never find debates to be informative or interesting to watch. I am further put off by most of the media’s discussion of “who won” rather than substantive issues. Debates are memorable mostly for the gaffes and zingers. If a person knows anything at all of past debates, they most likely know about Richard Nixon’s sweaty brow and 5 o’clock shadow (themselves the subject of post hoc fallacies), Gerald Ford’s “no Soviet domination of Eastern Europe”, Ronald Reagan’s “there you go again”, and Lloyd Bentson’s “you’re no Jack Kennedy”.
It has happened on nine occasions—four times due to assassination, four times due to deaths from other causes, and once due to resignation—that an American vice president has assumed the presidency. The quality of succeeding vice presidents has varied, but the transition itself was generally smooth in all but the first instance. After William Henry Harrison died in 1841, a month into his presidency, it was unclear if vice president John Tyler was the new president or merely an acting president who would serve until new elections could be held. It didn’t help that Tyler had severe differences with the governing Whig Party. But now the precedent of vice presidential succession is not only firmly established but explicitly codified in the 25th Amendment to the Constitution. The death of a president in office would be a traumatic event for the country, but it would not threaten stability.
Of greater concern is a situation in which a president is alive but unable to do the job effectively. Section 4 of the 25th Amendment contains procedures for the vice president and cabinet to declare presidential inability, but it has never been used, though it has been considered several times, such as after the March 30, 1981 assassination attempt on Ronald Reagan. It is not clear how credible this provision is.
The most serious episode of chronic presidential incapacitation followed Woodrow Wilson’s stroke on October 2, 1919, an event from which he never truly recovered. For the last year and a half of Wilson’s administration, the United States was without an effective chief executive, and the severity of the condition was kept secret from Congress and the American public by his physician Admiral Cary Grayson.
Many of the convulsions of the later Wilson years are familiar to contemporary American audiences, and they include the aftermath of a global pandemic, social upheavals, economic instability following World War I, and a terrorist attack in Lower Manhattan. But the most consequential issue was whether the United States would join the League of Nations, which was Wilson’s brainchild in the Treaty of Versailles. According to this paper, Wilson’s incapacitation is a decisive reason in why the United States did not do so. Whether the United States should have joined the League is beyond the scope of this writing, nor is it entirely clear that a healthy Wilson would have convinced the Senate to ratify the Treaty of Versailles over the objections of Henry Cabot Lodge and rising isolationist sentiment.
Much more speculative is Ronald Reagan’s health during his second term. As is the case today, Reagan was the oldest U.S. president up to that time, and chattering classes in the 1980s speculated about Reagan’s age and mental heath. In a 1984 debate with Walter Mondale, Reagan deflected the age question with humor.
I want you to know that also I will not make age an issue of this campaign. I am not going to exploit, for political purposes, my opponent's youth and inexperience.
Reagan was diagnosed with Alzheimer’s Disease in 1994, five years after leaving office, and most of us know from experience that evidence of cognitive decline is visible well before a formal diagnosis. This paper argues that Reagan’s speech patterns show evidence of decline over the course of his presidency, as does this one. Here, John Heubusch and Craig Shirley, director of the Reagan Foundation and author of several books about the 40th president respectively, argue that Reagan did not suffer from Alzheimer’s while in office. The president’s mental health may have even played a role in the Iran-Contra affair. But all of this discussion is inconclusive.
Age is not a decisive factor, nor is the public able to diagnosis problems through television, especially when partisanship is involved. I recall that I had my own brain aneurysm at age 38, before which I had no evidence of serious health problems, and the event prevented me from working seriously for months. It is a major vulnerability in the American system that so much rides on the single person of the president, and fixing this vulnerability is more important than finding a spring chicken to put at the top of the ticket in a given year.
Quick Hits
I’ve been working through the Foundations of Eastern Civilization by Craig Benjamin on The Great Courses. I will probably have more to say about that later. Benjamin is one of the major players in Big History, and the Big History approach is clear in the course, starting the analysis with geography and early hominid migration, and the particular emphasis on cultural development.
One of my niche genres of entertainment is tool-assisted speedruns, which is the use of emulators to beat video games in superhuman time via frame precision controls, luck manipulation, bug exploitation, and other tricks. Here is a video of Super Mario Bros. The video opens with a cart swap from Super Mario Bros. 3, starts the game at World N-1, which is normally inaccessible, and uses an arbitrary code execution trick. Then it gets really weird. This video from the creator explains.