For the past 18 months, the message from researchers was consistent: the more you lean on AI, the weaker your thinking gets. A 2025 survey of 666 participants found a significant negative correlation between AI tool use and critical thinking scores. MIT ran EEG brain scans and found that ChatGPT users showed the weakest neural engagement of any group — and when the AI was taken away, they couldn’t even recall their own earlier work. The researchers called it “cognitive debt.”
That research was real. But it was measuring the wrong condition.
The New Finding
In March 2026, Wang and Zhang published a study in the International Journal of Educational Technology in Higher Education. They studied 912 students across China, Europe, and the United States using a three-wave time-lagged design — measuring partnership orientation with AI first, then cognitive strategies two weeks later, then learning outcomes two weeks after that.
What they found flipped the narrative. When students treated AI as an intellectual collaborator rather than just a search engine, two things happened simultaneously: they became more critical of AI outputs, and they delegated more substantive work to AI. Both independently predicted deeper, transformative learning — the kind where students questioned long-held assumptions and fundamentally changed how they understood their subjects.
The paradox: the same mindset that made students delegate more also made them more critical. And both made them learn more deeply.
The Three Zones
The relationship between AI offloading and learning depth wasn’t a straight line. It was a U-curve. Wang and Zhang identified three zones.
Zone 1 — No offloading. The learner does everything manually. Full cognitive load, every step. Learning happens, but it’s slow and capacity-constrained.
Zone 2 — Scattered offloading. This is where most CPAs and finance professionals using AI actually live. Fixing a sentence here. Checking one fact there. Tidying a paragraph. According to the research, this is the worst zone. You’re still carrying almost all of the cognitive load, but now you’ve added the friction of managing the AI. More effort, no meaningful benefit. Zone 2 is worse than not using AI at all.
Zone 3 — Strategic offloading. The learner delegates entire categories of work to AI. All the source summarization. The full first-pass literature review. The entire first draft. The cognitive savings are large enough to genuinely free capacity — and that freed capacity gets invested in the work AI can’t do: critiquing frameworks, questioning assumptions, constructing original arguments.
Think about how you’d work with a junior associate. If you only trust them with one sentence at a time, you’re still doing all the work — and micromanaging them too. But if you say “take this client situation and give me a full first-pass memo,” and you step back and use that hour to think about strategy, that’s Zone 3. You’ve freed enough capacity to do something you couldn’t have done otherwise.
What This Means for CPAs
Two practical shifts.
First, audit your current AI use. Ask yourself honestly: am I in Zone 2? If you’re using AI to fix sentences, check individual facts, or generate one paragraph at a time, the research says you’d be cognitively better off not using AI for those tasks at all. Instead, think in categories. Can you hand AI an entire set of client documents and ask for a comprehensive summary? Can you ask it to draft the entire first version of a memo — with the understanding that you’re going to restructure and question it? Those are Zone 3 tasks. That’s where the benefit is.
Second, the framing matters as much as the usage. Treating AI as a partner — not just a tool — changes which cognitive pathways activate. When you see AI as a collaborator, you automatically become both more critical of its outputs and more strategic about what you delegate. The instinct to “just use AI a little” is the wrong instinct if your goal is to think more clearly or learn more deeply. Commit to it, or don’t use it.
One more thing: keep at least one part of your workflow fully AI-free. If you never evaluate your own thinking without a scaffold, you can’t tell whether you’ve built capability or just built dependency. Know which one you’re measuring.
Key Takeaways
- The prior “AI erodes thinking” research was measuring Zone 2 use — scattered, small assists that add overhead without freeing cognitive capacity.
- Zone 3 use — delegating entire categories of work — produces better critical thinking and deeper learning.
- Offload substantially, or not at all. Half-measures are worse than no AI.
- Frame AI as a collaborator. That framing activates both critical evaluation and strategic delegation simultaneously.
- Preserve one AI-free output. If you can’t perform without the scaffold, you’ve built dependency, not capability.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply