Future & Contribution
Using Automation Responsibly
Also called: Ethics of automation
- Personal interest
- Philosophical question
- Working interpretation
An AI agent can act fast and many times over, so its mistakes can spread and matter. I want automation that is useful but also supervised, keeps a clear record of what it did, has the right permission and stays within firm limits.
Which decisions should never be handed to a machine without a person reviewing them?
Why it attracts me
Automation multiplies whatever it is given, including mistakes. An agent that edits a hundred files or sends a thousand messages can do great work or great damage at the same speed. That makes its limits as important as its abilities.
The idea
Responsible automation rests on a few plain commitments. People review decisions with real stakes. The system keeps a record so anyone can trace what happened. The people affected have agreed to it. And the system can only do what it was clearly allowed to do. These follow from the partnership in Pairing AI's Reach With Human Judgment: the machine searches and proposes, while judgment about what matters stays with people. They also follow from Engineering That Respects How People Work, because real people live with the consequences.
An example
In my write-up on working with AI coding agents, I separate guidance from proof. An instruction to the agent is guidance. A test is evidence. A check that blocks bad output is enforcement. The same post prefers narrow tools that do one small job over a tool that can run anything. A Voice Before Words, a communication project on this site, applies the same idea to language: the software offers possible sentences, but the person chooses what is said.
What I think (and don't know)
I think the reflective question has no fixed answer, but asking it before automating anything is the real safeguard. What I don't know is how to keep human review meaningful when the volume grows and approval becomes a reflex.
Questions I'm still exploring
- Where should an agent be allowed to act, and where only to suggest?
- How do I keep a record of automated decisions that a person can actually follow?
- Who should give consent when automation affects people who never chose it?
Sources and further reading
- NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023 — A voluntary US framework for managing the risks of AI systems.
- Raja Parasuraman, Thomas B. Sheridan and Christopher D. Wickens, "A Model for Types and Levels of Human Interaction with Automation", IEEE Transactions on Systems, Man, and Cybernetics (2000) — A widely cited way to decide how much of a task a machine should take over.
Working interpretation: drafted from my notes and interests for review. It is not a direct quotation, and I may still change it.