For much of this year, Madison Avenue has had a tough task: prevent a trillion-dollar investment bubble from bursting by selling us new ways of living through artificial intelligence.
A recent 30-second spot from Apple — forever vying with other tech behemoths to colonize and monopolize our future — centers on a fundamental human practice: faking it till we make it. In the ad, a TV actress shows up for a Hollywood meeting unprepared, having not read the pitch. In mere seconds, “Apple Intelligence” boils a long email down to a few lines, allowing her to feign the necessary knowledge.
The ad isn’t just selling the latest iPhone. It’s selling what Silicon Valley sees as an apparent ideal: to have read without ever having to read. A suite of products offering summary shortcuts — “meta reading,” you might call it, though not in the Zuckerbergian sense — symbolizes this AI cult of efficiency.
Why ever engage with a text directly when a robot can just cut to the chase on your behalf?
Google’s NotebookLM heralds much the same efficiency: You can upload dozens of files that its AI system will read for you and then query that content for synopses, answers, and linked passages. It’s a “democratized CliffNotes” that can be applied to any text ever created.
Because “hallucination” — incorrect or misleading information generated by AI — is an ongoing glitch bedeviling AI accuracy and, therefore, trustworthiness and adoption, I tested NotebookLM out on a source I know well: 100,000 words from my last book, which is about America’s obsession with authenticity in media, culture, and politics.
I posed queries like: “How does social media create the desire for authenticity while simultaneously fostering inauthenticity?” “How does the influencer industry idealize ordinariness?” And “Explain the strategies that campaign consultants use to create the impression of genuine politicians.”
The system’s output was flawless. It captured my book’s broad themes and synthesized them in outline form.
And if even NotebookLM’s bullet-pointed outlines of a text remain too heavy a lift, the system can also whip you up an eight-minute podcast distillation of the content featuring chipper chatbot voices with NPR-style cadences bantering about what you didn’t have to read.
As far as I could tell from my first foray into NotebookLM, there was no reason to read my book. Or any book. Ever again.
Too long; didn’t read
Automated summary shortcuts are proliferating. Otter and Zoom apply them to meeting transcripts; Facebook to user comments sections; and Amazon to buyer review highlights. Newspaper chain Gannett even piloted AI-generated “key points” atop articles, precluding the need to read the work of their reporters in full.
Such tools portend an information landscape where everything is reduced to TL;DR — shorthand for “too long; didn’t read.” Perhaps this is an inevitable development in a world awash in data that’s exponentially expanding. Part of the challenge of contemporary life is not just figuring out what to pay attention to but cultivating habits of mind to screen out what not to pay attention to.
And not all source texts are created equal. We lose considerably less, intellectually speaking, by shoveling the 60,000 customer opinions about Amazon’s smart plug into the insatiable maw of AI than, say, feeding it the 120,000 lines of prose composing Shakespeare’s canon. Yet both examples abide the same logic: To save time and “know” more, one must read less.
That cult of efficiency is not a new ideal. When the Industrial Revolution unfolded centuries ago, it redefined not just labor but virtue as well. Machinery foretold what could be done and therefore should be done — which is the same sales pitch that AI ads spin today.
Sociologist Max Weber diagnosed this as the principle of rationalization, an ethos wherein the most efficient, instrumental means to an end must govern all human behavior and where nothing is left to chance. This powered capitalism — if factory assembly lines could churn out widgets in half the time, owners would reap twice the profit.
The invisible algorithmic bureaucracies shaping today’s information landscape serve an analogous ideal. If you can power through a text in half the time (or less), you can consume twice as much (or more). Yet has that text, in its emaciated form, still left its mark on you? What’s lost with these gains?
Reading — not unlike its intellectual cousins, learning and writing — is not always an efficient process. Unlike an assembly line, it can’t be Taylorized — a theory of management named for Frederick Taylor, one of its leading proponents, that analyzes and synthesizes workflows for maximum productivity. With reading, there are lots of dead ends and cul-de-sacs of thought. Yet we only figure that out in retrospect, a process I consider one of the many messy, sometimes inefficient joys of learning.
AI wants to strip out all that inefficiency. It presumes to know what we’re looking for, even when we might not yet know. It empowers the power-browse. It seeks to eliminate the wastefulness of digressive detail. Google’s product manager for NotebookLM might assure us, “There is no replacement for reading the actual thing,” even as the system seeks precisely that: replacing a slower approach to answering a question with a mechanized one that is faster and more predictable.
Destroy reading and you might also destroy writing. The news business stands to lose upward of 25 percent of web traffic and $2 billion in ad revenue if readers settle for Google’s AI summaries that dominate search results rather than clicking through publisher links. After two decades of slow-motion implosion — with falling revenues, dwindling subscribers, and gutted newsrooms — the newspaper industry does not seem poised to survive one more digital innovation.
Who will populate the AI summary sources when content creators can’t make a living?
Pancake people, resist!
Fifteen years ago, tech writer Nicholas Carr presciently observed that the internet was turning us into “pancake people,” our brains flattened by the habit of superficially zipping along the surface of information like digital jet-skiers zooming across hyperlinks. Whither the scuba divers, diving into deep focus and becoming immersed in a printed text?
To be sure, a world lacking any efficiency would be a maddening place. Efficiency enabled human evolution across millennia and underpinned countless facets of growth. But it’s not the most important, much less the only, value that matters in a culture. And higher education, where knowledge is valued for its own sake, ought to be a special space of resistance against the unsparing dominance of the tyranny of efficiency.
Sure, I could teach a classroom of 1,000 students. It wouldn’t be as good as my classes capped at 30 — for them or for me. Perhaps foremost, we’d lose human connection. Reading remains a not-insignificant spark for that human connection.
What we gain in AI’s meta-reading quantity, we lose in quality. Algorithmic culture idealizes a world in which you get only the information you want, because anything more would be a waste of time. As tech companies seek to convert us to their products, don’t forget to read between the lines about what’s being made obsolete.
Michael Serazio is a professor of communication at Boston College and the author of “The Authenticity Industries: Keeping It ‘Real’ in Media, Culture, and Politics.”