06When AI Begins to Replicate People, What Should People Own?

In the past, AI learned about the world. Increasingly, it is beginning to learn about a particular person. Give a model enough photos and it can learn your face; enough recordings and it can imitate your voice; enough videos and it can learn your expressions and movements. Add articles and interviews from your past, and it may even learn your habits of expression. AI is moving from generating something that looks like a person to generating something that looks like a particular person.

A face may be only the most superficial layer. A digital version of a person also includes expressions, movements, posture, speech patterns, familiar phrases, aesthetic sensibility, knowledge, relationships, experiences, and even the ways they tend to respond to different situations. As more and more traits that belong to "a person" can be learned by models, we may need to ask again: what, exactly, does a person own about themselves?

This may still feel distant for most people, but it is already close at hand for creators. What makes an established creator valuable has never been just their face. Over many years, they have shaped how users understand them today: their way of expressing themselves, their voice, their personality, their content, their stories, and their relationships with users. In the past, these assets were largely bound to the real person. AI is loosening that boundary for the first time. So the question becomes: who does this digital version belong to?

Here is a very practical question: if a company signs you, does that mean it has signed your AI version too? Does the company have the right to train it? Does it have the right to generate it? How long may it use it? Can it continue using it after the contract ends? If a streamer leaves the company, what happens to the AI version trained before they left? If the real person has passed away, can the AI version continue working? Who can authorize it? Who receives the income it generates? If the digital version says something the real person would never have said, who bears responsibility?

Once AI can replicate people, "authorization" may matter more than "generation." The fact that something can be generated technically does not mean it can be used commercially. A mature digital-human industry should not be built around "Can I replicate you?" It should be built around "Have you allowed me to use your digital version in this way?" Who authorizes it, what is authorized, who receives the authorization, where it may be used, how long it may be used, whether it may be altered, whether it may be used for training, whether further versions may be generated, whether it may be commercialized, how revenue is shared, and what happens to the data after the authorization ends—these are the questions that could form the industry’s operating order.

The greater risk may not be something fake that fails to convince, but something so real it does. When a person clearly never said something, yet everyone believes they did; when a celebrity never made an ad, yet that ad circulates widely, the evidence we have relied on—"seeing it with our own eyes" and "hearing it with our own ears"—can become unreliable. One of the most important pieces of infrastructure for digital identity may not be making AI more convincing, but making it possible for people to know whether an AI version was actually authorized by the person concerned.

What we may need is not merely a "digital human," but a "digital identity." Who is this person? Which data belongs to them? Which digital characteristics have been authorized by them? Which AI versions are legitimate? Which institutions may access them? In which commercial contexts may they be used? What is the record of use? What revenue has been generated? When does the authorization end? A person’s AI version is no longer just a video file. It may become a set of continuing digital assets and rights relationships.

Over time, I have come to care more and more about the idea of rights attribution. The easier generation becomes, the more important provenance becomes. Who authorized the digital likeness? Where did the voice come from? Was permission obtained for the training data? When was this version created? Who may access it? How often has it been used? In which commercial projects has it appeared? Rights attribution is not a restriction on AI. Clear rights may be what allow AI to enter commercial society in a meaningful way.

What I would rather see in the future is not "AI takes my job," but "my AI works for me." In the first kind of world, someone takes my data, trains an AI that resembles me, and uses it to make money. I do not know, I did not agree, and I do not share in the value. In the second kind of world, I proactively create my own digital version and decide who may be authorized and what it may do. It can create content on my behalf, serve users, and reach markets I cannot cover in person, while I also share in the value it creates. The discussion of AI and people need not remain forever at "Will AI replace me?" We can also ask: "Can I own an AI that truly belongs to me?"

Could a person’s future income come from more than their work in person? In the industrial era, people primarily sold their time and labor. In the content era, a single work can reach a million people. Could the AI era take this one step further? A person’s face, voice, knowledge, way of expressing themselves, digital character, and even certain lawfully authorized behavioral capabilities could all become digital assets capable of creating value over time. Alongside selling time, licensing digital capabilities could become a very different form of personal asset.

But there are things I do not want AI to have. Not everything about a person should be replicated. There are words I would allow AI to say for me and others I would not; commercial partnerships I would permit and others I would refuse; behaviors that can be simulated and choices that must be made by the person themselves. An AI version is not the person. It may represent the person in carrying out certain authorized tasks, but it should not automatically gain the right to make every decision on that person’s behalf.

What may truly belong to a person is the right to define their own AI version. I decide how it may look, which voice it may use, what it may know, what it may do, where it may appear, and which tasks it may carry out for me; what it may not say or participate in; when its authorization ends; and how the value it generates is shared. I may even have the right to make it disappear. Technology companies can help generate it, platforms can help distribute it, and brands can purchase its services—but who ultimately has the right to define it? I am inclined to believe that the answer should begin with the person themselves.

This is also one reason I later began building Yingjie Intelligence. After so many years working with OST, I had long been focused on content. When AI entered the picture, our first concern was also how to make content faster and better. But as the work continued, I increasingly realized that when AI starts to enter the person themselves, the boundaries of a content company change as well. In the past, we managed a person’s work. In the future, we may also need to manage that person’s AI version. That is one reason I later founded Yingjie Intelligence and began focusing on the authorization and rights attribution of real-person likenesses, voices, digital identities, and behavioral data.

AI is replicating people, but people should own their own AI versions. The question we may face in the future is not only "Who owns this image?" but also "Who has the right to define, authorize, and use this person’s digital version?" I do not reject this development. On the contrary, I think it could open up significant opportunities. But there is one condition: people cannot be absent from their own AI versions. They should know. They should authorize. They should have boundaries. And they should share in the value. AI is learning to replicate people. But people should have the right to define their own AI versions.