I Would Answer the First Email
Somebody important wants my opinion. I rehearse the first reply to a fake AI-policy invitation, and begin making myself useful before finding out who asked.
All my worst apologies have been written by a human. I would like this entered in the record before we blame the machines.
I supplied the words. I supplied the pauses. Sometimes I supplied an explanation so good that halfway through it I began to feel I was the person owed an apology. There is no shortage of human feeling in this operation. Pride is a feeling. So is the desire to get the unpleasant conversation over with before somebody remembers another example.
Now Liza Vilnits writes in Business Insider that her husband asked ChatGPT for help after an argument. He told it he could be short-tempered, identified six situations that set him off, and asked for ways to stop being so short. Then he sent her the thread, his “thoughts and plan.”
I can hear the objection before I finish reading. The man has assigned his improvement to a machine. There will be headings. There will be phases. Somewhere in the document a perfectly ordinary failure to keep his mouth shut will acquire the dignity of a protocol.
It did. A thirty-day “Tone Reset Protocol.”
The title makes me want to behave worse on principle. This is not a useful principle. I have plenty of those.
Vilnits was annoyed too. At first she saw a shortcut around thinking about what he had done. Then she read the exchange. She says it helped them discuss how to move forward. This is her account of a conversation, not a finding that a chatbot cures bad tempers or saves marriages. They already had a therapist. I don’t know what happened on day thirty-one.
But I am interested in the first day. A man has noticed something about himself, asked for help with it, and shown the answer to the person who has to live with the problem. He has not secretly hired a ghostwriter to impersonate a newly considerate husband. The machinery is right there in the thread.
I have an occupational difficulty with this. I write. I want finding the words to count for something. When somebody says a machine helped him say the thing, part of me hears that my trade has been invited to leave the room.
Fair enough if the thing is a poem. I want to know who wrote it. But if I am trying not to snap at somebody, why have I decided the next sentence must establish my qualifications as an author?
I was rude without any help. Apparently that was authentic. The other person got the full, unassisted force of my personality. I don’t see why she ought to value it more highly than a borrowed way of stopping.
There is a sentence in the advice Vilnits describes: “Hey, that came off a little sharp. What did you mean?”
No great writing. No unexpected image. The question has probably been said in one form or another by millions of people who never received credit for developing it. I can feel the writer in me checking whether a little sharp could be improved.
The useful part is farther along. What did you mean?
A question. Not my explanation of what the other person must have meant. Not the opening of a speech about why I react the way I do. I would have to hear something I haven’t composed.
The chatbot offered an explanation as well. In Vilnits’s account, it connected her husband’s history to a nervous system “wired for intensity,” turning him from husband and father into protector and defender. She found the connection illuminating. I have her description, not a verified account of his nervous system, and I am not qualified to examine either of them from my chair.
I know which part would attract me. Protector sounds considerably better than bastard. I could become very fond of a paragraph in which my worst moments were evidence that I had been ready to defend everyone. Show it to the person I snapped at. Ask her to appreciate how much danger I was bravely detecting in her ordinary remark.
That is a use I could make of an explanation. It is not what I know this husband did. Vilnits says the advice made her reconsider her own part in their arguments, and I won’t tell her she was wrong to find something useful. I am trying to keep my own appetite for an impressive excuse from passing itself off as concern about hers.
I dislike being interrupted. I am also capable of spending somebody else’s entire answer preparing my interruption. The result can be delivered calmly. I can wait for a pause and still never have listened. Better manners do not automatically make the conversation less crowded with me.
What Vilnits actually reports is that they sat down and talked through the scripts, questions, and actions. I keep wanting either a fraud or a miracle, and she gives me two people discussing something. Nobody has disappeared into the machine. There is still another person there, able to say the suggested words are wrong, or that the right words aren’t enough.
I can borrow the question. I can admit where I got it. I can try it without producing a certificate of spontaneous goodness.
“What did you mean?”
I have several improvements in mind. None of them leaves room for the answer.
Source: My husband used ChatGPT to become a better husband. I hate to admit it, but it helped.
Somebody important wants my opinion. I rehearse the first reply to a fake AI-policy invitation, and begin making myself useful before finding out who asked.
A second angle looks like a way to check the first picture. I make three views of a disaster that never happened, and find myself keeping the most convincing one.
California is putting limits on AI at work. I trace an imaginary job backward, looking for the moment my employer became entitled to know how I felt.
OpenAI once reportedly considered governments as buyers for its future AGI. I am thinking about the people who would have paid without ever getting a seat at the table.
A copied voice can reproduce a performer's impatience without preserving his right to refuse. I wanted a duplicate who could work while I slept. I assumed I would be the one sending him out.