Infinite Content: Chapter 10

Known Unknowns of GenAI: The Determinants of Disruption

This is the draft tenth chapter of my book, Infinite Content: AI, The Next Great Disruption of Media, and How to Navigate What’s Coming, due to be published by The MIT Press in 2026. The introductory chapter is available for free here. Subsequent draft chapters will be serialized for paid subscribers to The Mediator and can be found here.

---

In January 2007, I was in the final months of my career on Wall Street. One night, the analyst who covered Apple and I were both working late and he wanted to bounce something off me.

Apple had just announced the iPhone. The stock was on a tear, but he had a Neutral rating on it. He was in a quandary. Should he upgrade AAPL on the promise of the iPhone? Or had he already missed the run? Or, worse, was the iPhone overhyped, nothing more than “…the most expensive phone in the world [that] doesn’t appeal to business customers because it doesn’t have a keyboard…” as Microsoft’s then-CEO Steve Ballmer infamously and dismissively summed it up?

I told him, “You have to upgrade it. Apple is not creating a phone, it’s creating an ecosystem. Next year, it will launch an app store open to third-party developers, who will bring untold innovation to mobile. In coming years, it will add new features, like a better camera, video, and support for GPS, which will enable all kinds of new mobile applications. At the same time, carriers like AT&T, Verizon, and Sprint will invest tens of billions in next-generation wireless technology and infrastructure, increasing the coverage, speed, and reliability of wireless broadband. Within a decade, iPhones will be ubiquitous. People will use them to communicate, navigate, shop, document and share their lives, play games, and pay for everyday items. It will bring with it all types of societal changes. One day, people will select potential mates by swiping through photos of those who are nearby, then willingly pay to get into a stranger’s car to go meet them at a restaurant, and once there, take pictures of the food they’re eating so that all their friends and even strangers can see—all by using their iPhones.”

Under the theory that there are infinite multiverses, there is some universe in which I said exactly those words. Unfortunately, it wasn’t in this one. (That’s why I’m not writing this book from my private island.) I can’t remember what I said. I have a vague recollection of being skeptical about a touchscreen keyboard, so it’s more likely than not that I gave him bad advice. The point of this improbable soliloquy is that technology evolves in emergent and unexpected ways. No one knows how GenAI will evolve. Anyone who tells you he does is lying, to you, himself, or perhaps both.

Who can we turn to for help? The obvious answer: Former U.S. Secretary of Defense Donald Rumsfeld.

At a press briefing in 2002, Rumsfeld was asked about the lack of evidence that Iraq possessed weapons of mass destruction. He responded:

Reports that say that something hasn’t happened are always interesting to me, because as we know, there are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns—the ones we don’t know we don’t know. And if one looks throughout the history of our country and other free countries, it is the latter category that tends to be the difficult ones.

At first, it was considered evasive and became fodder for late night talk show hosts, but this framework is useful for grappling with uncertainty.

Last chapter, we discussed the known knowns of GenAI. By definition, we can’t know the unknown unknowns. In this chapter, we’ll attack that middle category and dissect the most consequential known unknowns about how GenAI will evolve.

Disruptive, But How Disruptive?

In Chapter 1, I introduced the idea that the effects of disruption differ based on the circumstances. Sometimes, as with Kodak and digital photography, disruption is devastating. Other times, as has occurred with AirBNB and traditional hotels, the disruptor(s) and incumbents settle into an uneasy coexistence. In Chapter 8, we applied this idea to media, showing how disruption played out differently across different media sectors.

All disruptors steal away some portion of the incumbents’ customers. The operative question is: how big a portion?

I previously proposed a framework for analyzing those circumstances, built around four questions:

• How hard is it for the new entrant to improve its product and move up market?

• How hard is it for the customer to adopt new products (“customer lock-in”)?

• How much does the new entrant change consumers’ definitions of quality (i.e., how important and appealing are the new features that the upstart introduces)?

• Can the incumbent head off the threat?

In Chapter 8, we used this framework to explain past disruptions of media. In this chapter and the next, we’ll flip it around and use the same lens to predict how GenAI might disrupt media in coming years. To start, we’ll examine the key known unknowns—the questions that will define just how far that disruption goes:

• How realistic will GenAI get?

• How much will it reduce the need for human labor and, therefore, costs?

• Will it enable enough creative control for the most demanding professionals?

• Will consumers accept it—and for which uses?

• Will unresolved legal issues slow progress and adoption?

• What new forms of media will GenAI enable and will they matter?

To be transparent, among all the chapters in this book, I’m most concerned about the shelf life of this one. Stuff is moving fast. But let’s again try to approach these questions from first principles.

How Realistic Will GenAI Get?

Let’s define “realistic.” In the last chapter, I mentioned the Turing Test (originally called “the imitation game”), Alan Turing’s theoretical exercise to test whether a machine could fool a human into believing it is communicating with another human.¹ You could also call this a test of realism. Is the generated output indistinguishable from something created by a real person or capturing a real thing? (A different question is whether people will accept content that is clearly not “real,” and for which use cases, but we’ll get to that later in the chapter.)

Turing didn’t conceive of different tests for different modalities, but you could easily imagine “Turing Tests” for:

• Text: was this written by a person?

• Voice: is this a human voice?

• Music: was this written and performed by people?

• Image: was this drawn/painted/designed by a person? Or, if photorealistic, is this a photo of a real thing or not?

• Video: is this video footage of real things?

Within each of these, you could subdivide the test further. In text, you could imagine different tests for journalism, poetry, and fiction. In music, you could have different tests for pop, jazz, and classical, and so on.

So, how realistic can GenAI get? There is no fundamental technical reason it can’t surpass all these theoretical “Turing Tests.” But some forms of media will be tougher to replicate than others.