“ it runs continuously and makes decisions on its own (within the boundaries you’ve set)”
More importantly, this setup of any other, is feeding in time externally. The agents themselves have no internal sense of time.
Pretty much all of biology has endogenous rhythms at various timescales (some bacteria and archaea, that live very short lives, may not, but even they may have their metabolism under a rhythm. Viruses definitely don’t have one). These rhythms continue to tick even when external time signals (light availability in day vs night being the big one, and tidal forces, for marine life) are removed. That is, in constant conditions, the rhythms keeps ticking.
This temporal awareness is baked into every cell in our bodies. And entirely absent in AI models. Which is why all kinds of higher level things we hear about, like consciousness, feeling, knowledge… they don’t make sense for AI. A foundational aspect of agency in biology has been nixed out and we keep ignoring this.
> They're called agents because they can act on our behalf. Which they do.
That’s not how most AI companies describe their agents. They describe them as being capable of acting on their own behalf.
I get that in computers, you can have, say, a “user agent” that deterministically provides certain information or takes some action on behalf of the user.
AI companies clearly do not mean this when they say their models are “agentic”, otherwise calling them “rogue” would make no sense. A user agent that screws up didn’t do so for its own purposes did it?