Chatting with AI Led Me to a Philosopher
Lately, while chatting with AI, I find myself asking a lot of philosophical questions. Digging into "Why isn't it doing what I intended?" led me to reflect on my own thinking and communication habits—and in that process, I met a philosopher named Grice.
AI doesn't automatically share "unspoken premises."
"You have pretty handwriting."
How do you feel when you hear this? It sounds like a compliment, but depending on the context, there might be a hidden barb: "You don't really have any other strengths..." We naturally pick up on these implications depending on the situation. For AI, that "depending on the situation" part might require context engineering.
Lately, I tend to use Claude Code and Codex together.
Finding a needle in the haystack of the web/code via a subagent,
Summarizing the context well and consulting Codex through a skill,
Taking that answer and having agents debate each other to quickly arrive
at the right direction—I've been experimenting with this approach.
The AskUserQuestion tool prompts Claude Code to ask questions. By explicitly including it, the AI sometimes offered new perspectives, and work got done while polishing up parts I had missed.
So I was going to put this rule right into CLAUDE.md, but while editing, I asked it this question:
Do we really need a guideline that says "don't ask questions when things are clear"? What do professional linguists and philosophers think about this?
I discovered that a philosophical concept aligned with this direction of my thoughts is Grice's Cooperative Principle.
Grice studied the gap between what is said and what is meant.
When we converse, we implicitly fill in each other's blanks.
Saying only as much as needed,
Saying only what is relevant,
Saying it in a way the other person can understand.
Between humans, shared experiences, organizational culture, and previous conversations build a "common ground." With AI, this ground is thin. So the AI fills in the blanks its own way, and we're left thinking, "Why is it giving such an irrelevant answer?"
Context engineering is the habit of making things explicit upfront, rather than expecting implicit meaning from the AI.
Acknowledging that we can't ask perfect questions, the key might be "pulling the premises out of my head to close the gap."
Philosophy is rising to the surface
Interestingly, it seems I'm not the only one thinking about this approach.
https://www.youtube.com/watch?v=I9aGC6Ui3eE
The video talks about how there's a growing trend to connect the domains LLMs understand with philosophical and principled instructions.
It feels like philosophy is becoming an increasingly practical tool in the AI era.