Excerpts: “Agreement changes in context: Are you addressing a grocer or a botanist? Plants do not grow with labels attached. Words attach to things indirectly, through use–at a time, in a place” (p. 55).

“Lovelace explained that ‘in studying the action of the Analytical Engine, we find that the peculiar and independent nature of the considerations which in all mathematical analysis belong to operations, as distinguished from the objects operated upon and form the results of the operations performed upon those objects, is very strikingly defined and separated'” (p. 56).

“Once we understand intellectual labor as labor, we should not be surprised to find tools, templates, and machines near it. But because mind and language are special to us, we like to pretend they are exempt from labor history” (p. 62).

“Wheel, table, pattern, template, scheme–the time has come for us to trace the last of our arcane figures–the Orion’s Belt–a string of words, arranged in a series of statistically probable continuations based on observed probabilities, called a Markov chain. An idea before its time, it gathered dust in the archives until computers became powerful enough to appreciate it” (p. 101).

“For instance, knowing nothing about oatmeal in the real world, our machine child could learn that the word often occurs next to words like bowl, eat, table, breakfast, and spoon. And it rarely occurs next to many other words, like bulldozer or munitions. From these facts, it could reasonably surmise that oatmeal has something to do with eating breakfast at tables with spoons, and not, say, war or construction. Such machine “learning” would depend on observing prior occurrence. The more a computer would read about oatmeal, the more it learned (metaphorically) about its common linguistic contexts. And it would be decades until machines had enough memory or processing power to retain enough information to start making sense” (p. 103).

“Sergey Brin, Lawrence Page, Hector Garcia-Molina, and Robin Li (the founder of Baidu) further built systems that weighted mutual citation. Documents that referenced other documents counted for more than those that were filed in isolation. Search engine technology was thus built out of parts found in library science, philology, and linguistics, using techniques like text alignment, indexing, cataloging, term frequency comparison, and citation analysis” (p. 113).

“Similarly, next time you hear the words ‘artificial intelligence,’ imagine not just a tangle of wires, but also the whole mess of complicity: people, practices, bureaucracies, and institutions involved at a distance. An AI gains it ‘smarts’ the way a team of trivia participants outplay a single opponent. There’s more of them. The remote action of smart devices obscures an essentially collaborative enterprise” (p. 132).

Tennen, Dennis Yi. (2024). Literary Theory for Robots: How Computers Learned to Write. New York: W.W. Norton & Company.