What does it mean to come up with an idea?
I keep returning to this because ideation seems to occupy so much of what I do, and yet I struggle to describe the process. I can ask an agent for ideas. But if I want it to work independently over a longer period, what exactly am I asking it to do?
By a primitive, I mean a basic operation that other processes can build on. If we could describe ideation at that level, I think it could become part of how we design autonomous agents.
What is the thing being produced?
I saved a passage from signull’s “design collisions” about ideas emerging through contact between a person and the world at a particular time. That made sense to me. It also sent me towards a different question: can we describe the idea itself?
Is an idea a unit of value? Does one idea equal another idea? Is it closer to a vector, with a direction and a magnitude? When I ask about its shape, I’m trying to find a way to articulate what I’m handling.
I don’t know which of those descriptions is useful yet. Calling an idea a vector doesn’t tell me what its dimensions would be. Calling it a unit doesn’t tell me what I’m counting.
But if we could get more precise, we could start asking what inputs are required and what happens between those inputs and the idea.
What could an agent do with that?
The part I’m interested in is whether we can make that process available to an agent. Instead of relying on a new prompt each time, could it use what it knows, encounter a problem, and develop a possible direction from there?
That would require us to understand more than how to generate a list. We’d need to articulate what it means to recognize and develop an idea, and how to decide whether it is worth pursuing.
I think that could be foundational for agents that work for long periods. But I haven’t defined the primitive. I’m still at the question that comes before it: what, exactly, is an idea?