CHAPTER 19

"Strange Loops and Abstractions: The Devil in the Stack"

p. 405

Hofstadter claims to have no interest in computers as such and has expressed skepticism of current "AI" models: In "The Mind Reader," New York Times Magazine, April 1, 2007. Like so many others, he ultimately changed his mind: "'Human Beings Are Soon Going to Be Eclipsed,'" New York Times, July 13, 2023.

406

To illustrate the shifty concept of recursion he not only recruits Bach's canons: Most particularly in the composer's extraordinary fugue Ricercar, part of a collection of pieces based on a single theme, originally improvised, called The Musical Offering. Hofstadter's explication of the structure of Ricercar early in Part One of the book and is among the most fascinating musical discussions I've ever come across. Afterwards I couldn't stop listening to the piece on the Hilliard Ensemble's Ricercar album.

Marshall McLuhan's famous sixties adage, "the medium is the message": From Understanding Media: The Extensions of Man, 1964.

407

Put another way, mediums propagate their essences, their forms, in all who use or encounter them: I write about this in more detail in my book Totally Wired: The Rise and Fall of Josh Harris and the Great Dotcom Swindle. On the subject of McLuhan I also recommend Douglas Coupland's entertaining biography, Marshall McLuhan: You Know Nothing of My Work!—named for the famous scene in Woody Allen's Annie Hall, where Allen produces McLuhan out of nowhere in a movie queue to set a fellow queuer straight about the Canadian thinker's ideas .

We shape our tools, and thereafter our tools shape us: This quote opens the door to multiple rabbit holes for McLuhan geeks. Culkin was explaining McLuhan's thought when he said it, but McLuhan doesn't ever appear to have claimed attribution. To complicate matters further, none other than Winston Churchill told his Parliament in 1943, on the subject of rebuilding the bombed House of Commons, that "We shape our buildings and afterwards our buildings shape us." For the very nerdy (which in this particular instance includes me, obviously), there's a satisfyingly labyrinthine discussion of attribution on quoteinvestigator.com.

408

The money's quintessence, corruption, replicated in everything it touched and may take a generation to cleanse: for astonishing accounts of this process, read Oliver Bullough's Moneyland (2019) and/or Butler to the World: How Britain Helps the World's Worst People Launder Money, Commit Crimes, and Get Away with Anything, or listen to Tortoise Media's "Londongrad" podcast.

409

a computer founded on similar "ternary" logic was built in the Soviet Union as long ago as 1958: this computer was called "Setun" and was considered to have some significant advantages over computers built on binary logic. See Wikipedia entries for "Setun," "Ternary computer" and "Decimal computer."

410

Every problem in computer science can be solved by another layer of abstraction: This is in fact a deliberate misquotation of something said by the Cambridge University computer scientist David Wheeler, specifically: "All problems in computer science can be solved by another level of indirection." Loosely defined, "indirection" refers to the process of subsuming a value under a convenient name or symbol, a good example being a value stored in a variable. The quote is sometimes given as "All problems in computer science can be solved by another level of indirection, except for the problem of too many layers of indirection." Arcana lovers can knock rhemselves out on "Indirection: The Unsung Hero Of Software Engineering," Forbes, August 20, 2020.

when I try to work out exactly how my Python gets to the metal—the bottom of the stack—it takes days of research to trace the steps: This passage owes a huge debt to Nicholas. It tells you something about the degree of complexity and depth of abstraction involved in the process being described, that after all my painstaking investigation I'd got very close but still not quite there. Nicholas consulted his colleague Dr. Antonio Cuni, respected as one of the world's leading authorities on Python compiler design, and helped immeasurably in producing the finished account you read in these pages. Giant thanks to both these incredibly smart people. So this quest involved us both in days of work . . . only afterwards did I think: we could have just bought Guido a beer and asked him!

a series of instructions called bytecode: As with the mostly hidden code objects, you would never normally see bytecode, but a natty tool called a disassembler allows you to, rendering it as something very like the ultra-low level Assembly Code used by programmers on the Apollo space program, before they had high-level languages to translate for them: long sequences of numbered codes and abbreviated three-letter commands.

a Python virtual machine—a simulated computer, written in C, that compiles to the C interpreter: Another one for lovers of the long grass. During my extended period of trying to piece together this process—it appears written down like this almost nowhere—I grasp for the first time that the Python programming language I experience exists independently of the particular implementation that interprets my own code, and that there is more than one such implementation—i.e. more than one available interpreter. The standard implementation such as I and most people use is called CPython because it is written in C and compiles to the C interpreter, while Jython compiles to Java bytecode in order to be processed by the Java interpreter. Yet the source code I write for each will look (and be) the same.

412

scientists began to flag a dangerous link between carbon emissions and global heating from the late 1960s on: the English engineer Guy Stewart Callendar began to publish evidence of a link in 1938, but thought the resultant warming would be beneficial (as an Englishman might.) From the late 1950s scientists began to model and monitor the effects of warming, leading to the first serious public warnings from the late 1960s. Politicians and the public engaged en masse from the late 1980s.

413

likelihood of committing a crime and therefore suitability to being searched or arrested: "Predictive policing algorithms are racist. They need to be dismantled." MIT Technology Review, July 17, 2022. Cathy O'Neil lays out the dangers of algorithmic decision-making and gives many, many examples of what she means in Weapons of Math Destruction.

research claiming mathematical proof that code fixes Facebook proposed for its algorithms would inevitably worsen their effects: See "Emergent dynamics of extremes in a population driven by common information sources and new social media algorithms," N. F. Johnson et al, Scientific Reports, 15 August, 2019. The abstract reads:

We quantify how and when extreme subpopulations emerge in a model society despite everyone having the same information and available resources -- and show that counterintuitively these extremes will likely be enhanced over time by new social media algorithms designed to reduce division. We verify our analysis mathematically, and show it reproduces (a) the time-dependent behavior observed in controlled experiments on humans, (b) the findings of a recent study of online behavior by Facebook concerning the impact of 'soft' and 'hard' news, (c) the observed temporal emergence of extremes in U.S. House of Representatives voting, and (d) the real-time emergence of a division in national opinion during the ongoing peace process in Colombia. We uncover a novel societal tipping point which is a 'ghost' of a nearby saddle-node bifurcation from dynamical systems theory, and which provides a novel policy opportunity for preventing extremes from emerging.

We need to understand that over time the logic of recommendation engines is to narrow individual and collective scope: I don't want to write these off as part of a balanced portfolio, however. I use the Tidal streaming music service and frequently appreciate tips from the algos—but only in conjunction with recommendations from friends, newspapers and magazines, radio, playlists in cafes and so on. On their own the algos would tend to narrow my listening. They can only work on the basis of what I've responded to in the past.

nutritional labelling on food (something the food industry fought tooth and nail, lest we forget): and still do. See the struggles around identifying genetically modified foods, sugar etc.

414

We've seen Machine Learning programs do remarkable things within closed systems (and also disturbing things, as per High Frequency Trading on the stock market): See my own "Fast money: The battle against the high frequency traders," The Guardian magazine, 7 June 2014, or (for more) Michael Lewis's book Flash Boys.

Frances Haugen's eloquent testimony before Congress made plain that Facebook had been grooming kids for monetization and knew it was harming teenage girls: "Here are 4 key points from the Facebook whistleblower's testimony on Capitol Hill," NPR, October 5, 2021.

415

Even these could be a handful to manage: See Julian Dibbell, My Tiny Life, ibid.

Most of the work I see coders doing on next-generation social networks resembles these first iterations: One I especially enjoyed exploring while making the book was Scuttlebutt, whose Rubyite proprietor, Rabble, is one of the smartest and most colorful people I spent time with. He disapproves of Python, thinking it way too restrictive. Nicholas still harbors a yen to flesh out TextSmith, too.

a lecturer named Nir Eyal taught a course on "behavioral engineering" at Stanford: "The professor who wrote the book on making addictive technology is having second thoughts," CNBC, January 30, 2013. Oh, really?