The proxy self

Generative AI offers an increasingly strange service: it can represent us. “Write this like me.” “What would I say to this?” “Make it sound like my usual tone.” At first this is simply convenient. We give the system examples, preferences and context, and it produces something close enough to our style that we can use it as a shortcut.

But a proxy becomes interesting once it stops merely assisting us and starts appearing in our place. Other people receive messages produced by a model trained on our preferences. We begin asking the system what we “would” say before we have actually decided what we want to say. The proxy does not need to be perfectly accurate to become influential. It only needs to become easier to consult than ourselves.

There is a subtle shift here from using AI to express a self toward using AI to maintain one. The system starts carrying a model of our tone, preferences, values and habitual reactions, and we can begin referring back to that model as if it were evidence about who we are. At some point the interesting question is no longer whether AI can imitate us convincingly. It is whether a tool designed to represent us can slowly become another layer between us and ourselves.