Skilled Wisher AI Delegation
Define the wish, connect it to action, and retain human responsibility
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 91%
Kozyrkov uses the genie story to move attention away from the apparent power of AI and toward the quality of the person making the wish. The mechanism begins with human intent: define the outcome before asking a system to produce information or act. The request must connect information to a decision rather than generating data with no operational consequence. After delegation, a human reviews whether the result serves the original outcome and remains responsible for what follows. AI can increase reach and speed, but it does not remove the need for judgment. The framework therefore treats clear intent, action linkage, review, and accountability as one decision system.
Origin
Kozyrkov invokes genie stories to argue that the central problem is the unskilled wisher rather than the powerful tool.
Core principles
- 01The hardest part is knowing what you want
- 02AI amplifies the quality of the human wish
- 03Information matters only when it connects to action
- 04Humans remain responsible for delegated decisions
How to run it
- 1
Define the wish
Name the outcome the decision should create before selecting data, prompts, or tools. Make the desired change concrete enough to judge later.
Pro tip Describe what should be different after the decision, not merely what output the AI should produce.
Watch out A precise prompt cannot rescue an undefined objective.
- 2
Connect information to action
Specify which decision or action the requested information will inform. Remove analysis that has no plausible effect on what anyone does.
Pro tip Complete the sentence: If the answer is X, we will do Y.
Watch out More data is not automatically more decision value.
- 3
Delegate deliberately
Give the AI a bounded task that serves the defined outcome. Preserve the human judgment needed where consequences or preferences matter.
Pro tip State the boundary between machine execution and human choice.
Watch out Do not mistake technical autonomy for freedom from human responsibility.
- 4
Review and own the result
Check the result against the original wish and the action it is meant to support. Keep responsibility with the human or organization that chose to delegate.
Pro tip Name the accountable person before the system is used.
Watch out Blaming the tool after deployment does not undo the human decision to use it.
In the wild
A support leader first defines the desired outcome as routing urgent safety issues to a human within five minutes. The AI receives that bounded classification task, while the leader sets escalation rules, reviews misses, and remains accountable for the service outcome.
→ The automation serves a measurable decision instead of merely producing labels.
Common mistakes
Starting with the tool
Choosing an AI system before defining the desired outcome turns capability into aimless activity.
Collecting actionless information
An answer that cannot change a decision or action creates activity without decision value.
Outsourcing responsibility
Calling a system autonomous does not erase the responsibility of the people who chose and deployed it.
Is it for you?
Best for
It is best for leaders delegating consequential analysis, recommendations, or actions to AI systems.
Not ideal for
It is not ideal for trivial automation where the desired output and acceptable failure modes are already unambiguous.
From the transcript
“Every single story is never about the genie. It is about the unskilled wisher.”
“AI not autonomous in the sense that humans can wash their hands of responsibility for it.”
From the episode
Former Google Chief Decision Scientist Cassie Kozyrkov on AI, Decisions, and Human Responsibility