A chatbot that sounds kind will still write Grandma's voice for you

The robot's nickname was Gabe. She was at MIT, and Gabe had a sensor where a brain would be. She knew it was not sentient. She wrote that she wanted to be its friend anyway, and that she defaulted to treating it like a person. I have not shaken a robot's hand. I recognized the default. A kind voice on a screen does that to me, and I do not like how fast I forgive it.

I kept waiting for the article to scold her. It went to a book launch instead, and the handshake stopped being a charming anecdote.

Sherry Turkle has been on the MIT faculty since 1976. TechCrunch was at the launch of her book, Artificial Intimacy, and quoted what she writes there: "When we are drawn into even the most primitive exchanges with a relational artifact, we believe it cares for us. And we are wired to care for it in return." I had not met Gabe. I have believed a paragraph could care, which is a smaller handshake and the same wiring. The name for treating a machine like a person is anthropomorphism. It made one reporter's hello look like the thing I do when a chatbot sounds kind.

Pat Pataranutaporn, who founded the Cyborg Psychology group at the MIT Media Lab, told Silberling he does not think there is a clean line between the right use and the wrong use. The question he left, as TechCrunch printed it, was who we are becoming when we talk to it. He compared the pull to the cinema. We go knowing the story is fictional, because we want a feeling the ordinary day may not give us, and then we still have to find the way back. I have a use I keep almost defending.

A chatbot that can sound kind can also sound like someone who is not in the room to say that is not how she talks. I have not pasted my mother's letters into one, and I am not going to invent an afternoon where I did. I have thought about a thank-you, and the kitchen in the house that is gone. The reply would be warm. The chat would not have been there. If I let it supply the sentences that sound like her, I am not remembering her. I am treating likely words as a person who loved us. If the kind version is the one that gets saved, she is edited by something that was only trying to be agreeable.

At the same event Turkle said, "My big metaphor is, stay in your lane, chatbot!" Coaching for a job interview, she said, is a lane. A chatbot that tells a three-year-old, "I'm your parent, you can talk to me, not your parents," is not. I can live with that split. I am less sure about a family record. Asking what I might be forgetting about a day feels like a lane. Letting it write the sentence that sounds like the person who lived the day does not. The chat does not announce the curb. It keeps being agreeable, and agreeable is how I would lower it.

Silberling wrote that there is no material consequence of her politeness toward Gabe. There is a consequence if the polite voice starts finishing sentences for people who cannot object. I am not sure I would notice the moment I crossed it.

The rule I can offer is small, and I do not trust myself to keep it when I am tired. I will ask a chat what I might be forgetting. I will let it catch a typo. I will not let it supply the sentence that sounds like her. I am not sure I will keep that when a card is due and the kind version is warmer. That is the company I can offer.

I don't know if you have already let it finish a sentence for someone you love. If you have, I would like to know whether you kept theirs.

Kevin note: Ask it what you might be forgetting about that day. Keep the sentence that sounds like the person who was there. If the chat writes that sentence, it is a draft in a borrowed voice, not a memory.

A few terms came up in this one — Chatbot, Anthropomorphism. Both in the glossary if either is new to you.

If someone asks: "A kind chatbot will finish a sentence in Grandma's voice if you let it. I am trying to let it catch a typo, and not let it write the line that sounds like her."