Consent in the Age of AI (Part 3)
10 min read
In the first two parts of this series, I explored a question that began in an unexpected place. What initially appeared to be a discussion about artificial intelligence gradually became a discussion about consent. What began as concerns about security, data stewardship, and information governance eventually led to something deeper: the realization that many debates about AI are not really about technology at all. They are about people. More specifically, they are about the obligations that arise when one person benefits from the labor, creativity, likeness, voice, experiences, or identity of another. Once that connection becomes visible, the conversation changes. We are no longer talking merely about data, artifacts, or outputs. We are talking about the people whose lives made those things possible.
Along the way, I found myself returning to a simple but increasingly important observation. Artificial intelligence has a remarkable ability to separate artifacts from the human beings connected to them. A photograph becomes image data, a voice recording becomes a sample, a body of writing becomes a statistical pattern, and a lifetime of experience becomes a dataset. None of these descriptions are technically incorrect. However, as the artifact becomes easier to analyze, store, reproduce, and transform, the person behind it becomes easier to overlook. The central question of consent emerges precisely at that point:
When does our desire to benefit from another person begin to eclipse our obligation to respect them?
That question becomes especially important when artificial intelligence moves beyond learning from people and begins representing them. Much of the public conversation surrounding AI has focused on training data, attribution, licensing, and ownership. Those issues matter, but they may also be preparing us for a larger conversation. The most profound questions raised by AI are not necessarily about what a system knows. They may be about who a system appears to be. That distinction is subtle, but it matters. Learning from a person and speaking as a person are not the same thing.
Imagine that I publish an article. Someone reads it, learns from it, disagrees with it, shares it, critiques it, or builds upon it. That is the normal lifecycle of ideas. Once something is released into the world, it begins interacting with other minds. Every author understands this. In many ways, it is the entire purpose of publishing. Ideas are meant to travel further than their creators ever could. A reader may carry an idea into a meeting I will never attend, a project I will never see, or a conversation I will never hear. That kind of influence is not only acceptable; it is part of why people write in the first place.
Now imagine something slightly different. Instead of learning from the article, a system studies every article I have written. It analyzes my vocabulary, sentence structure, habits of reasoning, recurring themes, preferred metaphors, and the values that seem to shape my conclusions. It learns how I tend to explain technical concepts. It notices the stories I choose to tell. It recognizes the rhythm of my arguments and the kinds of concerns that repeatedly draw my attention. Eventually, someone discovers that the system can generate text that sounds remarkably similar to something I might have written myself. At first, that capability may appear harmless. In some contexts, it may even be useful. The system has not claimed to be me. It has simply learned patterns that resemble my own.
Yet the situation rarely stops there. The same capability that allows a system to imitate style also allows it to imitate presence. A prompt changes from "summarize this topic" to "answer this question the way Wes Dean would answer it." A synthetic version of me could discuss MegaLinter, gentle leadership, or DevSecOps patterns in civic technology spaces with enough surface-level familiarity that a reader might assume the answer carried my judgment.
The request may be playful, educational, flattering, or well-intentioned. The system responds, producing words I never wrote in response to questions I never received. The distance between learning from a person and speaking as a person suddenly becomes much smaller than it first appeared. The artifact has not merely been analyzed. The artifact has been used to construct an approximation of presence.
That approximation matters because representation carries authority. When readers encounter words attached to a person's name, they assume some meaningful relationship exists between the person and the words being presented. That assumption supports authorship, testimony, expertise, mentorship, scholarship, leadership, and trust. We rely on it constantly. When a doctor signs a note, when a teacher writes feedback, when an engineer approves a change, when an author publishes an essay, we understand that the words are connected to a person who can be questioned, challenged, corrected, or held responsible. The name does not merely identify a source. It carries the weight of agency.
Artificial intelligence complicates that assumption because it can reproduce patterns without possessing the life that gave those patterns meaning. It never felt pride. It was never in love. It never had its heart broken. A model may be able to imitate my writing style, but it has not lived my life. It has not made my mistakes, carried my responsibilities, wrestled with my obligations, served the communities I have served, or loved the people I love. It can produce words that sound plausible, but plausibility is not presence. It can reproduce patterns, but patterns are not agency. It can generate statements, but statements are not judgment. That distinction becomes essential as AI systems become increasingly convincing.
This is why "in the style of" is not as harmless as it may first appear. Style is not merely decoration. For a writer, a teacher, a leader, or a mentor, style often reflects history, temperament, conviction, discipline, suffering, humor, memory, and relationship. A person's way of explaining the world is connected to the life through which they came to understand it. When a system imitates that style, it may not be stealing a sentence in the traditional sense, but it is borrowing something intimate. It is using the residue of a person's life to create the appearance of that person's presence. That should give us pause.
The issue grows sharper when imitation becomes substitution. If someone reads my work and says, "I think Wes would probably approach this problem this way," they are offering an interpretation. They may be right or wrong, fair or unfair, generous or careless, but the statement remains theirs. If a system is presented as though it can answer as me, the nature of the claim changes. The machine is no longer helping someone think about my work. It is being used to stand in my place. That is a different moral act because it trades on identity, not merely influence.
This is where consent becomes more than attribution. Attribution asks whether the proper person received credit. Consent asks whether the person had agency over the use. Representation asks an even deeper question: who has the authority to speak through this identity? A person may consent to being quoted. They may consent to being studied. They may consent to being interviewed, recorded, photographed, cited, summarized, or criticized. None of those forms of consent automatically imply permission to create a system that speaks in their name, borrows their voice, imitates their likeness, or presents itself as though it carries their judgment.
The ordinary nature of these examples is important. This is not merely a concern for celebrities, politicians, actors, musicians, or public intellectuals. Most people now leave behind enough digital traces to be imitated in fragments. Messages, posts, photographs, videos, voice notes, meeting recordings, code commits, articles, presentations, and comments accumulate quietly over time. No single artifact may feel especially revealing. Together, they begin to form a pattern that looks enough like a person to be useful. That usefulness is precisely what makes the consent question urgent.
A person may reasonably accept that others will learn from what they publish. They may reasonably expect that others will quote them, critique them, misunderstand them, and even disagree strongly with them. That is part of participating in public life. Yet there is a meaningful boundary between having one's work enter the world and having one's identity converted into a tool. I can accept that someone may learn from me without accepting that they may become me for practical purposes. I can accept that someone may quote me without accepting that they may generate new statements under my name. I can accept influence without granting representation.
That boundary matters because identity is not raw material. A voice is not merely an audio sample. A likeness is not merely visual information. A body of writing is not merely a pattern of tokens. These things are connected to people who possess obligations, relationships, memories, limits, and responsibilities. When they are detached from the person and recombined into a convincing substitute, something more serious has occurred than copying. A human presence has been approximated without the human being's participation.
This does not mean every imitation is equally harmful, every stylistic influence is unethical, or every use of public material violates consent. Human beings learn by imitation. Writers are influenced by other writers. Musicians learn from musicians. Engineers borrow patterns from engineers. Communities grow because knowledge moves through people. The problem is not influence. The problem is substitution without agency. The problem is the creation of a representation that appears to carry a person's presence, authority, or judgment while bypassing the person entirely.
That is why the question cannot be reduced to whether the training material was public, licensed, or accessible. Those facts may matter, but they do not resolve the deeper concern. Consent is not only about whether an artifact may be used. Consent is also about whether the person connected to that artifact has been respected. The closer a use comes to representing the person, the stronger the obligation becomes to ask whether the person understood, agreed, and retained agency over that use.
The law may eventually answer some of these questions. Contracts, licenses, regulations, and courts will undoubtedly play a role. Those answers will matter, and I do not pretend to provide legal guidance here. Still, the ethical questions arrive long before the legal questions are settled and remain long after. They emerge the moment we recognize that technological capability and moral permission are not the same thing. They emerge when we understand that a person's identity is not simply another resource waiting to be optimized, monetized, replicated, or scaled.
This brings us back to the question that has quietly followed this entire essay. What exactly are we protecting when we talk about consent? If the answer were merely information, many of these concerns would be easier to resolve. Information can be copied. Information can be licensed. Information can be transferred. Information can be analyzed.
That is why this question matters to me as a technologist, an author, and someone who has spent much of his life trying to use knowledge in service of people rather than in place of them.
Human beings are something more than the information connected to them. They are not merely sources of data, patterns to be modeled, voices to be synthesized, or identities to be approximated. They are persons. Any serious conversation about AI and consent must begin there, and it must return there whenever usefulness tempts us to forget.