AI and the democratisation of intellectual sparring. A position paper, April 2026 — the argument here in small pieces, the full text free on SSRN.
Position paper · April 2026 · free and open access on SSRN
The thesis
“What matters most about AI is not what it produces. It is what it does to how we think.”
Drawing on Ericsson's deliberate-practice framework, the paper argues that engaging with AI in a structured, intentional way works as a training ground for reasoning. AI does not just hand us information. It gives us something that used to be scarce and expensive: a sparring partner who will challenge our ideas whenever we want, for as long as we want, with no gatekeeping.
The practice
The benefit is entirely conditional on how you engage. The same tool sharpens one person's reasoning and dulls another's.
Formulate a precise question. Evaluate what comes back. Find the gaps, push back, refine — and arrive somewhere stronger. Each loop is one repetition of deliberate practice in reasoning.
Sharpens — you are doing the workAsk, accept the first answer, stop. No loop, no practice, no growth. Cognitive offloading erodes the very capacity needed to evaluate the answers — a 2025 study of 666 users found frequent AI use strongly negatively correlated with critical-thinking scores (r = −0.68).
Dulls — it is doing the work for youThe bottleneck
Books have been widely available for centuries; the internet put nearly all recorded knowledge within reach. Critical thinking did not equalise to match. What was missing was interactive intellectual challenge — someone who would engage with your thinking, push back, and expose the weak spots. And that was rationed:
“The bottleneck was never information. It was the sparring partner.”
AI is the first universally accessible capable interlocutor: any time, any domain, unlimited patience, no gatekeeping. A motivated teenager at an underfunded school, with nothing but a phone, now has access to something that approximates the engagement previously reserved for the privileged. The caveat, and it matters: the tool is democratised — the disposition to use it well is not.
The diagnosis
Once AI becomes widely available, one shared trajectory splits into three.
Upper arm
Question, refine, debate, push back. Better thinking leads to better questions leads to harder challenges. More reps of genuine sparring per day than any seminar ever offered — and the compounding shows.
Middle line
Keep developing at whatever rate their environment provides. Not declining — being outpaced. Relative stagnation: the ceiling rises for everyone who engages deliberately.
Lower arm
Offloading breeds dependence breeds weaker thinking. They may not even notice: AI output sounds fluent whether or not the person reading it can still tell good from bad.
The gap that changes the policy argument is not between the upper and lower arms. It is between the upper arm and the middle line: if deliberate engagement genuinely accelerates cognitive development, staying out is not staying still.
“Non-adoption is not neutrality. It is falling behind.”
The toolkit
Philosophy had prompt engineering figured out 2,400 years ago. Five techniques, drawn from Western and Eastern traditions, ordered as the natural arc of a serious enquiry. Philosophy provides the method; AI provides the repetitions.
01 · Setup
Confucian — the exchange begins before you write your prompt
Inquiry begins with precise naming. Check your own terms before you ask: sloppy language produces sloppy output, and sloppy output trains a sloppy thinker.
“What exactly do I mean by this term? Help me define it precisely.”
“What does this concept include — and what does it exclude?”
02 · Probe
Socrates — once the first response arrives, the real work begins
Never let a plausible answer end the inquiry. Refusing the first response is what turns a lookup into deliberate practice.
“What assumptions am I making about this?”
“Why does that assumption matter? What falls apart if I remove it?”
03 · Verify
How do you know what you claim to know?
AI's failure mode is confident, fluent wrongness. Asking “how would I know if this were wrong?” is the single most protective habit — the skill that most directly prevents the lower arm.
“What would count as evidence for or against this claim?”
“Could this be stated with confidence and still be wrong? How?”
04 · Stress-test
Build the strongest version of the other side — then answer it
The commonest misuse of AI is asking it to confirm what you already believe. Inviting the strongest counter-case transforms the exchange, and trains the intellectual honesty that algorithmic comfort corrodes.
“Make the strongest possible case against my position. Do not soften it.”
“Argue with me as if you genuinely disagreed.”
05 · Push to limits
Follow the logic past the point where most humans stop
Take a claim to its logical extreme and see whether it produces absurdity. The extreme exposes exactly where an argument breaks — a diagnostic skill that carries over to every other reasoning context.
“If this argument is true, what else must also be true? Follow the logic all the way.”
“Push this position to its logical extreme. What breaks?”
The caveats
The paper
Throw Punches: AI and the Democratisation of Intellectual Sparring. Tham Seng Choe, Singapore, April 2026. Free and open access on SSRN.
The AI is the most accessible sparring partner in human history. The question is whether people know they are allowed to throw punches.