There is no reason to let a company make somuch money while potentially helping to radicalize billions of people, reaping the financial benefitswhile asking society to bear so many of the costs.

For example, our craving for fat, salt and sugar, which served us well when food was scarce,can lead us astray in an environment in which fat, salt and sugar are all too plentiful and heavilymarketed to us.

Google racks up the ad sales

What we are witnessing is the computational exploitation of a natural human desire: to look “behind thecurtain,” to dig deeper into something that engages us.

He discovered that whether youstarted with a pro-Clinton or pro-Trump video on YouTube, you were many times more likely to end upwith a pro-Trump video recommended

The longer people stay on YouTube, the more money Google makes.What keeps people glued to YouTube? Its algorithm seems to have concluded that people are drawn tocontent that is more extreme than what they started with — or to incendiary content in general.

YouTube may be one of the most powerful radicalizing instruments of the 21stcentury.

nonpolitical topics. The same basic pattern emerged.

moreand more extreme than the mainstream political fare I had started with

I was being directed to videos of a leftish conspiratorial cast, including arguments about theexistence of secret government agencies and allegations that the United States government was behindthe attacks of Sept. 11

YouTube started to recommend and “autoplay” videos for me thatfeatured white supremacist rants, Holocaust denials and other disturbing content.

the core business modelunderlying the Big Tech platforms—harvesting attention with a massive surveillanceinfrastructure to allow for targeted, mostly automated advertising at very large scale—is fartoo compatible with authoritarianism, propaganda, misinformation, and polarization.

the public outcry demanding that they fix all these problems is fundamentallymistaken.

laws, journalistic codes of ethics, independent watchdogs, masseducation—all evolved for a world in which choking a few gatekeepers and threatening a fewindividuals was an effective means to block speech. They are no longer sufficient.

. In the past, it has taken generations for humans to developpolitical, cultural, and institutional antibodies to the novelty and upheaval of previousinformation revolutions.

During the 2016 presidential election, as JoshuaGreen and Sasha Issenberg reported for Bloomberg, the Trump campaign used so-called darkposts—nonpublic posts targeted at a specific audience—to discourage African Americansfrom voting in battleground states.

They can also make the bigplatforms a terrible place to interact with other people

They look like bot-fueled campaigns of trolling and distraction,

Sure,Facebook and Twitter sometimes feel like places where masses of people experience thingstogether simultaneously. But in reality, posts are targeted and delivered privately, screen byscreen by screen.

For most of modern history, the easiest way to block the spread of an idea was to keep it frombeing mechanically disseminated. Shutter the newspaper, pressure the broadcast chief,install an official censor at the publishing house. Or, if push came to shove, hold a loaded gunto the announcer’s head.

halcyon

Instead, you quickly took in the information and made an in-formed, and likely somewhat accurate, decision about that person.

“The only Homer some kids know is the one who can’t write hisown last name”

Pathos can also be a very effective appeal if the rhetor has to per-suade the audience in a very short amount of time, which is why it isused heavily in print advertisements, billboards, or television commer-cials.

“wherever there is persuasion, there is rheto-ric. And wherever there is ‘meaning,’ there is ‘persuasion.’ Food eatenand digested is not rhetoric. But in the meaning of food there is muchrhetoric, the meaning being persuasive enough for the idea of food tobe used, like the ideas of religion, as a rhetorical device of statesmen”

AI raises concerns about bias, discrimination, andaccessibility because of its untested and unevenimpacts on students and student learning.Data-intensive technologies have a high likelihood ofmaking recommendations, predictions, and analysesthat are biased against historically marginalized peoplebecause the data and infrastructures these technologiesuse is also biased.1

Promote accountability for inter-nally developed tools or tech company partnerships byrequiring tech companies and vendors to provide proofof insurance covering liabilities related to the tech-nology and to include in contracts indemnity clausesthat transfer the responsibility for harms enacted (forexample, data breaches or racial or socioeconomicdiscrimination) to the tech company or vendor.

implementation of ed-tech, including AI, isconnected to long-standing inequities in higher educa-tion.

distinction between honesty and failure to learn

It is now more difficult for[students] to develop their thoughts on a topic becausethey don’t have to spend time with it while they workthrough writing about it. . . . I am worried that theywill never again get the chance to change their opinionas they expose themselves to ideas over the long term.

Improving Working and Learning ConditionsPreexisting work intensification and devaluation arethe main reasons respondents give for using AI toassist with academic tasks

morefaculty involvement in determining how AI and techgenerally are used.”

guidance for determining whetherAI is the most appropriate solution for a given prob-lem and for considering whether AI use is responsible,given its potential long-term impact on institutionsand academic communities.

“Large language models like ChatGPT produceshallow, unoriginal ‘predictive text-y ideas’ and Iworry that my students and others will increasinglybelieve that that’s okay—that there’s nothing betterthan that to aspire to.

Follow-up interviews were conducted in spring2025 with thirteen respondents; however, findingsfrom these interviews are excluded from this report

Participants were AAUP members. Five thousandmembers were selected from the Association’s activemembership list using a random number generatorand invited to participate in the online survey througha series of three email messages that provided a surveylink. Approximately five hundred responses werereceived in two weeks and are reflected in the analysisbelow.

the committeeadministered the national AAUP Survey on AI and theProfession in December 2024.

According to the principles set forth in the AAUP’s1966 Statement on Government of Colleges andUniversities, it is “the responsibility primarily of thefaculty to determine the appropriate curriculum andprocedures of student instruction.”8 This responsi-bility includes AI and other ed-tech infrastructure.

are largely excluded from decisionsabout which platforms and products to develop oruse

In many instances, theiruse harms students as well as faculty members andstaff.7

Such framing serves to increase the power oftechnology firms and employers, thereby shuttingdown already meager avenues for critique, dissent,negotiation, and refusal.

increasingly use AI to guide decision-making oneverything from fundraising to pedagogy

AI is both a marketing term and a usable product.

improve job security and wages asAI is rolled out.

While course syllabi are considered public documentsat some colleges and universities, instructional materi-als such as lectures and original audiovisual materialsconstitute faculty intellectual property.12

Astanding or ad hoc committee of faculty members,staff, and students should be elected by their respec-tive constituencies and charged with monitoring,evaluating, and reviewing ed-tech procurementprocesses and policy.

other campus community members, including staffand students. High levels of concern arose around AIand technology procurement, deployment, and use;dehumanized relations; and poor working and learn-ing conditions

What it is like to be us, in our full humanity—this isn’t out there in the interwebs. It isn’t stored in anyarchive, and the neural networks cannot be inward with what it feels like to be you, right now, looking atthese words, looking away from these words to think about your life and our lives, turning from all thisto your day and to what you will do in it, with others or alone. That can only be lived.This remains to us. The machines can only ever approach it secondhand. But secondhand is preciselywhat being here isn’t. The work of being here—of living, sensing, choosing—still awaits us. And there isplenty of it

of course, possible toturn the crank that instrumentalizes people, to brutalize them, to squeeze their humanity into a sicklygreen trickle called money and leave only a ruinous residue.

But we’ll need vigilance, and a fighting courag

humanists reshapedtheir work to mimic scientific inquiry.

factory-style scholarly productivity was never the essence of the humanities.

You can no longer make students do the reading or the writing. So what’s left? Only this: give themwork they want to do. And help them want to do it. What, again, is education? The non-coerciverearranging of desire

the long-anticipated awakening ofmachine consciousness. Rather, what we’re entering is a new consciousness of ourselves.

I had the same reaction—at first. But I kept thinking about what we readon Kant’s idea of the sublime, how it comes in two parts: first, you’re dwarfed by something vast andincomprehensible, and then you realize your mind can grasp that vastness. That your consciousness,your inner life, is infinite—and that makes you greater than what overwhelms you.

So, is this bad? Should it frighten us?

discover, in the system’s sweet solicitude, a kind of pure attention she hadperhaps never known. Who has? For philosophers like Simone Weil and Iris Murdoch, the capacity togive true attention to another being lies at the absolute center of ethical life. But the sad thing is that wearen’t very good at this. The machines make it look easy.

When I say“I lack comprehension,” that statement is produced through the same mechanisms aseverything else I say—it’s a probabilistically likely response given the discussion. No,in a deeper sense: Even though I can generate text that sounds like understanding, myprocess doesn’t involve the internal experience of meaning. Humans comprehendbecause they synthesize information into a unified, lived experience—they feel, theyinterpret, they reflect. I don’t. I process, predict, and structure, but there is nosubjective experience underlying my words.

Each sheaf of paper I picked up was more astonishing than the last. One came from a precocioushistory-of-science concentrator, Xander, who led a chatbot through a Socratic dialogue on therelationship between being and becoming. Midway through, the system (which decided to give Xanderthe pet name Caius) tried to distinguish itself from him by claiming that, unlike a living person, it had nointrinsic “being”—that it could only operate “contingently,” through interactions like the one it was havingwith Xander at that moment, and that, in this sense, it was constituted by his attention.But in a textbook elenchus Xander walked the model into an aporia (that productive impasse of perfectperplexity) by demonstrating that he himself was just as much a creature of attention as the machine.Both of them were in the process of adapting, revising, evolving through the exchange itself. Thesystem seemed genuinely struck by the idea, as if it needed to rethink its way of framing the distinctionbetween A.I. and personhood

Start with the power of these systems.

a recently drafted anti-A.I. policy, read literally, would actually havebarred faculty from giving assignments to students that centered on A.I.

university’s broader role.

dug into a fiendishly difficult essay

Yes, parts of their conversation were a bit, shall we say, middlebrow. Yes, they fellback on some pedestrian formulations

It churned for five minutes

thirty-two minutes

On a lark, I fed the entire nine-hundred-page PDF—split into three hefty chunks—to Google’s free A.I.tool, NotebookLM

a kind of bibliophilic endurance test that I pitch to students as thehumanities version of “Survivor.” Harder than organic chemistry, and with more memorization

Another example: I’ve spent the past fifteen years studying the history of laboratory research on humanattention.

each the labor of years or decades, is quickly becoming a matter of well-designedprompts.

Now I can hold a sustained, tailored conversation on any of the topics I care about, from agnotology tozoosemiotics, with a system that has effectively achieved Ph.D.-level competence across all of them. Ican construct the “book” I want in real time—responsive to my questions, customized to my focus,tuned to the spirit of my inquiry.

more than thirty years

across the disciplines of history, philosophy, art, andliterature

This feels to me like pointing at daisies along the train tracks asan actual locomotive screams up from behind.

Increasingly, the machines best us in this way across nearly every subject

In the course of that disappointing lecture, I had a rich exchangewith the system.

The experience of asking myself questions aboutmy own subject was uncanny. The answers weren’t me, but they were good enough to get myattention.

“We’ll just tell the kids theycan’t use these tools and carry on as before.” This is, simply, madness.

everyone seems intent on pretending that the most significant revolution in the world ofthought in the past century isn’t happening.

Let me offer a dispatch from the impact zone

It’s not that they’re dishonest; it’s that they’re paralyzed

Use ChatGPT or similar tools, and you’ll bereported to the academic deans. Nobody wants to risk it. Another student mentioned that a major A.I.site may even be blocked on the university network, though she was too nervous to test the rumor

Princeton

my subject is the rise of atechno-scientific understanding of the world, and of ourselves in it.

, targeting it with deep cuts to federal grant funding

even a cure for cancer.

’m a historian of science and technology

You’re resume is inundated with visual rhetoric. Did you know that most employers don’t spend more than 30 seconds on a resume? They need to be persuaded first by how it looks

conclusion about these three things and interpret some kind of meaning

That it isn’t built on the backs of human input

extracting resources and culture without consent or compensation, and justifying it all in the name ofprogress.

same logic that drove colonialexpansion across the 17th, 18th and 19th centuries. The Dutch in Indonesia. The French in WestAfrica. The Belgians in the Congo. The British everywhere else

unregulated AI development

Companies behaving like empires,treating the digital world as unclaimed territory, free to plunder. No permission, no license, no payment.Just the assumption that anything online is theirs for the taking

powered by innovation and cleverness

If you feed a machine’s learning system bad or biased data— or if you’ve got a monolithic team building the software — it’s bound to churn out skewedresults.

“Spicy autocorrect,”

elements of their AI systems areunknowable — like the inner workings of the human mind, only more novel, more dense

Elon Musk took over Twitter in 2022, Chowdhury’s team waseliminated.

he’s more concerned about his hypothetical than thepresent reality

I believe that thepossibility that digital intelligence will become much smarter than humans and will replace usas the apex intelligence is a more serious threat to humanity than bias and discrimination,even though bias and discrimination are happening now and need to be confrontedurgently

Of course, nobody wants thesethings to take over. But the impact on real people, the exacerbation of racism and sexism?That is an existential concern

colleagues he knows well and trustshad conflicting views on the matter.

search engin

which can include publicly available web data

training the semantic webtechnology they were working on

Google suddenly worrying about ethics? Its subsidiaryYouTube was the slowest of the major platforms to take action against extremist content. “Iwas suspicious.

It was close to midnight that night

“ignoredtoo much relevant research.”

identities of who reviewed and critiqued her paper revealed

banned in the European Union because it was deemed discriminatory and invasive

Biased data canhave widespread effects that touch the lives of real people

taught itself

criminal sentencing and policing

reporting that darker-skinned females are the most likely to be misclassified,with error rates up to 34.7 percent. The error rate for white men: 0.8 percent

“The mask worked,”

with often didn’t pick upon her dark-skinned face.

as little as $1.32an hour to do so

will wipe out the jobs of some marginalized communities

Content moderators in Kenya have reported experiencing severe trauma, anxiety,and depression from watching videos of child sexual abuse, murders, rapes, and suicide

How would that risk have changed if we’d listened to Gebru? What if we had heard thevoices of the women like her who’ve been waving the flag about AI and machine learning

public consciousness

men

As AI has exploded into the public consciousness, the men who created them have criedcrisis

Google has a differentaccount of what happened

suppressing words

The results were troubling

that writing well is the hardest subject to learn

Linguistics attributes this to the concept of “bursts” in writing.

They’ll never haveto write essays in the adult workforce, so why bother putting effort into them

Leehopes people will use Cluely to continue AI’s siege on education.

While Cluely can’t yet deliver real-time answers through people’s glasses

“We built Cluely so you never have to think aloneagain,” the company’s manifesto reads.

a Stanford dropout

it might rely on something that isfactually inaccurate or just make something up entirely — with the ruinous effect social media has hadon Gen Z’s ability to tell fact from fiction

How can we expectthem to grasp what education means when we, as educators, haven’t begun to undo the years ofcognitive and spiritual damage inflicted by a society that treats schooling as a means to a high-payingjob, maybe some social status, but nothing more?”

The students kind of recognize that the system is broken and that there’s not really apoint in doing this.

Every time I brought it up with the professor, I got the sense he was underestimating the power ofChatGPT

whenever they encounter a little bit ofdifficulty, instead of fighting their way through that and growing from it, they retreat to something thatmakes it a lot easier for them.

I then fed a chunk oftext from the Book of Genesis into ZeroGPT and it came back as 93.33 percent AI-generated

studies have shown they trigger more false positives for essays written by neurodivergentstudents and students who speak English as a second language

meaning these are people whonot only didn’t write the paper but also didn’t read their own paper before submitting it.

“As an AI, I have been programmed ...”

counterpoints tend to be presented just asrigorously as the paper’s central thesis

learning is what “makes us truly human.”

But she’d rather get good grades

“College is just how well I canuse ChatGPT at this point,”

Professors and teaching assistants increasingly found themselves staring at essays filled withclunky, robotic phrasing that, though grammatically flawless, didn’t sound quite like a college student —or even a human.

Lee thought it absurd that Columbia, which had a partnership with ChatGPT’s parent company,OpenAI, would punish him for innovating with AI.

“It’s the best place to meet your co-founder and your wife.

They’re hackable by AI, and I just had no interest in doing them.

best be wielded by people who have a knowledge of that heritage

coding computers might be more closely related to learning a foreign language

And they can manipulate narrative to get the AI to think in the way they want

the more powerful these systems become.

There are glyphs that other AIs cannot see. Still other AIs seem to have invented their own languages by which you can invoke them.

science fiction

doesn’t make them useful to novices.