Decision Making in the Age of AI

In March 2023, a financial advisor in New York started using ChatGPT to help clients make retirement planning decisions. Within two months, he stopped. Not because the AI gave bad answers, but because clients started treating AI suggestions as decisions rather than inputs. They stopped thinking.

"They'd come in and say, 'ChatGPT told me to put 60% in index funds,'" he told me. "They had no idea why. They couldn't explain the reasoning. They just wanted me to confirm what the machine said."

This story captures the central tension of decision-making in the age of AI: these tools can dramatically enhance human judgment, but only if you use them as thinking partners rather than thinking replacements.

What AI Does Well in Decision-Making

Let's be specific about where AI genuinely helps.

Pattern recognition across large datasets. If you're trying to find trends in ten years of financial data, market reports, or customer feedback, AI processes volume that's impossible for humans. A venture capitalist I know uses AI to screen 500 pitch decks per week, flagging the 30 that match patterns of previously successful investments. He still reads those 30 himself. The AI handles volume; he handles judgment.

Reducing information asymmetry. Before making a major decision -- buying a house, choosing a medical treatment, selecting a vendor -- AI can rapidly compile relevant information from dozens of sources. This doesn't replace expertise, but it levels the playing field. A first-time homebuyer using AI to research neighborhood data, comparable sales, and inspection red flags enters the negotiation significantly better informed.

Stress-testing assumptions. One of the best uses of AI in decision-making is as a devil's advocate. You can present your reasoning for a decision and ask the AI to find flaws, identify blind spots, and generate counterarguments. This is particularly valuable for solo founders and independent professionals who lack a sounding board.

Generating options you hadn't considered. AI excels at divergent thinking. When you're stuck between Option A and Option B, AI can often suggest Options C through F that you hadn't considered. This breaks binary thinking and expands the decision space.

What AI Does Poorly in Decision-Making

Values-based decisions. Should you take the higher-paying job or the more meaningful one? Should you move closer to family or stay in the city you love? AI can list pros and cons, but it cannot weigh them according to your values. Only you know what matters to you, and that knowledge isn't in any training dataset.

Decisions involving trust and relationships. Hiring, partnerships, and personal relationships involve reading intentions, assessing character, and making bets on people. AI can analyze resumes, but it can't tell you whether someone will be loyal under pressure.

Novel situations without historical precedent. AI learns from patterns in past data. When you're facing a truly unprecedented situation -- a new market, a personal crisis unlike anything you've experienced, a once-in-a-lifetime opportunity -- historical patterns may be misleading. The AI will confidently extrapolate from data that doesn't apply.

Decisions where the framing matters more than the data. The most important part of many decisions isn't choosing between options -- it's figuring out what question you're actually trying to answer. AI takes your framing as given. If you're asking the wrong question, AI will efficiently give you the wrong answer.

The AI-Augmented Decision Framework

Here's a practical framework for integrating AI into your decision process without surrendering your judgment.

Step 1: Define the Decision (Human Only)

What are you actually deciding? What are your constraints? What are your values? Do this yourself. AI can help you brainstorm criteria, but the final framing must be yours.

Step 2: Gather Information (AI-Assisted)

Use AI to research, compile data, and summarize relevant information. Ask it to find perspectives you might be missing. But verify critical facts independently -- AI hallucinates, and a decision based on fabricated data is worse than a decision based on incomplete data.

Step 3: Generate Options (AI + Human)

Ask AI to generate a wide range of options. Then add your own. The combination of AI's breadth and your contextual knowledge produces the richest option set.

Step 4: Evaluate Options (Human-Led, AI-Assisted)

Use AI to model outcomes, identify risks, and stress-test your reasoning. But make the evaluation yourself. You can use structured decision frameworks -- platforms like KeepRule offer prompt-based tools that help you systematically evaluate options through proven decision principles, combining AI capabilities with human judgment.

Step 5: Decide (Human Only)

The decision is yours. Not the AI's. Not a committee's. Yours. Own it.

Step 6: Review (AI-Assisted)

After the outcome, use AI to help analyze what happened and why. Feed back the results to improve your future decision process.

The Automation Trap

The biggest risk of AI in decision-making isn't that it gives bad advice. It's that it atrophies your decision-making muscle.

Every time you outsource a decision to AI without engaging your own judgment, you get slightly worse at deciding. Decision-making is a skill. Skills degrade without practice.

I've watched this happen with GPS navigation. Twenty years ago, most people had reasonable spatial awareness and could navigate with a paper map. Today, many can't drive to a familiar location without Google Maps. The skill atrophied because the tool made it unnecessary.

The same will happen with decision-making if you let it. Use AI as a bicycle for the mind, not a wheelchair.

Prompt Engineering for Better Decisions

If you're going to use AI for decision support, the quality of your prompts determines the quality of the output. Here are prompts that actually help:

• "I'm deciding between X and Y. What am I not considering?"
• "What are the second-order consequences of choosing X?"
• "Argue against my current position: [your position]"
• "What would [specific expert/framework] say about this decision?"
• "What's the worst realistic outcome of each option?"

Bad prompts: "What should I do?" "What's the best option?" These invite the AI to decide for you, which defeats the purpose.

The Future of Human + AI Decisions

The most effective decision-makers in the next decade won't be the ones with the best AI tools. They'll be the ones who best understand the boundary between what AI should handle and what humans must handle.

AI handles data, patterns, volume, and speed. Humans handle values, meaning, relationships, and novel judgment. The combination is more powerful than either alone -- but only if you maintain the boundary.

Your decisions define your life. Don't outsource that to a language model. Use the language model to decide better, then decide yourself.