Writing · AI and human responsibility
The Fixed Point in Human-AI Work
By Richard K. Marshall · Originally published
The next decade of human-AI work won't be decided by who has the largest model.
It will be decided by who keeps a fixed point when tools get faster, cheaper and more persuasive.
That fixed point is the Marshall Principle:
Artificial intelligence may assist human decision-making, but responsibility always remains with humans. Authority cannot be automated.
Not optional. Not "when the model is confident." Non-delegable.
Why a fixed point matters
Every few months the stack changes. New agents. New "autonomous" workflows. New demos where nobody is in the screenshot.
Without a constitutional rule, organizations drift into a quiet lie: "The system decided."
That sentence is useful for blame and useless for trust.
Customers, employees, regulators and your own sleep at 3 a.m. all need the same answer: who owns this?
If you can't name a human, you don't have governance. You have a vibe.
What the principle allows
The Marshall Principle isn't anti-AI.
It allows:
- Research help
- Drafting
- Diagnosis
- Pattern spotting
- Boring automation inside clear rails
It refuses:
- Orphan authority
- "The model said so" as a decision
- Shipping material actions just because the tool could
Capability is welcome. Unowned power is not.
Applied to the future of human-AI work
1. The workplace agent. Your agent can file tickets, draft emails and update the CRM. Fine. When it speaks as the company, spends money or changes a permanent record, a named human still owns the outcome.
2. The customer front door. People already ask machines who you are. AI can help you see that picture. It can't become the owner of your reputation. You still fix the public file. You still answer the angry email.
3. The care setting. Summaries and checklists assist. Duty of care doesn't transfer to a vendor's model card.
4. The board packet. AI can brief. Directors still deliberate. The minutes still need a human secretary of truth.
A one-page test
Before you deploy a new AI workflow, ask:
- Who is the accountable human?
- What may run unattended?
- What requires explicit human authority?
- How do we verify the outcome, not just the API status?
- How do we shut it off?
If any answer is "the system," rewrite the design.
The fixed point doesn't move
Models will improve. Prices will fall. Hype will rotate.
The principle stays: AI may assist. Responsibility stays human.
— Richard K. Marshall Marshall Intelligence · Lexington, Kentucky
The full public policy is free to read at marshall.net/ai-governance.html.
Originally published on X: https://x.com/RichMarshall/status/2088840197882478737 · . Refreshed .