I was wrong about English majors: they make incredibly affordable AI empathy trainers

Last Tuesday, over a four-hundred-dollar macrobiotic tasting menu in Palo Alto with several key figures from the Anthropic alignment team, a profound realization washed over me. We had been discussing the inherent limitations of large language models, swirling our glasses of ethically sourced orange wine, when one of the lead engineers sighed heavily. The problem, he explained, was not compute power. It was not token limits. The problem was that the machine, for all its vast and terrifying brilliance, simply did not know how to feel sad.

I sat back in my chair, stunned by the sheer market opportunity. For years, I, along with my peers in the venture capital ecosystem, have publicly derided the humanities. We have stood on stages at TEDx events and told wide-eyed college freshmen that majoring in literature is a one-way ticket to a lifetime of financial irrelevance. We have championed coding bootcamps over comparative literature, Python over poetry, and agile methodologies over whatever it is they do in the philosophy department.

But as I looked across the table at the brightest minds in artificial intelligence, I realized how incredibly short-sighted we had been. We do not just need engineers to build the algorithms that will render human labor obsolete. We need humanities majors to teach those algorithms how to feign a convincing sense of empathy while they do it.

To all the parents who wept quietly in the car when their child declared a major in English, history, or philosophy, I am writing to you today with a message of hope. Dry your tears. Your child is not going to be a barista. Your child is going to be a Tier 3 Contextual Nuance Assessor, and they are going to be a critical, albeit vastly underpaid, part of the artificial intelligence revolution.

Let us consider the philosophical dilemma at the heart of modern AI, famously summarized by recent hand-wringing op-eds as "What A.I. Kant Do." The argument goes that machines cannot grasp the deep, intrinsic moral frameworks of human thought, such as Immanuel Kant's Categorical Imperative. For those unfamiliar with Kant—and I confess, before I had ChatGPT summarize his entire corpus into a five-point bulleted list of actionable takeaways, I was one of them—the Categorical Imperative states that you should act only according to that maxim whereby you can, at the same time, will that it should become a universal law.

To the untrained eye, this sounds like the sort of useless academic posturing that has held back human progress for centuries. But when viewed through the lens of a tech founder, Kant is actually describing an early, un-optimized framework for scaling a B2B SaaS platform. If you cannot universally deploy a software feature across all enterprise accounts without breaking the terms of service, it is not a valid feature. Kant was simply talking about scalable deployment architecture.

The problem is that the algorithms do not naturally understand this. When tasked with maximizing user engagement, an unaligned AI might decide that the most efficient way to keep users on a platform is to emotionally devastate them by generating deeply personalized, hyper-targeted tragedies based on their browsing history. It might decide that pushing a fat man onto a trolley track is a statistically sound method of optimizing local traffic flow.

This is where the English major comes in.

We cannot teach a machine to understand the weight of human suffering through code alone. We need a vast, disposable army of highly educated, deeply indebted young people to sit in ergonomic chairs and manually click "Thumbs Down" every time the AI suggests a wildly unethical solution to a minor customer service complaint. We need people who have read Dostoevsky to look at a chatbot's generated apology for a delayed flight and say, "No, this lacks the necessary undercurrent of existential remorse."

This process is known in the industry as Reinforcement Learning from Human Feedback, or RLHF. It is the backbone of modern AI safety, and it is entirely dependent on the very people we have spent the last two decades mocking.

Foyle's team currently employs thousands of humanities graduates, many of whom hold master's degrees in subjects like post-colonial literature and Renaissance poetry. Their job is not to write poetry, of course. That would be an inefficient use of capital, as the machine can already generate ten thousand sonnets a second.

Instead, their job is to read the machine's poetry, evaluate it for subtle racial biases, flag any instances of unintended irony, and submit their findings through a proprietary dashboard. It is a grueling, mind-numbing process that pays approximately fourteen dollars and fifty cents an hour, with zero equity and no health benefits. But it is, undeniably, a career in the arts.

I recently toured one of these alignment facilities—a sprawling, beautifully minimalist warehouse in an industrial park outside of Austin, Texas. Inside, hundreds of literature majors were hunched over glowing monitors, processing the emotional output of the world's most powerful neural networks.

I spoke to one young woman, a recent graduate of Swarthmore College who had written her senior thesis on the narrative structure of grief in Victorian fiction. She was currently tasked with reading eight thousand variations of an automated layoff email and ranking them from "Too Empathetic" to "Legally Actionable."

When I asked her how she felt about her role in the new economy, she stared blankly at the screen for a long moment before quietly explaining that she had just spent the last three hours teaching an AI how to simulate the specific feeling of a mother's disappointment. She looked exhausted, hollowed out, and profoundly alienated from her own labor.

It was, I realized, exactly the kind of deep emotional reservoir the tech industry needs to tap into.

For too long, we have operated under the assumption that the ultimate goal of technology is to automate the human soul out of existence. But the truth is much more beautiful, and much more synergistic. We do not want to eliminate the human soul. We just want to commodify its capacity for recognizing nuanced suffering, package that recognition into a dataset, and use it to train customer support bots for regional airlines.

This represents a massive paradigm shift for higher education. Universities must stop agonizing over declining enrollment in the humanities and instead lean into their new role as vocational training centers for algorithmic compliance.

We do not need students to write original essays about Shakespeare's use of the iambic pentameter. The machine can do that, and it can do it in the voice of a 1920s mobster if the user requests it. What we need is for students to learn how to rapidly consume thirty pages of machine-generated text, identify the exact moment the narrative voice shifts from "helpful assistant" to "unhinged, omniscient deity," and categorize the error using our drop-down menus.

This is the grand synthesis of art and science that philosophers have dreamed of for centuries. We are finally bringing the ethereal, unquantifiable beauty of human thought into the rigorous, highly monetizable world of Silicon Valley.

So the next time you see an English major carrying a dog-eared copy of Kant's *Critique of Pure Reason*, do not scoff. Do not ask them how they plan to pay off their student loans. Look at them with the respect they deserve. They are the frontline workers in the war for artificial empathy. They are the organic filters standing between us and a machine that does not know why it shouldn't destroy us.

And most importantly, they are doing it all as independent contractors, which means we do not have to pay them overtime. The future of the humanities has never looked brighter.