N.B. This essay is about three interrelated topics: (a) the art and soul of human writing, (b) the ethics of intellectual delegation via AI, and (c) the mechanistic reality of intelligence.
Many years ago, I sat in a university cabinet meeting as we debated some forgotten, but seemingly momentous issue. At some point during the discussion, one of the other vice presidents turned to the provost and asked, “Who’s doing your thinking on this topic?” It was an innocuous question, albeit poorly phrased, seeking only to know who wrote the provost’s briefing papers.
The provost took umbrage, sat up straight, looked squarely at the interlocutor and said, “I’m doing my own thinking on this topic.” This exchange elicited a few wry smiles around the conference table, and a sheepish apology from the vice president who posed the query. In today’s world, the question might have been about which AI the provost was using to synthesize position papers.
The Crucible of Writing
In my professional lifetime, I have read thousands of student essays, academic memoranda, and research proposals. Some were painfully poor, a mutual exercise in masochism by writer and reader alike. A rare and memorable few were brilliantly parsimonious, aiming unerringly for the intellectual heart of the topic, simultaneously uplifting the reader with their eloquence. Some were pedantic and pedestrian. As for the rest, they are why words such as turgid, meandering, convoluted, and hackneyed were invented, a literary purgatory where Strunk and White suffer eternally.
Crafting well-written prose is an art, one requiring diligence, practice, and most certainly, talent. While clarity of thought, sound grammar, and a rich vocabulary are essential for competence, rhetorical artistry is a different matter. The latter is a rare and precious gift, one where a writer’s words can paint a vivid and captivating masterpiece, one unique in each reader’s mind.
Three foundational principles separate egregiously bad writing from the sublime.
- First, do not annoy the reader. If your reader finds themselves lost in a miasma of grammatical errors, tortured syntax, logical fallacies, and muddled messages, gnashing their teeth in frustration, you have failed. Whether you are a budding novelist, a frustrated student writing a term paper, or a proposal writer beseeching a funding agency, once the reader is annoyed, all hope of a successful outcome is irretrievably lost.
- Second, do not bore the reader. If your prose is pedantic and plodding, your readers may find themselves pining to inspect their cuticles, bemoaning the deforestation caused by your missive, or reading the nutritional labels on their canned goods in a desperate search for the intellectual stimulation you failed to provide.
- Third, excite and satisfy the reader. Hook them with a compelling story, something doubly important for a fact-based document. Make them want to keep reading, personally invested in what comes next and how the story ends. This is what separates great writers from the merely pedestrian ones. Whether a memorandum on deferred building maintenance or the meaning of life itself, good writers do not merely inform, they motivate and inspire.
The goal of writing is to communicate and connect mind to mind, not to impress. (See Simplifying Communications.) Current AIs suggest that I sometimes violate this dictum, losing myself in the rhythm of language. I continue to struggle, because writing brings me intellectual and emotional comfort. (See My Balm in Gilead for my personal reflections on writing.)
Generative AI as a Disruptive Force
The power of deep neural networks, transformers, and mixture-of-experts (MoE) architectures far exceeds what I might have predicted even a few years ago. Our current AIs are statistical predictors, albeit astonishingly good ones, capable of synthesizing documents and taking actions based on an enormous corpus of human writing and digital records. As such, they have enormous value as semiautonomous AI agents capable of offloading many routine and non-routine cognitive tasks, including generating the documents that are the detritus of the modern knowledge economy. AIs can also be effective collaborators, suggesting approaches and ideas beyond the experience and insight of any individual.
As with all disruptive technologies, we humans are wrestling with Amara’s law: we tend to overestimate the effect of a new technology in the near term, but we underestimate the effect in the long term. In that interregnum between technological emergence and societal normalization, we struggle to define social norms and the ethics of acceptable use. Generative AI is but the latest example of the disconnect between the accelerating pace of technical advances and the slower evolution of human social understanding. The lesson of history is that this is a combustible combination with uncertain outcomes.
Any new technology heightens the importance of thoughtful, informed debate; AI is no exception. All too often, digital echo chambers sharpen ill-considered positions based on rumors, deceptions, and AI slop. Serious debate is a long form process, grounded in facts; it is not a collection of small digital soundbites. One need not be an expert on the numerical stability of backpropagation to grasp the essence of large language models as statistical predictors, nor does one need to be a student of macroeconomic history to appreciate its disruptive effects. (See Contemplative Reflection and Instantaneous Communication.)
AI As Mimic
The power of generative AI also challenges our sense of humanity in ways mechanical augmentation never did. Out of self-centered hubris, we have long viewed thinking, reasoning, and communicating as our evolutionary crown. In the name of efficiency, we are delegating that most human of tasks – communicating ideas, meaning, and nuance – to our AIs.
I asked Google Gemini AI to rewrite the preceding paragraph in the literary styles of Hunter S. Thompson, Ernest Hemingway, and William Faulkner. Because Gemini had read everything these three had ever written, it is no surprise it was able to statistically predict and mimic their styles and rhythms of writing. The irony is that Gemini implicitly engaged in self-criticism.
Faux Hunter S. Thompson. We are running headlong into a savage, neon-lit nightmare, and nobody is checking the brakes. For centuries, we strutted around this polluted planet like cosmic kings, convinced that our ability to think, reason, and babble was God’s own gift to our pathetic species. It was the ultimate, self-centered high. But look at us now. In the name of ‘efficiency’—that foul, corporate buzzword—we are voluntarily ripping out our own brains and feeding them to the silicone monsters. We are delegating the very soul of human communication to a bunch of faceless algorithms, and the terrifying truth is that we’re doing it willingly, with a smile on our faces and a cocktail in our hands.
Faux Ernest Hemingway. The machines of the past changed what a man could do with his hands. The new machines are different. They challenge what a man is. For a long time, men believed that thinking and speaking made them better than the beasts. It was an old vanity. Now, men want things to be easy. They give the thinking and the words to the machines. They give away the meaning and the nuance. It is a bad trade, and it leaves a man empty.
Faux William Faulkner. And it was not merely the mechanical extension of bone and sinew—which man had long since endured and accepted—but a stranger, darker usurpation that struck at the very marrow of his collective ghost; because man, in his enduring and tragic vanity, had for generations unreckoned deemed his capacity to reason, to articulate, to project into the uncaring void his own small, fragile griefs and triumphs, as his supreme inheritance; yet now, under the bloodless banner of expediency, he was willingly surrendering that ancient, agonizingly wrought prerogative—the brokering of essence, meaning, and the fragile, fleeting nuance of human breath—unto a cold, synthetic deity of his own reckless forging.
The crucial point is that these paragraphs are mimicry, brilliant, but still mimicry. We recognize the styles because each of the three writers was an original voice, explaining the human condition in their own distinctive way, writing from lived and unique experiences.
AIs have been trained on everything the three authors ever wrote but have experienced nothing – at least not yet. Is experience required? No, not in every case, but empathy and understanding are, and those cannot be found solely in a statistical recapitulation, however clever.
Delegating Expression
What gets lost in our AI delegation? It is a brewing conflict that pits expediency and least effort against historical ethical norms, human creativity, and the satisfaction of personal accomplishment. The technology may be complex, but the ethics are simple. If you use AI to accomplish a task, you must acknowledge that reality, just as you would a ghost writer. Concealment is the ethical violation, not the collaboration.
In that spirit, I recently told a group of academics that generative AIs already write better and much faster than most of our students ever will. The students know this too. Despite academic prohibitions, the use of AI to write student term papers is ubiquitous. In the endless cat-and-mouse game of academic ethics that pits teachers against students, companies actively profit by selling tools to make AI-synthesized prose flawed enough to pass as human.
Like many of you, I sometimes use AIs to evaluate my writing, and I occasionally adopt a turn of phrase, but I do not use it to create content under my byline. Asking an AI to write prose you then claim as your own is a bit like asking the gym manager to lift weights on your behalf. You may learn something about exercise techniques, but it will not improve your health. Seeing is not doing, and delegation is not ownership. If your name is on it, you bear the ethical responsibility for its creation and its attribution.
AI Prompts and Speaking Truth to Power
If I have learned anything as a leader – and some may well doubt that I have – it is that those with whom you work must feel free, compelled even, to tell you the truth, even when – especially when – you are self-rationalizing painful realities. Hence, one of my bedrock directives to those who work with me is simple: when you think I have erred, your most important job is to close the door, look me in the eye, and speak truth to power, “Dan, you screwed that up. Here is why, and here is how you must fix it.” It can be humbling and frustrating, but it is the lifeblood of a healthy organization.
All of which brings me to AI as sycophant, rather than an honest, dispassionate partner and writing critic. By default, AI heaps unearned praise on even doggerel. In contrast, a good editor tells you what you cannot see, or worse, do not want to admit to yourself.
This is why the quality of the AI prompt matters and why the phrase “prompt engineering” has entered our lexicon. Do not just ask in generalities, “Is this good?” or “How can I improve this?” Instead, ask direct, pointed questions: “Be direct and brutally honest. Does this essay have clear arguments? Does it address the question posed? What are the weakest parts? Now suggest how to strengthen them.”
Equally important, if you want concrete feedback about writing, establish the role of the AI. For example, “You are an experienced editor at a nationally respected publication. Read this with a critical eye to uphold the editorial standards of your organization.”
Delegating the Future
Despite having experienced the computing revolution firsthand, I remain astounded by the rapidly advancing capabilities of generative AI. (See Meditations on AI: History, Hype, Myth, Reality, and Futures for some personal perspectives.) However, in my technical opinion, grounded in four decades of computing research, we are still far from artificial general intelligence (AGI).
I do believe we will achieve AGI someday, and perhaps even artificial superintelligence (ASI). Admittedly, that is an expression of faith, but it is grounded in scientific precedent. Nature offered proof that heavier than air machines were possible, and we gleaned the physical principles of lift and aerodynamics from those examples, then used those principles to replicate flight via other mechanisms (e.g., engines and propellers or jet turbines). Pointedly, aircraft fly far faster and higher than any biological organism ever has.
Theologians, philosophers, and scientists have long explored what it means to think and to possess a theory of mind. Like most scientists, I have a mechanistic view of intelligence, namely, that complex cognitive behaviors such as learning, reasoning, and perception emerge from physical mechanisms. The famous Turing test is one manifestation of operationalism.
Just as with flight, nature shows us the way. Intelligent, thinking machines do not require gigawatts of power or a massive data center. After all, the human brain operates roughly 20 watts.
Using our imperfect, high-energy silicon approximations, we are beginning to decode the real thing. To date, we have mapped the complete connectome of model organisms such as C. elegans and Drosophila melanogaster, as well as a cubic millimeter of the mouse brain. Computational modeling of those connectomes replicates aspects of the biological neural network’s functionality. In a self-referential twist, generative AI is accelerating connectome mapping by automating the identification of connections in microtome sections. This progress suggests the biological mechanisms of intelligence are understandable and replicable.
Dystopian Fears and Utopian Dreams
Echoing dystopian science fiction themes, such as that in Ex Machina, several respected experts believe advanced AI, particularly ASI, could be the death of humanity, with goals and objectives antithetical to our own.
One day the AIs are going to look back on us the same way we look at fossil skeletons in the plains of Africa. An upright ape living in dust with crude language and tools, all set for extinction. Ex Machina (2014)
The risk and possible mitigations are topics worthy of thoughtful discussion, one more grounded in fact than dystopian science fiction, however entertaining it may be. These are deep philosophical and technical questions that encompass self-awareness, cognition, and theory of mind, topics at the center of what it means to think and feel, the core of our humanity.
Regardless of perspective – AI as a modern miracle that can uplift humanity, an inevitable extinction event, or a belief that AI, like all technologies, can be used for good or ill – what matters most is thoughtful and informed debate about how and where we use AI. That means thinking, listening, and debating using the proven human tools of fact-based reasoning and logic, enabled by effective oral and written communication. (See Just the Facts, Ma’am: Reasoning Is Not Dead, Jim)
Let me be clear, I am an AI proponent; it can be an effective intellectual lever when used wisely.
Echoing Danny Hillis, from the halcyon days of Thinking Machines, perhaps we should aspire to build a machine that will be proud of us and the stories we tell.
Until then, when the world asks who is creating our ideas and telling our stories, the answer must remain poignantly clear: We are still doing our own thinking. Cogito, ergo sum.
Please leave a comment …