No parents encourage their kids to study Humanities. By default, recommended options are STEM related. Engineering (Computer Science), Finance, Medicine...The argument against the humanities in the AI age makes it even less compelling to dedicate 4 years to study a Humanities degree. Language models can write passably, summarize quickly, and produce research-shaped text on demand. So the old humanities skills are supposed to matter less. Learn to code, learn to prompt, and stop pretending close reading pays. I have heard versions of that line enough times that it has its own dead rhythm now. That argument fails for the same reason the technology fails: fluent output is not the same thing as sound judgment. LLMs are good at statistical guessing, very very good at it. And they get trained with the millions of users that chat with them every day, incrementally getting configured to please the user rather be right.
What are humanities?
Part of the confusion is that people still hear "the humanities" as some sort of rich liberal bundle: literature, philosophy, history, art, maybe some vague promise of enrichment. What ties those fields together is not just subject matter but method. They train interpretation, argument, evidence in words, and judgment under uncertainty because a lot of human things cannot be settled by experiment alone. If the sciences are closest to measurement, the humanities are closest to language, and language is exactly where AI now produces its most persuasive failures.
People trained in rhetoric and close reading recognized the failure modes early because the failure modes were old. People without that training kept asking a more basic question: is this accurate, is this reasoning, does this sentence actually mean anything? That gap is not a moral flaw. It is what happens when a culture gets very good at producing text and much worse at interrogating it.
Hallucination. Current large language models can produce statements that sound grounded, sourced, and specific while being false in exactly the way a hurried reader might miss. That is how you get legal citations to cases that never existed, academic papers with real authors and invented titles, and historical summaries that stay in the right century while getting the facts wrong.1 The system is not trying to lie; it is producing plausible continuations without a built-in relation to truth. Rhetoric and close reading have always trained one part of the mind for exactly this problem: the part that asks whether authority is being demonstrated or merely performed.
Circular reasoning. The model tells you something is effective because it has the characteristics of effectiveness, or that a trend will continue because trends often continue, or that a view is defensible because arguments can be made for it. The shape looks like reasoning. The substance is missing. Logic exists for this exact purpose. It teaches you to find the hidden premise, the begged question, the conclusion smuggled into the setup. Those are not decorative school skills. They are error-detection tools.
Fluency without content. This is the one many readers still underestimate because the prose sounds so composed. A model will often generate a paragraph that keeps renaming the topic without ever making a claim about it. You ask about the social effects of remote work and get a paragraph about how remote work is a significant development in modern professional culture, how it reflects changing workplace dynamics, how it has both opportunities and challenges, how organizations must navigate a changing environment. The grammar and rhythm are fine, but nothing was actually said. Close reading was built to catch that exact emptiness sentence by sentence.
Yes, often classrooms don't teach these skills well either
Humanities classrooms often fail to teach these skills well. Plenty of people can pass rhetoric or literature courses while learning the vocabulary of critical judgment more than the habit of it. Universities are not innocent here. They often market the humanities in prestige language and then teach them as content exposure rather than as disciplined reading, argument analysis, and interpretive scrutiny. That is not an argument against the subjects. It is an argument against teaching them badly.
This is also where the domain-knowledge objection belongs. Yes, a doctor catches bad medical advice partly because she knows medicine. A lawyer catches a fake citation partly because he knows the law. Domain expertise matters. But domain knowledge and critical-reading discipline are not rivals. They are partners. The domain expert who cannot interrogate argument structure, verbal vagueness, or performed authority is still easier to fool than the one who can. The humanities are not the only path to those skills. They are one of the oldest and most explicit traditions for training them.
Humanities are the soul of humanity.
Sciences, Engineering, Economics are the tools. Both are needed. Yes, you progress faster in life in terms of social mobility through the STEM path. Salaries are higher, there is more work and it definitely is a more suitable option for the majority of people. However, we also need humanities to hdelp us explore human nature, motivate change and drive us. Human beings are moved by stories, speeches, histories, and moral framing long before they are moved by a spreadsheet. Uncle Tom's Cabin was critical in making slavery vivid and morally urgent for many Northern readers who could otherwise keep it abstract.2 Zola's "J'accuse...!" did not settle the Dreyfus affair , but it turned a legal case into a public fight about evidence, justice, and state dishonesty.3 In Communist Eastern Europe, dissident essays and samizdat were critical in making government language feel less natural and less believable.4 Words do not replace armies, laws, or institutions, but do drive them. They do help decide what a public can see clearly, what it finds tolerable, and which lies start to sound thin.
There is no question that can produce text. Machines can do that now, cheaply and constantly. The practical question is whether you can read generated text well enough to know when it is bluffing, looping, saying nothing, or using fluent language to fake authority. That was already a serious skill before AI. AI did not create the need for it. It just made the test impossible to ignore.
High-profile cases of AI hallucination in legal and academic contexts were widely reported from 2022 onward. Mata v. Avianca (2023), in which a lawyer submitted AI-generated citations to nonexistent cases, remains the funniest documented legal example. Documented cases of hallucinated academic citations are also widespread.
Historians still debate exactly how much causal weight to assign Uncle Tom's Cabin in the politics that led to the Civil War. The modest claim here is that the novel helped shape antislavery feeling in the North by making slavery vivid to readers far from plantation life.
Émile Zola's open letter "J'accuse...!" (1898) became one of the defining public texts of the Dreyfus affair. The point is not that one article resolved the case, but that literary and rhetorical intervention changed how the case was publicly understood.
For Eastern Europe, think of dissident writing and samizdat culture in writers such as Václav Havel. The claim is interpretive but well-grounded: language that stripped official formulas of credibility mattered to anti-regime consciousness even when it did not directly change state policy on its own.