There is a version of the AI story that goes like this: STEM wins, humanities loses, learn to code or get left behind. It is a tidy narrative and it is wrong. IBM Technology's Jeff Crume makes the counterargument in a recent video, and it is worth sitting with — particularly for anyone who spends their days working at the intersection of AI systems and human organizations.
The argument is not that the humanities are a consolation prize for people who can't do math. It is that the humanities are load-bearing infrastructure for an AI-driven world, and we have been systematically underinvesting in them at exactly the moment they matter most.
What AI Actually Does — and What It Doesn't
AI systems are extraordinarily good at pattern recognition at scale. Feed a model enough data and it will find structure in that data, generate outputs that fit the statistical shape of what it has seen, and do so faster and more consistently than any human. That capability is genuinely transformative and genuinely useful.What it is not is understanding. An AI system has no lived experience. It has no stake in the outcome. It cannot feel the weight of a decision that affects a real person, cannot recognize when a technically correct answer is contextually wrong, and cannot supply the ethical judgment that determines whether something that can be done should be done. These are not gaps that more data or better models will close. They are structural — they reflect what AI is rather than how mature it is.
STEM fields are essential for building and deploying AI systems. They are not, on their own, sufficient for governing them. The question of what AI should do, whose interests it should serve, and how we should think about accountability when it fails — those are humanities questions. Philosophy, ethics, history, literature: these are the disciplines that have spent centuries developing frameworks for exactly the kinds of questions AI is now forcing on everyone.
The Interpretation Problem
AI outputs are probabilistic. A language model doesn't know what is true — it produces outputs that are statistically plausible given its training. That means every AI output is a starting point for human interpretation, not a finished product. The question is not just whether the output is technically fluent. It is whether it is accurate, whether it is appropriate for the context, whether it carries embedded assumptions or biases that the reader should account for, and whether the confidence of the presentation matches the actual reliability of the content.Humanities training has a name for this kind of work: hermeneutics — the study of interpretation and meaning-making. It is the discipline of asking not just what a text says but what it means, in what context, for what purpose, and with what assumptions built in. Applied to AI outputs, it is the skill that separates people who use AI effectively from people who use it credulously.
The practical version of this skill shows up constantly in real AI deployments. A clinical decision support system recommends a treatment pathway. A fraud detection model flags a transaction. A content moderation system removes a post. In each case, the AI has produced an output. The human receiving that output has to interpret it — has to ask whether the confidence is warranted, whether the context was correctly understood, whether the recommendation should be acted on or questioned. That interpretive work is a humanities skill applied to a technical artifact.
Prompt Engineering Is Rhetoric
One of the more interesting reframes in Crume's argument is the relationship between prompt engineering and rhetoric. Rhetoric — the art of effective communication, of structuring language to achieve a specific effect on a specific audience — is one of the oldest disciplines in the humanities. It was considered foundational to education for most of Western history. Then it became unfashionable, associated with manipulation rather than craft.Prompt engineering rehabilitates it. Getting useful output from a language model is fundamentally a communication problem. You are trying to convey context, intent, constraints, and desired form to a system that will interpret your words probabilistically. The better you understand how to structure language, how to establish context, how to specify what you want and what you don't want — the better your outputs will be. That is rhetoric. The technical wrapper is new. The underlying skill is ancient.
This matters for how we think about education and workforce development. Organizations investing heavily in AI tooling should also be investing in the communication and critical thinking skills that determine how effectively those tools get used. A workforce that can write clear, precise, contextually appropriate prompts will consistently outperform one that can't — and that workforce is built through humanities training as much as technical training.
Critical Judgment as the Scarcest Resource
Crume's central claim is that the most valuable skill in an AI-driven world is critical judgment — the ability to evaluate whether an AI-generated answer is not just technically fluent but ethically and contextually appropriate. This is not a soft skill in the pejorative sense. It is the hardest skill to automate and the one that determines whether AI deployment goes well or badly.Critical judgment requires knowing enough about a domain to recognize when an AI output is wrong even when it sounds right. It requires ethical frameworks robust enough to evaluate tradeoffs that don't have clean technical solutions. It requires historical awareness of how similar technologies have played out, what failure modes have recurred, and what interventions have and haven't worked. It requires cultural competence to recognize when a technically correct answer is wrong for a specific human context.
None of those requirements are met by technical training alone. They require the kind of broad, integrative thinking that humanities education develops — the habit of asking not just how but why, not just what but whether.
The Division of Labor That's Actually Emerging
The practical picture of how AI is reshaping work is becoming clearer, and it looks less like "AI replaces humans" and more like "AI absorbs certain kinds of cognitive work, leaving other kinds more concentrated in human roles." The work AI is absorbing is primarily pattern-matching at scale — the retrieval, summarization, and generation tasks that consumed significant portions of knowledge work. The work that remains concentrated in human roles is judgment, interpretation, relationship, and accountability.That shift doesn't make technical skills less valuable. It makes the combination of technical and humanistic skills more valuable than either alone. The person who understands both how an AI system works and how to evaluate its outputs critically, communicate its limitations honestly, and govern its deployment ethically is not competing with AI. They are the essential layer that makes AI deployment functional.
Crume's framing is that AI handles the mundane, freeing humans to focus on the big questions of purpose and value. That is an optimistic read, and it's probably right directionally. The big questions of purpose and value are humanities questions. An era that produces more of them rather than fewer is an era in which humanities matter more, not less.
What This Means in Practice
For organizations deploying AI, the practical implication is that workforce development programs focused exclusively on technical AI skills are leaving capability on the table. The employees who will use AI most effectively are the ones who combine technical literacy with strong communication skills, critical thinking habits, and ethical reasoning frameworks. Hiring and development programs should reflect that combination.For individuals navigating an AI-transformed job market, the implication is that humanities backgrounds are not a liability to apologize for. They are preparation for exactly the interpretive, communicative, and judgment-intensive work that AI deployment concentrates in human roles. The question is whether to combine that background with enough technical literacy to work effectively with AI tools — and the answer to that question is clearly yes.
For the broader culture, the implication is that the decades of defunding and devaluing humanities education have been a strategic error precisely when those disciplines are becoming more important. Reversing that trend is not a luxury. It is infrastructure investment for an AI-driven economy.