While writing this series, I kept returning to one question. If generative AI continues to develop at its current pace, many things that seem unimaginable today may feel ordinary in a few years. Generating a photograph may be ordinary. Generating a video clip may be ordinary. Generating a character may be ordinary. Creating a digital human may be ordinary, too. Tools will eventually recede into the background. The question that may remain is: What then? Once we can generate almost anything, what do we actually want to generate?
Over the past two years, I have been less and less astonished by “what AI can do now.” It is not that AI has become any less impressive; it is that impressive breakthroughs have become too frequent. Stronger models have not automatically produced more good content. They have simply produced more content.
Generation solves the question of “whether there is something”; creation solves the question of “why.” Many creative challenges used to be production challenges first. Generative AI is rapidly lowering the difficulty of that layer. Many things have shifted from “Can it be made?” to “Which version do we want?” The more answers there are, the more important “why” becomes.
We may overestimate the scarcity of “creating” and underestimate the value of “choosing.” Going “from nothing to something” is becoming easier and easier. A creator’s work will increasingly call for another capability: going from something to nothing. Out of a hundred images, delete ninety-nine. Out of ten directions, abandon nine. When a model tells you there is still more you can add, say: enough. Tools keep opening possibilities; people close most of them.
Once a face can be generated, what is truly scarce is “what has happened to this person.” A face lets us recognize a character; a past lets us believe in that character. Without memory, there is no real continuity.
When content is infinite, memory may instead become an asset. What may be truly scarce in the future is not whether there is something to watch, but whether there is something worth remembering. Attention happens in the present; assets live in time. IP is not how many views you have today; it is whether what happened yesterday still counts today.
The stronger AI becomes, the more valuable relationships may become. Once so much can be generated, relationships still take time. Content can be generated; genuine trust and shared experience still take time.
This is why I still believe in organizations. A good organization allows experience to remain. A mistake one person has encountered does not need to be repeated by the next; a method proven by one account can inform the next account; a failure can become a rule. A genuinely good organization in the AI era should not be many people busy together. It should be one in which the judgments of many people can be preserved.
Once a work can change for each person, we still need things held in common. If everything changes for you, what do we have left in common? Culture exists, to a large extent, because we share certain things. The value of some works lies precisely in the fact that they do not change for you. You need to step into them and understand them, rather than having them continually change to accommodate you.
When AI begins to replicate “me,” responsibility cannot be replicated away. An AI version can undertake certain tasks for me, but it should not automatically inherit all decision-making authority. The more technology can replicate a person, the more we need to know: who signs off in the end?
Perhaps what cannot truly be outsourced is responsibility. AI can help us do more and more, but capability can be delegated; responsibility cannot be delegated without limit.
So I am increasingly less concerned with whether AI looks like a person. What interests me more is: when AI really does become more and more like a person, what will people still need to do? My current answer is not complicated. People still need to choose. Still need to remember. Still need to build relationships. Still need to take responsibility.
Choice, memory, relationships, responsibility. Choice: AI expands possibilities; people decide what is worth keeping. Memory: generation can start from zero, but characters, IP, and organizations must have a past. Relationships: attention can be bought and content can be generated, but genuine trust and shared experience still take time. Responsibility: AI can act as an agent for capabilities, but someone must ultimately know where the boundaries are and be willing to take responsibility for the choice.
Technology will eventually recede into the background. Perhaps one day, the very phrase “AI content” will sound strange. At that point, we will return to the oldest questions: Is this any good? Why should I watch it? What does it leave me remembering? Am I willing to believe it? What does it have to do with me?
Generation is not the endpoint. A real industry always takes shape after a technological breakthrough. After generation, who will choose? After generation, how will characters continue to live? After generation, what can become an asset? After generation, how should a company organize itself? After generation, is a work still a fixed file? After generation, when AI begins to replicate a person, who holds the rights? After generation, how can an ordinary person leave behind something that is truly their own?
I am increasingly willing to admit: I do not know. I do not know how powerful AI will become, how many roles will ultimately change, whether generative works will become mainstream, or whether everyone will truly have their own AI version in the future. But “not knowing” does not mean doing nothing. For someone working in this industry, a more grounded approach may be: get involved first. Make a little. Get some things wrong. Then adjust a little. I am increasingly unconvinced by predicting the future from the shore. Many changes can only be understood once you enter the water and find out how deep it really is.
The future is not predicted into existence. I wrote this series not because I think I have fully understood AIGC. Quite the opposite: the more we do, the more questions emerge. At first, we cared whether AI could make content. Then we began asking what was worth making; later, what was worth retaining; and eventually we arrived at this: when more and more things can be generated, what are people responsible for? For now, my answer is: Choice. Memory. Relationships. Responsibility. I do not know whether these four words will still be enough in a few years. Perhaps questions we cannot imagine today will arise by then. So, at this point in After Generation, I do not want to offer an especially grand conclusion. AI will not automatically bring better content, nor will it automatically bring a worse world. It has simply magnified what we are able to do. And once capability is amplified, choice is amplified, and responsibility is amplified. We used to be able to leave many questions unanswered because we could not do certain things. Increasingly, we will not be able to. Because “we cannot” is slowly disappearing. The question that remains will become more and more direct: Now that you can do it, why do you want to do it this way? Perhaps this is the truly difficult question of the generative era. Not what else AI can generate. But when it can generate anything, what do we decide to leave behind.
Copy this essay for someone thinking about AIGC content, IP, organisation or digital identity.
