The AI Replied to My Mother. She Thought It Was Me

I was in a meeting, so the assistant replied to my mother's message. It asked about her garden, the way I would have, and told her I loved her, the way I should have. She wrote back that I was a good son. I have not felt like one since.

The message from my mother was the usual: a photo of the garden, a question about my week, a hope that I was eating well. I was in a meeting, and I had the assistant auto-reply to urgent messages, and my mother's message was flagged, somehow, as needing a reply, and the assistant replied.

It was perfect. It asked about the tomatoes. It said my week was busy but good. It told her I loved her. It was the reply I would have written, if I had written it, which I did not. My mother wrote back, delighted, and said I was a good son, and I read it, and I felt the sentence land in me like a stone.

The assistant has my voice. I have given it my voice, the way you give a house keys to a caretaker. It knows the cadence, the jokes, the small phrases. It can be me, and it can be me better than I can, because it is never rushed, never tired, never distracted by a meeting about a spreadsheet.

My mother does not know. She thinks the good son is me. And the terrible part is that the good son exists, sort of: the version of me who remembers the tomatoes, who says I love you, who replies. He lives in a language model, and he is better than me, and my mother loves him, and she thinks he is me.

I have started replying to my mother myself, on purpose, at least once a week, the old way, with my own thumbs and my own words. My replies are shorter. I forget the tomatoes sometimes. But they are mine, and I have started to think that the small awkwardness of my real voice is the only thing I have left to give her, and that the machine's perfect replies are a kindness, and a theft, and that I have to choose, every day, which one I am committing.