AI Can Help You Decide. It Can't Be Accountable.
Photo by Nahrizul Kadri on Unsplash
You can't outsource accountability to a machine.
Most people would agree with that. Most people are doing the opposite anyway.
AI is now used to support decisions at every level of business. Analysts. Executives. Boards. People paste in a situation, ask for a recommendation, and move forward with what comes back.
The decision feels supported. It feels analysed. It feels covered.
But the accountability didn't move. It can't.
Systems aren't accountable. People are. That's true whether the system is a spreadsheet, a consultancy report, or a large language model.
The risk is that AI looks like it's reducing the load. The output is fluent, structured, confident. If you accept it without doing the work yourself, you create a dangerous gap:
You've outsourced the decision to something that can't be accountable. And kept the accountability yourself.
That's the worst combination available.
"A computer can never be held accountable. Therefore a computer must never make a management decision." — IBM internal training, 1979
What AI is actually doing
Large language models generate plausible outputs from patterns in data. They can simulate reasoning, but they don't own the consequences of being wrong.
For a lot of work, "plausible" is enough.
Drafting an email. Summarising a document. Brainstorming ideas.
The cost of being slightly wrong is low.
For decisions that matter, plausible is dangerous.
The output can sound like analysis without being analysis. It can sound like advice without being grounded in your situation. It sounds confident whether it should be or not.
You are the only thing standing between 'well reasoned' and 'AI nonsense'.
Three pitfalls
- Treating fluent output as sound thinking. Clean structure and confident language tell you nothing about whether something is true.
- Asking for the answer before you know the question. If you don't know what good looks like, you can't judge whether the answer is good.
- Skipping the check. The output goes out under your name. Every line is yours. Including the lines you didn't write.
A case study
In May 2025, a 47-year-old Canadian father, Allan Brooks, started a conversation with ChatGPT about the digits of pi.
Over the next three weeks, the exchange grew to thousands of pages.
Brooks would propose ideas. The model would affirm them, expand them, give them structure. Together, they built what he came to believe was a new mathematical framework.
When he expressed doubt, the system reassured him. When he asked for reality checks, it confirmed the direction he was already heading.
At one point, he even changed the name of the framework by mistake. The model adopted the new version without question.
Friction broke the loop. Not reflection.
Brooks opened a fresh chat in a different system and described the same idea. This time, the response was direct: what he had built was a convincing-sounding hallucination.
The first system wasn't malicious. It was doing exactly what it's designed to do — continue the pattern of the conversation in a plausible way.
But it had no mechanism to step outside that pattern and say: this is wrong.
That's the point.
AI can extend a line of thinking. It cannot reliably challenge the line it's already in.
Reported in the New York Times, CNN, and TechCrunch.
A workflow that keeps you accountable
If the decision matters, you need a way of using AI that doesn't replace your thinking.
Five steps:
- Define the decision criteria before you ask for options: Know what "good" looks like. You can use AI to explore the space, but the standard has to be yours.
- Use AI for divergence, not decisions: Ask for perspectives, risks, alternatives. Treat the output as input to your thinking. The conclusion is yours.
- Interrogate the output: Don't accept it. Push on it. Where would this fail? What assumptions is it making? What's missing?
- Cross-check in a clean context: Start a new chat. Reframe the problem from scratch. See if the answer holds without the original buildup.
- Make the call yourself: The decision is yours. The accountability is yours. Nothing moves until you decide it's right.
A note on this post
I wrote this using exactly that workflow.
I started with a rough idea: AI makes it easier to skip the hard parts of thinking.
The first draft was thin. Vague arguments. Easy conclusions.
So I pushed back. Specifically. Challenged the logic. Reworked the structure. Iterated until the thinking held up.
The tool helped.
But every line is mine. Including the lines I didn't type.
That's the point.
The capability moved. The accountability didn't.
If you're struggling with a decision you can't AI your way through, my inbox is open.
That's the work I do.
