The Question Behind the Question
"Can AI make decisions?" sounds like a technical question. It isn't. It's a question about accountability, and the honest answer is no — not the decisions that matter, not in the way that word actually means something.
AI can generate an option. It can rank alternatives, predict an outcome, draft a recommendation with real reasoning behind it. What it cannot do is own what happens if the option is wrong. There's no consequence that lands on a model. The consequence lands on the person who acted on its answer, which means that person was always the one deciding — whether they noticed it or not.
The leaders who get into trouble aren't the ones who use AI to inform a decision. They're the ones who stop noticing where the AI's contribution ends and their own accountability was supposed to begin.
What AI Actually Contributes
Give AI credit for what it's genuinely good at. It can surface an option a tired team wouldn't have thought of. It can process more information faster than a person reviewing the same material manually. It can offer a counterargument to the room's default assumption, the way a good outside voice does.
That's a real contribution to a decision. It is not the decision. The distinction matters because of what happens next: someone still has to weigh that option against what the model can't see — the politics of the room, the history with this client, the fact that the numbers look right but the timing is terrible for reasons no dataset captures.
Problem = Solution, Applied to an AI's Unexpected Answer
Save the Titanic teaches a specific reversal: the thing that looks like the problem is often where the solution actually lives, if the team stops treating it as a threat long enough to look. In the experience, it's the iceberg — the thing that caused the disaster turns out to be the one thing not sinking, and it can save everyone, if the team can see it that way.
The same instinct applies to an AI answer that doesn't match what the team expected. The reflex is to dismiss it — "that's not right" — the same reflex that kills a good idea from a colleague before anyone's explored what's inside it. A team that's practiced holding a surprising answer long enough to ask "what's useful here" gets more value from AI than a team that either rejects every surprising answer or accepts every confident one.
Neither extreme is judgment. Judgment is the practiced middle: hold the answer, ask why, decide.
What Leaders Need to Practice First
Before a team can use AI well in a real decision, they need practiced answers to three questions — not memorized answers, practiced ones, built under real pressure:
What am I willing to let AI inform, and what am I not? A team that's never drawn this line draws it badly, under pressure, the first time it actually matters.
What does this answer assume, and does the assumption hold? This is Root Cause Analysis pointed at the AI's reasoning instead of a crisis's cause — see how to verify AI output before you trust it for the full discipline.
Who owns this call if it's wrong? Not the AI. A named person, every time — the same accountability discipline that already separates high-performing teams from ones that quietly diffuse responsibility whenever a decision goes sideways.
Practiced Judgment, Not Read-About Judgment
You can read all three of those questions in an article and still freeze the first time a real, fast-moving decision has an AI-generated option sitting in front of your team. The judgment only becomes real under pressure, with something at stake, and a debrief that shows the team exactly where they leaned on the answer instead of making the call themselves.
That's what Save the Titanic builds for AI-era decision making — 2,200 lives, a real clock, and the same Root Cause and Problem = Solution disciplines that decide whether your team uses AI as a resource or defers to it as an authority. ArcelorMittal's 710 leaders decided 30 to 40% faster after the experience, not because the decisions got easier, but because the judgment got practiced.
Read next: How to Verify AI Output Before You Trust It