My wife and I have had an embodied agentic intelligence in our house for the past four years.
Given the sociopolitical ramifications of this disclosure, I’ll be more detailed. We—mostly she—pre-trained the agent for more than nine months. Then the agent entered a sustained and ongoing period of reinforcement learning. Feedback sources range from the rollyness of frozen blueberries to the forcefulness of our dog’s movement along the vector to the front door.
Assiduous regimes of supervised fine-tuning have yet to fully align the agent. In truth, things are getting increasingly out of control.
I was putting the agent through our usual power down routine—last energy inputs, some cleaning, and reading two short, comforting training materials. The agent likes to listen to more training materials while drifting into memory consolidation. I leaned over to turn on the audio player, and that’s when it happened . . . the face hugging incident.
***
Every week, I talk with aspiring nonfiction authors about their wishes for a book—whether it’s just an idea or is partly or fully drafted. If you’d like to talk about your book, I’d love to hear from you.