Note: Eighteen years ago, Kevin Kelly wrote an essay titled “Better Than Free.” In this essay he focused on the ease of making and sharing copies of things with the advent of digitality and the internet. It stood to reason, he argued, that when the price of making copies drops to zero, the price of the goods drops to zero as well. The question he tried to answer was what is of value in a world where digital copies are free to make and share. What would people still pay for, and why?
I read the essay and wrote a blog post in response, titled “Value in an age of free,” focusing particularly on education. I was reminded of that essay recently, given the advent of a new technology, Generative AI, that can produce all kinds of things, at scale, on demand, including things I had assumed only a person could make, like an explanation pitched exactly at your level or a fresh reading of a poem. Copying was what got cheap with the advent of the internet, imitation of intelligence is what has gotten cheap since, and I wanted to explore what that meant going forward.
“When copies are free you need to sell things which cannot be copied.”
Let’s start with the meaning of the word “value.” Words, as we know, are slippery, and “value” is a particularly interesting case because it holds two meanings that pull in opposite directions: monetary versus moral. Something can have value in the sense that you would pay for it, or value in the sense that it matters, that it shapes who you are. (To be fair money matters – so in certain contexts these two meanings do dovetail.)
But in most situations these two meanings coexist, sometimes uneasily, together. In exploring the question of value in a world of free digital copies, Kelly identified eight qualities he called “generatives,” things that cannot be copied, cloned, or faked, qualities that must be grown, cultivated, nurtured over time. And since these could not be copied their value would actually rise. His list included the following: Immediacy. Personalization. Interpretation. Authenticity. Accessibility. Embodiment. Patronage. Findability. These were, Kelly argued, “better than free.” They were what people would still pay for, yes, but also what would still matter.
Aside: It is interesting to notice that what Kelly called “generatives” has shifted in meaning dramatically over the years. In 2008 it meant qualities that must be generated organically, grown over time through human effort and relationship, things that could not be produced artificially. Now “generative” means the opposite: it describes what machines produce, instantly and at scale. The irony of this shift does not escape me.
In that response I argued that in an age of OpenCourseWare and free lectures on YouTube, universities needed to understand what they were really providing beyond mere information. I also added a ninth generative to Kelly’s list: Belongingness, that deep human need for community, identity, and membership that universities uniquely cultivate.
Kelly’s fundamental move, that in a world of free copies, value migrates to qualities that can’t be copied, has held up remarkably well. It anticipated the entire creator economy, the shift from products to experiences, the premium on “real” in an increasingly synthetic world. Even patronage, a concept that I had difficulty grasping when I first read his piece, has held its ground. Patreon, Substack, and the whole “pay the artist directly” economy have validated it more than I might have expected. Though what people are paying for there is perceived authenticity, the sense that a particular person is at the other end.
What happens when we have, at our fingertips, a kind of intelligence? I say “a kind of” intentionally, because what we have is something like a supremely confident, occasionally brilliant, frequently drunk intern who has read everything ever written and remembers none of it reliably. This intern never sleeps, never charges overtime, never asks for health insurance, and will generate a plausible answer to any question you pose, whether that answer bears any relationship to reality. It is biased in ways it cannot recognize, sycophantic when challenged, and incapable of saying “I don’t know.” It is intelligence of a sort. It is available pretty widely (though money does get you better versions). And it has scrambled the economics of expertise the way the internet once scrambled the economics of information. And yes, the models will get better, the errors will be rarer, and the intern will sober up somewhat. But confabulation is not a defect waiting to be patched out, it is inherent to the very nature of how these systems generate their outputs. None of the losses below depend on the machine being wrong; they happen because it is right often enough.
So, what happens to Kelly’s list when it collides with the easy availability of cheap, unreliable, overconfident intelligence?
The first thing we have to realize is that though Kelly gave us a flat list (i.e. there was no hierarchy, clusters, or organizational scheme) his list contained two very different kinds of qualities. A few of them are transactional (findability, accessibility, immediacy, personalization), i.e. they’re about delivering something to someone efficiently. Others are existential (interpretation, embodiment, authenticity, and belongingness), they’re about the kind of thing that can only come from being human. And this is a critical distinction in the age of AI.
I started with the assumption that the transactional ones were the most at risk, that they would be consumed by a technology that delivers their surface form while hollowing out their substance.
Immediacy is the most straightforward loss. Kelly meant the value of getting something first, of instant access to a human response. AI is immediate in a way no human can match. It answers at 3 AM with the same fluency it has at 3 PM. Whatever value once attached to “I can get you an answer right now,” the machine has claimed it completely. The fact that the answer might be wrong does not diminish the immediacy; it just changes what immediacy is worth.
Findability has been similarly consumed. AI does not merely help you find information; it generates answers with the confidence of someone who may or may not be making things up. The old model (skilled human labor organizing and locating relevant material) had the virtue of being accountable. What replaced it is faster, cheaper, but with little accountability to truth. Accessibility tells the same story from a different angle: AI translates, simplifies, and explains across languages and levels, which is honestly miraculous, though you can never really be sure of the output (as I learnt often to hilarious effect during my recent trip to Brazil).
Personalization, which I once argued was a key educational generative, now belongs substantially to the machines. AI seemingly adapts to your learning style, remembers your struggles, adjusts its pace, all at a scale that would have required impossible teacher-student ratios when I wrote about this in 2008. Whether what it personalizes is actually correct remains, apparently, a secondary concern.
That is four out of nine, gone or fatally compromised. Which leaves interpretation, authenticity, belongingness, and embodiment, with patronage riding on whichever way authenticity goes. When I first worked through this exercise, I stopped here and felt cautiously hopeful. Four survivors. Qualities that require a body, a community, a biography. Qualities that accrue only to creatures who exist in time, who have histories, relationships, and stakes.
But I gradually came to realize that matters were not that simple or straightforward.
For instance, interpretation, which I initially thought might survive in some “situated” form, is also lost. Not because AI cannot interpret well (sometimes it can), but because when interpretation becomes infinitely abundant, you lose the ability to judge it. A student can generate five competing readings of a poem in thirty seconds. Which one is good? On what basis do you decide? The scarcity that once made expert interpretation valuable (you sought out this professor because her reading of Milton was extraordinary) dissolves when the machine produces plausible readings at the push of a button. Interpretation unmoored from a life, from the accumulated weight of someone’s experience with texts and ideas, becomes indistinguishable from noise. What made that professor’s reading worth seeking out was that she had accumulated it over decades of living with those texts, which means interpretation too rests on having a body.
Authenticity, I realized, can be faked. AI can write “I struggled with this for years before I understood it” with the same fluency it applies to everything else. It can generate first-person narratives of failure and recovery that read as lived experience. Deepfakes put real faces on fabricated performances. The surface signals of authenticity (the personal story, the hard-won insight, the voice of experience) are producible at scale. Authenticity survives only when you can verify it through sustained physical presence, when you have watched someone be wrong and recover in real time, over months and years, in a body. We will trust college transcripts less, but we will still go to hear the band play live. Textual authenticity falls; embodied authenticity holds. Which means authenticity is really a special case of embodiment, not a standalone generative.
Belongingness can be hollowed out too. People are already forming genuine attachments to AI companions. Students report feeling more comfortable talking to a chatbot than to a person. AI communities, AI friendships, AI therapists… the felt sense of belonging is producible, even if what you belong to has no one on the other side. Whether that constitutes “real” belonging is a philosophical question I am not sure we know how to answer yet, but the signal is there, and for many people the feeling is real enough to compete with the human version. What survives in it is the part that needs other people in the room with you, which makes it embodiment shared rather than a quality of its own.
So we are left with one: Embodiment. One generative out of nine. The material world remains stubbornly material. You cannot learn surgery from a chatbot. You cannot develop the felt sense of throwing a pot on a wheel, or the tacit knowledge of reading a patient’s body language, or the physical intuition of a chemist who knows by the smell of a reaction that something has gone wrong. Skills that require physical practice, demonstration, and imitation remain intact. A lecture that pivots based on a student’s confusion, a studio critique where collective energy shifts the room, a lab session where something unexpected happens and everyone leans in… these require bodies in space, attending to each other.
There is an analogy that I find clarifying. When recorded music became free (through piracy, then streaming), what became valuable was not better recordings. It was the live show: embodied, unrepeatable, risky. The music industry discovered, painfully, that when copies have no value, what remains valuable is presence and the authenticity of real performance. Education, I think, will learn the same lesson.
Kelly could not have seen that in 2008, because nothing then could counterfeit the other eight well enough to show what lay at the foundation of it all—embodiment. What having a body actually buys us is what I will take up in the next post.
This post builds on Kevin Kelly’s Better than Free and my 2008 response, Value in an age of free.
