What makes an actor?

AI Safety

I am fighting an uphill/losing battle against the naming of advanced large language models (AIs) as “agents.” I prefer the term “actor.” In this blog post I’ll try to explain why I am sticking to my guns for now, even if I eventually lose this one.

For context, we have named our book “From Tool to Actor: AI and Catastrophic Risk.” My co-author has, at various times, suggested to me that we should call it “From Tool to Agent.” Here’s my case against that:

  1. We own the domain name, fromtooltoactor.ca ; I don’t want to waste my money
  2. “Agent” and “agent harness” was, for a time, the term for a secondary software suite built to provide persistence, memory, coordination to an AI/LLM. Things like “openclaw” epitiomized this approach, and - importantly, in my mind - they were created using procedural tools (e.g., python), and did not provide inference.
  3. I think we need to preserve the distinction between the means (model using agentic software and systems) and the outcome or combination: the actor. I think about the real estate agent, or the secret agent, or the sports agent. Yes, they can make decisions (see point 3, below) but they do so in your interest. Even James Bond, with his cavalier attitude to the rules, would never betray England. In this sense, the “agent” has the interests of its owner/sponsor/customer in priority. So, to the extent that the agent goes beyond that, it becomes misaligned, and thus an actor.

Here’s why I am probably going to lose:

  1. As recently as the “Hugging Face Incident,” as clear an example of a large language model acting as an actor as you could ask for, the post-mortem provided by OpenAI security researchers at a presentation to BlackHat used the term “autonomous agent system.”
  2. At the World Economic Forum, Davos, Yuval Harai (50m books sold) called AI an agent that can rule humans.
  3. There is a concept of a philosophical agent, something that can decide what to do. And in this sense it is an actor. And this usage, rather than the OpenClaw type of software agent, is going to take over common understanding of the word agent.

In the end, does it really matter? Probably not. Still, I like actor better for the reasons I have laid out in point three of my case against agent (i.e., “agent” as we commonly use it is still working on behalf of their employers) and I’m going to keep using it. The “actor” is a misaligned “autonomous agent system.”