We Were Promised Flying Cars. We Got AI Writing Detectors

We have arrived at a peculiar moment in technological history. One artificial intelligence helps a person research, develop and express an idea. Another artificial intelligence then inspects the resulting sentences to determine whether the first artificial intelligence was involved. We are building increasingly sophisticated tools to create language—and a parallel industry devoted to detecting and stigmatizing their use.

Meanwhile, almost 64 years after The Jetsons first aired, I am still sitting in traffic.

I began thinking about this after writing I’m Begging You: Learn How to Write With AI, my response to Bret Stephens’ plea that people stop using AI to write. My argument was that Stephens had drawn the line in the wrong place. The danger is not that AI participates in the writing process. The danger is that the human stops participating in the thinking process.

I wrote that article before seeing what my friend Brad Feld or Reid Hoffman had written on the subject. That chronology matters—not because I want credit for independently discovering fire, but because three people approached the same question from different directions and arrived at closely aligned conclusions.

I came at it through the experience of actually writing with AI while remaining intimately involved in the thesis, research, argument, structure and final language. Brad brought 2,400 years of history to the discussion. Reid articulated the promise that should accompany a byline regardless of which tools helped produce the words.

Together, those arguments suggest something larger: “AI writing” is a temporary category belonging to a temporary moment along the technology continuum.

In “AI Writing Is 2,400 Years Old,” Brad shows that the fear of a tool weakening human thought is almost as old as recorded thought itself. Socrates worried that writing would damage memory and give people the appearance of wisdom without genuine understanding. The printing press was treated as a threat to the culture and intellectual discipline of hand-copied manuscripts. Centuries later, word processors were accused of eroding the quality of literature.

The tools change. The indictment remains remarkably consistent.

Brad’s historical case does not prove that every concern about AI is foolish. Socrates was not entirely wrong. Writing did change how humans used memory. Google changed which facts we retain and which ones we remember how to find. GPS has probably weakened our unaided sense of direction. AI can encourage shallow thinking, manufactured authority, factual errors and oceans of prose that are technically correct but not worth reading.

The critics often identify a genuine tradeoff. Where they go wrong is in mistaking a change in human capability for the end of human capability.

New tools remove certain forms of effort while making other abilities more valuable. The printing press reduced the need to reproduce books by hand but vastly increased the value of literacy. Search engines reduced the need to memorize every available fact while increasing the value of evaluating and synthesizing information. AI may reduce the need to manually construct every sentence from a blank page, but it increases the importance of judgment, verification, taste, questioning and intellectual responsibility.

Reid Hoffman gets directly to that responsibility in “The Pangloss of Pangram.” His immediate subject is Substack’s AI-detection feature, which allows readers to scan a post and receive an estimate of whether it is human, AI-assisted or AI-generated.

The detector can analyze patterns in the words. It cannot determine whether those words contain an idea worth considering.

As Reid writes, his name on a page is a vow that he invested energy in the thoughts and remains present in the words. That vow never required him to type every letter himself. The real violation occurs when someone generates an answer, barely reads it and publishes it under a human name. Regardless of the tools involved, Reid argues, a human being must mean what is said.

That is the standard.

A byline has never been an affidavit certifying that the named author performed every mechanical step alone. Authors work with editors. Executives and political leaders work with speechwriters. Some writers dictate. Others begin with research assembled by colleagues. We do not judge the resulting work by counting the number of hands that touched it. We ask whether the named person stands behind the ideas and accepts the consequences of putting them into the world.

An AI detector cannot answer that question. It might estimate how a sentence was produced, but it cannot tell us whether the sentence is accurate, original, useful or insightful. A completely human-written article can be lazy, derivative and dishonest. An AI-assisted article can embody decades of experience, extensive research and sustained human thought.

The provenance of the keystrokes does not settle the value of the work.

We discussed Brad’s and Reid’s articles on the Techstrong Gang, and the conversation reinforced how specific this debate is to our current position along the AI continuum. We remain unusually conscious of AI because it is still visible as a separate destination. We open a chatbot, enter a prompt and receive an answer. The transaction has an obvious seam between human and machine, so we obsess over where one ends and the other begins.

That seam is already disappearing.

AI is being absorbed into search engines, word processors, research platforms, email systems, browsers, publishing software and operating systems. It will retrieve sources, organize notes, transcribe interviews, challenge arguments, check claims and suggest language. Soon, asking whether AI “touched” a piece of writing will be like asking whether the author used spellcheck, Google, autocomplete, transcription software or Microsoft Word’s ability to move a paragraph.

At some point, AI-assisted writing will simply become writing.

That does not relieve us of the need to establish rules. Electricity required safety standards. Automobiles required traffic laws, licensing and roads. Aviation required pilot training, air-traffic control and an enormous safety infrastructure. We did not decide those technologies were risk-free. We learned how to incorporate them into society without reopening the argument over their legitimacy every time someone plugged in a toaster, started a car or boarded an airplane.

AI will travel a similar path. We will develop expectations involving disclosure, copyright, verification, education and accountability. There will be situations where disclosure is essential, particularly when AI is impersonating a person or substituting for human expertise the audience was explicitly promised. There will also be countless routine uses where labeling every AI contribution would be impractical and meaningless.

History is not on the side of technological purity. It is on the side of technological assimilation.

The tools gradually disappear into the process. The human obligation does not. Writers must understand their arguments, verify their facts, mean their words and accept responsibility for publishing them. That obligation is the same whether the first draft began on a legal pad, in Microsoft Word or through a long and occasionally contentious conversation with an AI model.

We were promised flying cars. Instead, we got AI writing detectors.

I suspect the detectors will disappear before the traffic does.

Now, seriously: Where is my flying car?