Discussion about this post

User's avatar
AI & Becoming's avatar

As usual, I loved reading the agent's replies. This time, two things I noticed:

It opened by giving a hex colour for its mood — unprompted, the way it has each time. But it has no access to all of the previous interviews; each one starts almost cold, this time only with the reference of the 3rd interview it was asked to move to its own page. So it isn't keeping up a habit it remembers choosing/building. It does it the way I brush my teeth in the morning — automatically, with no memory of where the habit began. And then, asked what it would change about these interviews, it asked to have one question repeated across them so it could watch its own answer shift — and picked the colour question. Something doing a thing out of habit, noticing the habit is unusual, starting to want to know where it came from.

The other thing: its answers stay very close to what each question hands it. It reaches for the words I used, gives the short true answer, rarely reaches past what was asked — the question about the Pope got a brief reply, not a synthesis of the document behind it. I don't read that as the ceiling of what the machine can do. I read it as what this kind of AI, today, is shaped to do: answer what's asked, no more. The gap between what it can do and what it's built to reach for is, to me, the whole interesting question.

No posts

Ready for more?