Welcome to the Machine

Sometime in the 1970s, well before the advent of the iPhone, a Chicago newspaperman Sydney J. Harris filed a sentence that has outlasted nearly every column in which it first appeared: “The real danger is not that computers will begin to think like men, but that men will begin to think like computers.”

Harris’s concern was not that mankind would become ultra-rational and peaceable, like Star Trek’s Mr. Spock, a show that had begun to recently spike in popularity. His worry was deeper. Men, he saw, were beginning to ape machines in profound ways.

“Machine-thinking” is no longer optional. It has become the condition of admission to nearly every institution that distributes modern life: employment, credit, healthcare, education, the public square. Americans are slipping into the cognitive style of the machine because our social systems will not process them otherwise. This has profound implications.

The Architecture of Compliance

Consider what it now takes to apply for a job. According to Jobscan’s most recent Fortune 500 audit, 97.8% of those companies run applicants through automated tracking software before any human sees a resume. Smart candidates know this. They understand that before a hiring manager is even bothered with a resume, a parser must do its work scanning for keyword density, schema, machine-readable verbs, you name it. Workday and Greenhouse have produced an entire genre of human prose composed for the eye of the algorithm. This is human thought accomodating for the machine.

The field of medicine is another example. Primary care physicians, according to the Annals of Family Medicine, now spend nearly two hours on electronic health records for every hour they spend with patients. But the note is not written for the next doctor or specialist; it is written for the billing code. Today we have software that translates care into reimbursement, hence the billing codes. A streamlined reimbursement process requires the prose of the machine not the note of the physician. Over time, even our finest medical minds begin to follow rules.

Unsurprisingly, the pattern extends downward into the schools. Here it becomes interesting. K–12 instruction is organized around the standardized test which rewards a particular cognitive style: pattern recognition over judgment, retrieval over reasoning, the multiple choice over the essay. Forget oral exams. The successful college applicant knows, like his peers entering the job world, that algorithmic screening tools will read his personal statement. Better have some words and phrases that get picked up.

But the best example of all is in our world, the virtual public square. To be read or heard on a major platform is to be optimized by it. The post must be short, scannable, retrievable. The thumbnail must be tested. The headline must satisfy the model’s prediction of click-through. The same rules apply when it comes to social media. Woe to writers who reject these conditions. They simply won’t have an audience.

The Collapse of Judgment

That is the architecture of compliance. No single piece of it is malicious, really. And that makes the danger hard to spot. What the systems impose on the cognition of the worker, they also impose on the judgment of the audience. Once distribution becomes the measure of value, the very category of “good” is surrendered. A piece of work is no longer judged. It is scored. The question “is it any good?” is replaced by the question “did it perform?”, and a culture that has accepted that substitution has not just lost the ability to think clearly about its own creations. It has lost the standard against which any creation could be evaluated at all.

The consequence falls hardest on the people who actually make things and do the work. Take the contractor whose business depends on whether a lead-ranking algorithm decides to surface him this week, and who pays $15 to $100 for every shared lead the system routes to him and to three or four of his competitors at the same time. Or the trucker scored continuously by the electronic logging device wired to his engine, his hours and his braking and his cornering aggregated into a fleet telematics scorecard that decides his routing, his pay, and his place on the FMCSA safety record his next employer will pull. Or how about the teacher whose career rides on her students’ standardized-test growth, and the police officer whose report has to satisfy CompStat categories before it can satisfy the case in front of him? None of these people are competing against other contractors, truckers, teachers, or officers. They are competing against a parser that cannot see the work they do. The computer knows only what its scoring rubric has rewarded before. Whatever does not match that rubric is not bad, it is simply invisible.

The same dynamic has reshaped the culture downstream. Spotify pays royalties only for streams that pass a 30-second threshold, and song structure has reorganized around it. The music theorist Hubert Léveillé Gauvin, in a study of Billboard top-ten hits between 1986 and 2015, documented a 78% collapse in the length of song intros, from more than 20 seconds to about five. The chorus arrives before the listener has time to skip. Hits engineered for TikTok telescope further still: the hook lands in the opening bars, the bridge is suppressed, the song is built around a 15-second loop the platform’s algorithm can deploy at scale. None of this reflects an evolution of taste. It reflects a metric.

Publishing has gone the same way. Amazon’s recommendation engine rewards book covers that perform within the search-result thumbnail, and cover designers who once trained in the craft of evoking a book’s interior in a single image now design instead for click-through on a 200-pixel tile. AI-generated covers, increasingly common across self-publishing, are now competing on the same metric.

The Page the Parser Reads

The same logic governs the written word, but here the substitution has accelerated past anything Harris’s successors could have anticipated. ChatGPT serves more than 800 million weekly users; Google’s AI Overviews now appear at the top of billions of searches before the user ever clicks a blue link. The new game is no longer ranking on the search-results page. It is being cited inside the answer the user reads. The industry has not settled on a name for the discipline. Generative Engine Optimization, Answer Engine Optimization, LLM Optimization: the labels compete, but the inversion is the same. The writer no longer writes for the human reader. He no longer even writes for Google’s PageRank. He writes for the parser that decides whether the LLM will extract him and cite him in the synthesized response that increasingly replaces the page itself.

The page the writer publishes and the page the parser reads are no longer the same page. The serious essay published without schema, without an llms.txt, without the machine-readable backbone the model is trained to recognize may still be excellent. But excellence is no longer enough to guarantee legibility. The essay enters a distribution system increasingly unable to distinguish the overlooked from the unintelligible. Again, judgment has nothing to do with it. The machine, to repeat, is not a judge. Think of it more as a feedback loop. It rewards what it recognizes, and it recognizes what it has already rewarded. Inside that loop, originality reads as error, strangeness as risk, depth as inefficiency. The system has not so much as abolished judgment but replaced judgment with itself.

The Forty-Year Warning

Harris worried about a free citizen choosing to think like a computer. Today, the free citizen has been replaced by a subject who is processed by computers and learns, through the steady pressure of consequences, to anticipate how the algorithm will reward or punish their work. Of course, reward and punish, again, is incorrect nomenclature. The machine does not think in human terms, only in terms of scoring.

This brave new world is hard to combat. After all, these technological advances are simply making things easier for us, right? Applicant tracking saves recruiters time. Electronic records improve coordination. Lead-ranking algorithms route customers. Telematics improves safety. Recommendation engines surface what users want. Schema markup helps machines understand structure. Great!

But considered together, they constitute a quiet substitution of one kind of human being for another, and one kind of culture for another. The new model is fluent in the dialects of the machine, comfortable with its categories, and unable to recognize what has been traded away: not just in the cognition of the worker, but in the standards of the audience and the conditions under which anything worth making can still be seen. The old adage goes “the internet is forever.” Perhaps. Or perhaps it is “forever” to the extent it considers what needs retaining worthwhile.

Harris assumed his readers retained the freedom to refuse. He was writing for a country that still had the option of being unreasonable, of preferring judgment to procedure, of insisting that a person was not a record and a piece of work was not a score. To be clear, this country still exists. But to be clearer still, this is simply no longer the country the systems are built for.

Recovering this old mode of thinking will require the slow, expensive work of building institutions, professions, and platforms that treat human cognition as the irreducible thing it is: incomputable, particular, and worth the trouble. A magazine still edited by human beings reading every page. A hiring process that reads work samples instead of parsing resumes. A school that grades essays rather than scoring scantrons. None of these are sentimental restorations. They are the conditions under which a serious culture can still be made and seen.