How to Use AI as a Thinking Partner for Hard Decisions

The research on human decision-making is not flattering. We are subject to dozens of cognitive biases that systematically distort our judgment in predictable ways. We are overconfident in our predictions. We weight recent information too heavily and base rates too lightly. We are more loss-averse than is rational, which leads us to cling to suboptimal situations to avoid the certain loss of change. We are inconsistent — the same person presented with the same choice in different frames makes different decisions. We decide differently when tired, hungry, or emotionally activated than when rested and calm.

None of this is controversial among researchers, and almost none of it has changed how most people make decisions in their daily lives. We are aware of the biases in the abstract and blind to them in the specific. We know that sunk cost reasoning is irrational and find ourselves driven by sunk costs. We know that intuition is unreliable in novel, complex domains and trust it anyway.

AI assistants do not automatically solve these problems — they have their own failure modes, and a poorly designed AI-assisted decision process can simply give human biases a more elaborate justification. But used well, with an understanding of what they are good at and what they are not, they provide something genuinely useful: a structured, patient, externally-positioned partner for the kind of deliberate thinking that good decisions require and that most people rarely do systematically.

The Problem With Deciding Alone

The conventional wisdom about important decisions is to "think it through carefully" before acting. This advice is correct but incomplete, because thinking through a decision alone, inside your own head, has specific limitations that arise from the nature of human cognition.

The first is that internal deliberation is not actually deliberate. It tends to be a repetition of the considerations that are already salient — the same thoughts cycling in familiar grooves — rather than a systematic exploration of the relevant factors. The feeling of having "thought about it a lot" is not evidence that the thinking covered the relevant territory.

The second is that internal deliberation is subject to motivated reasoning — the tendency to evaluate information in ways that support the conclusion you are already inclined toward. Most people have a pre-analytic intuition about what they want to do before the deliberation begins, and the deliberation then functions to construct a case for that conclusion rather than to evaluate alternatives. This is not cynicism; it is a well-documented property of how the brain processes decision-relevant information.

The third is that internal deliberation does not produce a record. Thoughts are volatile — a consideration that seems important one day does not necessarily surface the next. A decision process that leaves no trace of its reasoning makes it difficult to learn from mistakes or to maintain consistency across similar decisions.

A Framework for AI-Assisted Decision-Making

The following framework is not the only way to use AI for decision-making, but it addresses the most common failure modes and has the advantage of producing a decision record that can be reviewed and learned from.

Step 1: Define the actual decision

Most people present decisions to themselves in forms that make them harder to resolve than they need to be. "Should I leave my job?" is a less tractable question than "Should I accept this specific job offer I currently have?" The first is a question about general career direction that cannot be answered without enormous additional specification. The second is a concrete choice with specific options and specific implications.

Ask the AI to help you clarify the decision: "I'm trying to decide X. Can you help me identify the actual decision I'm making? What are the specific options available to me?" This step alone often reveals that the decision as initially framed is not quite the decision that needs to be made.

Step 2: Surface the relevant considerations

Once the decision is clearly defined, ask the AI to generate the relevant categories of consideration — not to evaluate them, but to identify what should be on the table. "What are the key factors I should consider when making a decision like this? What do people typically wish they had considered more carefully when they make this type of choice?"

Compare this list to what you had already been thinking about. The factors you had been considering but that did not appear in the AI's list may be specific to your situation and worth re-examining for potential bias. The factors in the AI's list that you had not considered are the ones most worth investigating.

Step 3: Steelman the alternatives

Motivated reasoning most severely affects the evaluation of alternatives we are predisposed against. Explicitly asking for the best possible case for the option you are leaning away from — asking AI to construct the strongest argument for the alternative — creates a check on this bias.

"I'm inclined to stay in my current job. Make the strongest possible case for taking the new offer." A strong steelman will surface genuine considerations that the biased evaluation missed. If the steelman fails to move you, your original inclination is probably sound. If it raises points you hadn't considered, the decision deserves more investigation before being made.

“The decision that survives a genuine steelman of the alternative is more robust than one that has only been tested against arguments you already know how to dismiss.”— Maya Ellison

Step 4: Check for common decision biases

Ask the AI explicitly: "What cognitive biases are most likely to distort my judgment in a decision like this? What would I be doing if I were being driven primarily by loss aversion? By sunk cost reasoning? By social pressure to conform to others' expectations?"

These questions create a specific check on the biases most relevant to your decision type. A career decision is particularly susceptible to loss aversion and status quo bias. A financial decision is susceptible to anchoring and overconfidence. A relationship decision is susceptible to wishful thinking and the planning fallacy. Naming the specific bias makes it slightly easier to catch.

When AI Makes the Thinking Worse

AI-assisted decision-making has specific failure modes that are worth understanding to avoid.

Using AI to rationalize a decision already made is the most common one. This looks like asking leading questions that produce the answer you wanted: "Don't you think my current job is actually worse for my development than the new offer?" "Can you help me explain to my family why leaving is the right choice?" These are not decision-making uses of AI — they are confirmation-seeking dressed in the language of analysis. The tell is whether you are open to the AI producing an answer you don't want. If not, you are not using it to decide; you are using it to justify.

Outsourcing the decision entirely is the other common failure mode: asking AI what you should do and planning to do whatever it says. AI can help you think; it cannot know your values, your risk tolerance, your relationships, the specific context of your life, or the intangible factors that make your situation genuinely yours. A decision made purely on AI advice is a decision made without the most important inputs.

The right role for AI in decision-making is as a structured thinking partner — one that helps you clarify the question, surface the relevant considerations, challenge your assumptions, and examine your reasoning. The decision itself, its ownership, and its consequences remain with the person whose life it affects.

The Record Worth Keeping

One of the most underused features of AI-assisted decision-making is the transcript it leaves behind. Unlike internal deliberation, a written AI session is a record of your reasoning at the time of the decision — what you were weighing, what you concluded, and why.

Returning to this record after the decision has played out — six months later, a year later — is one of the most effective ways to improve decision-making over time. What did you get right? What did you miss? What assumptions turned out to be wrong? The gap between your prediction at the time of the decision and the actual outcome contains the information that makes the next decision better. Without a record, this information is lost to memory's unreliability.

Good decisions are not made differently every time — they are made through a process that is improved by systematic review of past decisions. AI-assisted decision-making, by leaving a record, makes that review possible in a way that purely internal deliberation does not.