Thinking Without Proxies: Keep Your Judgment in an AI World

There is a particular kind of intellectual helplessness that arrives quietly. It doesn't announce itself the way writer's block does, or the way burnout does. It seeps in through convenience. One day you realize you haven't formed an opinion from scratch in weeks. You've read AI summaries instead of source material. You've accepted AI-generated analysis instead of sitting with a problem long enough to feel its shape yourself. You've outsourced not just the labor of thinking — but the thinking itself.

This is the central cognitive risk of the AI era, and it's one we're not talking about nearly enough.

We spend a great deal of time worrying about AI hallucinations, job displacement, and model bias. Those are real concerns. But there's a quieter threat that may ultimately matter more: the gradual erosion of first-person judgment in a world where AI increasingly mediates our relationship with information, decision-making, and even self-understanding.

I want to explore what that erosion looks like, why it happens, and — most importantly — how to resist it without rejecting the genuine value that AI tools provide.

The Proxy Problem: What Happens When We Let AI Think for Us

A proxy, in the cognitive sense, is anything that stands in for your own reasoning. We've always had proxies — trusted experts, authoritative institutions, respected publications. We consult them, weigh their views, and integrate them into our thinking. That's healthy epistemic behavior.

The problem isn't using proxies. The problem is substituting them for independent thought entirely.

AI systems are uniquely powerful proxies because they are extraordinarily fluent, always available, and authoritative-sounding even when wrong. A 2023 study published in Science found that people exposed to AI-generated arguments became significantly more likely to change their views — and significantly less likely to remember that they'd changed them. The persuasion was invisible.

When persuasion is invisible, it can't be interrogated. And when it can't be interrogated, it can't be resisted or refined.

This is the proxy problem in its most acute form: not that AI influences your thinking, but that it does so without leaving a trace that you can examine.

Why the Brain Defaults to AI Deference

To understand why this happens, it helps to understand something about cognitive effort. Thinking — real, effortful, first-person thinking — is metabolically expensive. The brain is a prediction machine that is constantly optimizing for efficiency. When a high-quality, fluent, confident answer is available with a single prompt, the brain's reward system registers it as a successful outcome.

There is no built-in alarm that fires when you accept an AI answer uncritically. The sense of closure you feel when you read a well-constructed AI explanation is neurologically similar to the sense of closure you'd feel if you had worked the problem out yourself.

Researchers studying "cognitive offloading" — the practice of delegating mental tasks to external tools — have found that while offloading can extend human capability, it can also reduce the quality of unaided cognition over time when practiced habitually without deliberate re-engagement. A 2021 meta-analysis in Psychological Bulletin covering 38 studies found that heavy reliance on GPS navigation correlates with measurable declines in spatial reasoning ability — a preview of what domain-specific cognitive offloading can produce at scale.

The analogy to AI-mediated reasoning is imperfect but instructive: if you never navigate, you lose the ability to orient yourself. If you never reason through uncertainty, you lose the tolerance for it — and the skill of managing it.

The Three Patterns of Cognitive Surrender

Cognitive surrender to AI doesn't usually happen all at once. It tends to follow one of three patterns:

1. The Confirmation Shortcut

You have a half-formed view, and you ask an AI to develop it. The AI, trained on vast human text, almost always finds a way to frame your intuition compellingly. You receive your own idea back, polished and footnoted. The problem is that the idea was never stress-tested. It was dressed up. This is the difference between thinking that has been validated and thinking that has merely been articulated.

2. The Summary Trap

You need to understand a complex subject quickly, so you ask for a summary. Summaries are compressions, and compressions are lossy. The texture of a real argument — its hesitations, its qualifications, its contradictions — gets smoothed out. When you rely on summaries long enough, you stop developing the tolerance for complexity that genuine expertise requires. You begin to mistake fluency for understanding.

3. The Decision Displacement

You face a hard decision with genuine moral or strategic weight. You ask an AI what to do. The AI gives you a structured answer. You follow it, more or less. The problem isn't just that the AI might be wrong. It's that the struggle of making a hard decision is where values get clarified and character gets built. Skipping that struggle doesn't make you more efficient — it makes you less yourself.

What Independent Judgment Actually Requires

Before we can protect independent judgment, we need to be clear about what it is. Independent judgment isn't stubbornness. It isn't refusing outside input. It's something more demanding and more valuable:

Independent judgment is the capacity to form, hold, revise, and act on beliefs through a process you can account for.

Notice the components. You form the belief — meaning you do the actual epistemic work. You can hold it — meaning you're willing to defend it under scrutiny. You can revise it — meaning you're genuinely open to evidence, not just performing openness. And you can account for the process — meaning you know why you believe what you believe, and that story doesn't begin with "the AI said so."

This is a high bar. It's always been a high bar. AI makes it simultaneously more achievable (we have better information access than any generation in history) and more endangered (that information arrives pre-processed, pre-argued, and pre-concluded).

A Framework for Thinking With AI Without Thinking Through AI

The goal isn't to avoid AI. It's to use it without losing yourself in it. Here's a framework I find useful — not as a rigid system, but as a set of orientations:

Stage 1: Think First, Then Verify

Before you query an AI on any substantive question, spend at least a few minutes formulating your own position. Write it down if you can. Ask yourself: What do I already believe about this? What would I need to see to change my mind? What am I uncertain about?

This doesn't take long, but it creates a baseline. Now when the AI responds, you have something to compare against. You're evaluating its reasoning rather than inheriting it.

Stage 2: Use AI as a Challenger, Not an Oracle

Reframe how you prompt. Instead of asking "What should I do about X?" ask "What are the strongest arguments against what I'm inclined to do?" Instead of "Explain Y to me," try "What do smart critics of Y argue, and how valid are those criticisms?"

This posture keeps you in the driver's seat. You are using the AI to pressure-test your thinking, not to replace it.

Stage 3: Return to Primary Sources Selectively

You can't read everything from scratch. But for the questions that matter most — the ones that shape your strategy, your values, your understanding of the world — make a habit of going back to primary sources. Read the original paper, not the summary. Read the speech, not the transcript analysis. Read the case study, not the listicle.

Primary sources are slower. That slowness is a feature. Friction creates retention. Retention enables reasoning.

Stage 4: Name Your Uncertainty

AI systems are trained to minimize the appearance of uncertainty. They hedge with phrases like "it's worth noting" and "some experts argue" — but they rarely say "I genuinely don't know, and you shouldn't trust me here." Build the habit of doing what AI doesn't do well: name your uncertainty explicitly, and treat it as data.

"I'm not sure about this" is the beginning of rigorous thinking. It's also increasingly rare.

The Difference Between Augmentation and Delegation

The distinction that matters most in this conversation is the one between augmentation and delegation.

Augmentation means using a tool to extend your capability while keeping your judgment intact. A pilot uses autopilot to reduce workload on long flights, but retains situational awareness and the ability to intervene. A surgeon uses imaging technology to enhance precision, but makes the incision herself. The tool amplifies the human; the human remains the locus of decision.

Delegation means transferring both the task and the accountability to the tool. This is fine for low-stakes tasks — spell-checking, formatting, retrieving factual information. It becomes a problem when it extends to judgment-laden tasks: strategic decisions, moral choices, interpretive questions.

Most people are somewhere in the middle of this table. The question is which direction they're drifting — and whether they're drifting deliberately.

Institutions, Not Just Individuals

The cognitive stakes here extend well beyond individuals. We are entering an era in which AI systems mediate the reasoning of organizations, governments, and civic institutions. When a hospital deploys AI diagnostic tools without building cultures of clinical challenge, the AI becomes an institutional proxy — its errors propagate across every patient encounter. When a newsroom generates AI-assisted coverage without editors who can interrogate underlying assumptions, the AI shapes the narrative before any human has thought critically about it.

A 2024 survey by the Reuters Institute found that 62% of news consumers already struggle to identify AI-generated content from human-authored reporting — a figure that underlines how quickly the mediation of institutional judgment is becoming invisible.

The same dynamic applies in boardrooms, classrooms, courtrooms, and legislative chambers. The question of how to maintain independent judgment isn't just personal development advice. It's a question about the integrity of every institution that depends on humans doing hard, effortful, accountable thinking.

Why Epistemic Courage Matters More Than Ever

There's a phrase I keep coming back to: epistemic courage. It's the willingness to form and defend a view even when it's easier not to — even when an authoritative source, human or AI, offers you an off-ramp.

Epistemic courage has always been uncommon. Social pressure, institutional conformity, and the cognitive cost of genuine inquiry have always made it easier to defer. AI adds a new and potent layer: the instant availability of a confident, well-articulated answer makes the off-ramp smoother than it has ever been.

Strengthening epistemic courage in the AI era looks like:

• Disagreeing with AI outputs in writing — not just mentally, but actually recording your objections and your reasons
• Choosing difficulty deliberately — reading long-form arguments, engaging with primary sources, following a complex thought to its conclusion rather than outsourcing the final mile
• Practicing the statement "I don't know yet" — resisting the pull of premature closure, especially when an AI has already given you an answer that feels complete
• Building slow-thinking rituals — journaling, debate, unstructured reflection time that isn't prompted by any tool

None of this requires rejecting AI. It requires deciding, consciously and repeatedly, who is doing the thinking.

The Long Game: What We're Really Protecting

Here is the deepest reason this matters: independent judgment is not just a cognitive skill. It is the basis of agency — the capacity to be a self-determining actor in your own life and in the life of your community.

A society that outsources its judgment at scale isn't just intellectually weaker. It's structurally more fragile. The ability of citizens to evaluate competing claims, resist manipulation, hold institutions accountable, and revise their own beliefs in light of evidence is the operating system of democratic self-governance. When that operating system degrades, the consequences aren't academic.

The most important AI literacy skill isn't learning how to prompt — it's learning how to think independently in a world where prompting is the default.

We are in the early stages of a multi-decade reconfiguration of the relationship between human minds and artificial intelligence. How that reconfiguration goes depends, in no small part, on whether each of us decides to remain the author of our own thinking — or whether we let that authorship quietly slip away.

Thinking without proxies isn't a refusal of the future. It's the precondition for navigating it well.

For more on the human dimensions of AI transformation, explore how AI is changing the nature of expertise and what it means to prepare for AI as a person, not just a professional on Prepare for AI.

Last updated: 2026-04-05