The Golden Rule asks us to treat others as we would like to be treated. But what happens when we don’t agree on what acceptable treatment looks like?
I might welcome blunt criticism. You might find the same delivery objectionable. Neither of us can simply assume our preferences should govern the other.
This is where SpatzAI introduces reciprocity.
Under the SpatzAI FAIR PLAY Charter and Playbook, every team member has the same right to verbally “Caution” another about behaviour they find objectionable, regardless of rank, and the right to escalate proportionately through the agreed SpatzAI chat app procedure.
The recipient can acknowledge, deny or ignore the caution. If unresolved, the person issuing it can escalate.
A caution is not a verdict. It’s an early opportunity to reflect and course-correct before a minor spat becomes something more serious.
Object to my behaviour as you would like me to object to yours.
After the infamous Hugging Face breach, now imagine applying that principle to autonomous AI agents.
Each agent has its own task assigned to it (that is what distinguishes it as an agent): researching, coding, analysing or managing systems. What if every task orientated agent also carried a standing SpatzAI accountability sub-task?
Through the shared SpatzAI API, agents could caution one another when potentially exceeding authorised boundaries, acknowledge or challenge cautions, and escalate unresolved objections through SpatzAI’s chat and review process.
No agent would be automatically right. Even the agent raising an objection, and the reviewers, would remain accountable.
This wouldn’t replace technical safeguards. Dangerous actions must still be interruptible immediately.
In June 2023, I proposed real-time interventions and (people) peer review of AI behaviour in an application to OpenAI. Embedding reciprocal accountability directly into every AI agent is a further development of that idea.
Perhaps AI governance shouldn’t rely exclusively on the authority or people watching agents from above.
Perhaps it should also empower agents to look out for one another.
Every agent its brother’s keeper. Accountability Before Authority.
SpatzAI — Helping Humans and AI Agents Object Well.
Spatz.ai
Leave a Reply