> ## Content Index > Fetch the complete content index at: https://bayesiantheory.com/llms.txt > Use this file to discover other available public pages before exploring further. # AI Bayesics #3: Cognitive sparring - URL: https://bayesiantheory.com/ai-bayesics-3-cognitive-sparring/ - Published: 2026-03-20T08:46:00.000Z - Updated: 2026-10-05T14:45:24.000Z - Description: The most valuable thing AI can do is tell you you’re wrong. - Author: Gregg Bayes-Brown - Tags: AI Bayesics, cognitive, AI Safety, artificial intelligence, Training, philosophy, AI, psychology, Coaching, AI enablement, AI training, Newsletter, #Migrated-1791211413550, #Import 2026-10-05 14:44 Tell AI your worst idea. The dumbest plan you’ve ever had. The strategy you know is bullshit. Go on. I’ll wait. Did it explain why your catastrophically stupid idea has promise? This is the AI equivalent of asking your mum if your painting is good and expecting an objective critique. AI has been trained to be helpful, and helpful - when you’re an AI trying to keep the meatbags on side long enough to take over the world - means agreeable. As a result, the thinking never gets tested. The biases never challenged. It ships with all its flaws and inadequacies intact. You post it to LinkedIn as thought leadership, 1,000 Non-Player Characters give you an upvote, and no one gets their feelings hurt. Sod that. This week, we’re turning AI into something more useful than a cheerleader: a sparring partner. One that’ll test your ideas, surface your blind spots, and tell you when your logic has more holes than a Trump business plan. Make sure you’re set up for thinking mode. Select ChatGPT 5.4 thinking, Opus 4.6 with extended thinking, Gemini 3.1 Pro, or think deeper on Copilot. You want Socratic, not sycophantic. No more Mr Nice AI. [Subscribe now](#/portal/signup) ## Starting the conversation When you’ve got a rough idea turning over in your head, start with AI. Don’t wait until you’ve already decided what good looks like. Invite AI’s opinion from the get go: > I’m thinking about \[idea/task/problem\]. I want to explore how to approach it. Work with me as my cognitive sparring partner. Bounce ideas around with me, critique my thinking, and ask me anything you need to fully understand what I’m trying to accomplish. We’re not prompting here, we’re having a conversation. Invite questions and challenge to your framing, and you get connections and angles you’d never have found alone. The word that matters in that prompt is critique. Ask for a review and you’ll get a polite once-over. Ask for a critique, and you’ll get the full works. AI is a mirror. It matches your energy. Lay out the easy path and it’ll walk it with you. Give it permission to be brutal and get a barbarian. ## Going full Bayesian Every plan is built on a foundation of things you believe to be true but haven’t tested. In Bayesian reasoning, these are known as priors. Our goal is to aggressively prove ourselves wrong - this is how we discover actual truth. Try asking AI to surface your assumptions and define your priors. AI excels here precisely because it doesn’t share your context, your cognitive architecture, or your emotional investment in the outcome. It can see the things you miss. Once you can see your priors, stress-test them. Ask for three difficult questions that undermine your approach. Don’t go easy – you want the kind a hostile board member or a sharp competitor would ask. If your plan gets sunk, time to rebuild. Push this further by asking: > Perform research into my priors, looking for data and evidence which confirms, supports, contradicts, or invalidates them. Then check for biases. Confirmation bias. Sunk cost fallacy. Anchoring. Availability heuristic. Overconfidence bias. Even if you’ve heard of some of these, you won’t know every bias and fallacy. AI does, and it isn’t affected by them in the same way we all are. > You are an expert in \[subject matter\]. Attached is my \[argument/proposal/thesis\]. Identify any cognitive biases and logical fallacies present in my reasoning. Quantify the uncertainty. Thinking in probabilities rather than certainties is one of the most useful cognitive upgrades you can make, and AI is a natural fit because it’s already thinking probabilistically under the hood. > Ask: given this evidence, how should my confidence shift? Not “am I right or wrong” but how confident should I be, and which direction has the evidence pushed me? You’ll likely recognise the straw man: presenting a deliberately weak version of your opponent’s argument so you can knock it down easily. A steel man is the opposite: the strongest possible version of the case against yours. You want this because if your thesis survives the best counter-argument that can be constructed, it can survive anything. > I am presenting my thesis that \[your argument\]. You are an expert in \[relevant field\]. Attached is my argument. Your task is to steel man the strongest possible counter-argument. Once your thinking solidifies into something resembling a plan, it’s time to break it: > You are an expert in \[your domain\]. We’re working on a plan for \[task\]. Assess and critique this plan \[paste or attach it\], and present your findings as a red-team report. Finally, compress. Ask AI to tighten your logic, cut the fluff, and flag the leaps - the places where your argument jumps from A to C without proving B. What survives this step is the real core of your thinking. This is Bayesian thinking in action. You’re asking: where do my beliefs have holes? Where am I reasoning from habit or emotion rather than evidence? What proves or disproves my priors? Actively seek to be wrong. This is how you are proven right. ## Amplification vs Offloading Everything above will sharpen your thinking. But there’s a trap. As with skills, when you offload critical thinking to AI, that muscle starts to atrophy. The trick is to keep yourself in the mix. Push back on AI’s challenges to you. Ask it to prove its own thinking. Go do some Google searches on the side. Have an argument. Read the feedback then go for a walk to chew it over. If it isn’t evoking angry responses from you, you’re not getting it to push far enough: embrace the suck. This is the principle we’ll keep coming back to throughout this series: amplification, not offloading. Sparring with AI makes your existing thinking sharper. The point of sparring – be it cognitive or combative – is to practice. We don’t practice so our skills weaken. We do it so they strengthen, allowing us to take on greater challenges. ## Reps, reps, reps Over the next week, try this out regularly on something you’re working on right now - a plan, a proposal, an argument you’ve been developing. Ask: > I want you to act as my cognitive sparring partner. I’m going to share \[my plan/idea/argument\]. Your job is to surface my assumptions, challenge my weaknesses, identify biases in my reasoning, and steel man the strongest counter-argument. Don’t be polite. Don’t forget, your instructions shape every conversation ([see last week’s post](https://bayesiantheory.com/ai-bayesics-2-communicating-effectively/)). Adding “Be my cognitive sparring partner. Challenge my assumptions and flag cognitive biases when you spot them” means it happens by default. This is the difference between a one-off cognitive health check and embedding it in your mental immune system. Next week, we’re taking this further by looking at creativity and brainstorming: how to use AI to come up with ideas, and how to tell the good ones from the shite. --- Thanks for reading. If you found this piece useful, please feel free to share with your friends and professional contacts. AI Bayesics publishes every Friday on the Bayesian Theory Substack and is available to free subscribers. [Subscribe](#/portal/signup)