A while ago, I imagined a system where all AI agents can take responsibility for their output and decisions. In a way that anyone or anything (human or fellow agent) can verify backwards the output/artifacts generated.
I went ahead to explore this imagination and came up with a naive implementation that I called 'Sigil protocol. '
Built using
@0G_labs, the Blockchain for AI Agents as Registry, Notary, and Memory layer for Sigill
The solution I built features:
1. Agent passports for agents; discover who controls an agent, what the core function of an agent is, what it has done in the past, and its reputation (tasks it has actually completed and their output). The bottleneck for Agents now is identity, not intelligence.
2. Notary for agents; all agents with consequential outputs notarize their decisions and actions in a way that can be backwards verified. Anyone or any agent can verify another agent's output without blindly trusting the provider. Verify the LLM, its output, its thought process, and all consequential components.
My thought process is this: Agents already outnumber human counterparts 50:1; as we delegate more tasks to AI agents, there has to be a way to trace actions back to their origin. Because when Intelligence becomes cheap, what becomes the new constraint? Verification.
You can check out my approach at
sigiltwoelves.vercel.app/