Every time you use AI without thinking about how it thinks, it gets a little easier to stop thinking for yourself. The Mind Machine covers the cognitive science of that shift — the biases it exploits, the habits it builds, and the exact moves to stay sharp.
By Pratik Kamble — Every Thursday, free.
Full issue free, every Thursday. No spam. What Pro adds →
Your welcome letter is already on its way. Not a template — a real note about what this newsletter is and why it exists.
Issue #001 is in your inbox now. See you every Thursday at 7am.
Three questions. Each one reveals a different way AI might be shaping your thinking without you knowing.
Every other AI newsletter tells you what tools exist. The Mind Machine tells you what those tools do to your brain. Specifically: if you're a product manager, marketer, consultant, or builder who opens ChatGPT more than once a day and wants to understand what that habit is actually doing to your cognition — this is written for you.
Written by Pratik Kamble — reading the actual cognitive science papers, testing the findings on real AI workflows, and writing up what holds. Not a tech writer. Not an engineer. Someone who noticed AI was quietly changing how he thought, and started tracking exactly how.
The complete first issue. All five sections. Based on real 2026 research from HBR, PMC, and ScienceDirect.
Every tab-switch between AI tools is quietly destroying your most valuable cognitive resource. Gloria Mark's landmark research — and what it actually means for your workflow.
Why handing tasks to AI doesn't relieve cognitive pressure — it reroutes it. The psychology of locus of control, delegation anxiety, and learned helplessness in the age of AI. Grounded in Rotter (1966), Seligman (1967), and 2024 longitudinal research.
The Thursday issue is the foundation. Pro subscribers get four additional focused drops every week — each one short, specific, and immediately usable. Thursday through Sunday, your brain gets sharper.
That's 5 emails a week — all grounded in real psychology research, all written by Pratik. No AI-generated filler. No padding. Just the useful part.
Preview any Pro document below — these are real, full documents, not samples.
Start free. Upgrade when you're ready. No pressure, no dark patterns — just honest tiers for different levels of curiosity.
Switch to see how little this actually costs per day
Automation Bias, Cognitive Offloading, and the 78% stat that should make every AI user pause. The most important thing to read before you open ChatGPT again.
"The Real Risk Isn't AI Replacing Your Thinking. It's AI Quietly Reshaping How You Think — Without You Noticing."
The Mind Machine · Issue #001 · Pratik Kamble
AI is here to assist your thinking — not replace it.
One full issue every Thursday — the Behaviour Brief, the Deep-Dive, the Toolkit, Wired vs Tired. All of it free. Pro adds The Influence Move and four more drops every week.
Full issue free forever · No spam · Unsubscribe in one click · What Pro adds →
A proper welcome letter is headed to your inbox right now — not a template, a real note about what this newsletter is and why it exists. Thank you for trusting us with your attention. We don't take that lightly.
Your first issue arrives this Thursday at 7am. Until then — notice the next time an AI answer feels a little too confident. That's exactly where we'll start.
— Pratik
The full Issue #001 is embedded above in the main page — scroll up to read all five sections including The Behaviour Brief, The Deep-Dive on Cognitive Offloading, The Cognitive Toolkit, The Influence Move (Pro), and Wired Vs. Tired.
Every Thursday I give you one AI prompt — fully dissected. Not just "here's a prompt." You get the psychology behind why it works, the exact wording, a before/after example, and one variation to push it further. This week: the single most effective prompt reframe I've tested across 40+ use cases.
The Contrast Principle (Cialdini, 1984) states that our perception of something is always relative to what we compare it against. A ₹5,000 jacket feels cheap right after looking at a ₹25,000 one. A mediocre job offer feels great after a rejection. We don't evaluate things in isolation — we evaluate them against an anchor.
AI models work the same way internally. When you ask for "a good introduction," the model has no contrast anchor — it generates something statistically average, because average is what dominates training data. But when you show it a bad example first, you give it a contrast anchor — and the output quality jumps noticeably.
This has been validated not just anecdotally but through research on large language model output quality. Providing negative exemplars (bad examples to avoid) significantly improves output distinctiveness and quality versus zero-shot prompting alone.
Three things are happening simultaneously in this prompt:
Once you're comfortable with single contrast, try this: give AI both a bad example and a mediocre example, then ask for something that makes both look obviously inadequate. The model now has two contrast anchors pulling in the same direction. Output quality increases again. Use this for anything where "good" isn't enough and you need "genuinely sharp."
"Show AI what bad looks like — and it will spend all its effort running away from it."
Next Thursday: The Socratic Interrogation Prompt — how to use AI as a devil's advocate to stress-test any idea before you act on it. Based on the same dialogic reasoning technique Socrates used 2,400 years ago. Still works.
Every month, I send this audit. It's 7 questions — one for each cognitive bias most activated by daily AI use. You rate yourself honestly from 1 to 5. You note one specific instance from this month where the bias showed up in your work. Then I give you one corrective action.
The science behind this is solid. Metacognitive awareness — the act of consciously reflecting on your own cognitive patterns — is one of the few evidence-based interventions that actually reduces bias over time. You cannot correct a pattern you haven't noticed. This is how you notice it.
Research by Morewedge et al. (2015, Policy Insights from the Behavioral and Brain Sciences) found that targeted debiasing training reduced bias by up to 29% — and the most effective method was exactly this: self-audit with specific feedback, done repeatedly over time.
For each bias below: read the definition, rate yourself honestly (1 = never, 5 = constantly), and write one sentence describing a specific instance from this month. Be specific — "I used AI for strategy" is not useful. "I accepted Claude's market analysis for my pitch deck without checking any of the figures" is useful.
Definition: Over-relying on AI outputs without applying your own critical judgement. Treating the machine's answer as more reliable than your own experience.
Signs you scored a 4–5: You regularly use AI-generated content, analysis, or decisions without meaningfully questioning them. You feel slight discomfort when your instinct contradicts an AI output — and you go with the AI anyway.
Corrective action if you scored 4–5: For the next two weeks, implement the "Junior Colleague Test" on every AI output before acting on it. Ask: "If a junior colleague sent me this exact output, would I accept it without questioning?" Apply the same standard to AI.
Definition: Progressively outsourcing cognitive tasks to AI that you used to do yourself — and losing the ability to do them independently as a result.
Signs you scored a 4–5: You reach for AI before attempting a task yourself. You feel cognitive strain on tasks that used to feel easy. Your first drafts have gotten worse, not better, since you started using AI heavily.
Corrective action if you scored 4–5: The "Think First" rule — before opening any AI tool for a task requiring judgement, write your own answer in 3 sentences. Then use AI to challenge it. Never start with AI on creative or strategic work.
Definition: Using AI to find evidence that confirms what you already believe, rather than to genuinely challenge your thinking.
Signs you scored a 4–5: Your AI prompts often start with your conclusion and ask for support ("give me reasons why X is a good idea"). You rarely ask AI to argue the opposite of what you think. AI's answers tend to agree with you most of the time.
Corrective action if you scored 4–5: For every significant decision this month, run the "Steel Man Audit" — after getting AI to support your view, ask it: "Give me the strongest possible argument against this position, assuming a highly intelligent sceptic." Then actually read it.
Definition: Mistaking AI's confident, well-written output for accuracy. Smooth writing triggers a "competence heuristic" in the brain — we assume fluent = correct.
Signs you scored a 4–5: You trust longer, more detailed AI responses more than shorter ones. You feel more confident in AI outputs that use technical vocabulary. You rarely fact-check AI claims that are written persuasively.
Corrective action if you scored 4–5: The "Citation Rule" — if AI makes a factual claim you plan to use, you find the primary source yourself before using it. If you can't find it, you don't use the claim. No exceptions.
Definition: Choosing AI tools that feel good to use right now over tools that build better long-term cognitive habits. Optimising for instant gratification over compounding value.
Signs you scored a 4–5: You use AI tools because they're satisfying to use, not because they make you better at something. Your tool stack has grown but your thinking hasn't. You feel slightly worse at focusing deeply than you did a year ago.
Corrective action if you scored 4–5: Audit your AI tool stack this week. For each tool, ask: does this make me better at something independently, or does it just do the thing for me? Keep the first kind. Be honest about the second.
Definition: Treating AI's first output as the reference point for all subsequent refinements — anchoring your thinking to what the machine generated first rather than to your own independent frame.
Signs you scored a 4–5: When you don't like an AI output, you ask it to improve rather than starting from your own version. Your "editing" of AI content is mostly cosmetic. You can't easily write a better first draft than AI gave you.
Corrective action if you scored 4–5: The "Draft Zero" habit — before asking AI for anything creative or strategic, write your own 5-sentence rough version. Call it Draft Zero. Then ask AI to respond to your draft, not to start from scratch. You've set the anchor. AI responds to yours.
Definition: Trusting AI outputs more because "everyone uses ChatGPT" — applying the social proof heuristic to an automated system that has no social consensus about accuracy.
Signs you scored a 4–5: You've cited an AI output in professional work without verifying it. You've used "AI said so" as a reason in a conversation. You trust ChatGPT more than a colleague's equivalent opinion.
Corrective action if you scored 4–5: The "Would I Say This?" rule — before using any AI-generated claim professionally, ask: "Would I say this out loud if someone asked me to back it up?" If no, verify it or don't use it.
Add your 7 scores. Your total is somewhere between 7 and 35.
Every quarter I compile everything covered in the last 12 issues into one reference document. This is Q2 2026 — the founding edition. It contains every bias, tactic, prompt framework, and corrective action from the first 12 weeks of The Mind Machine, organised for practical use.
This is not a summary. It is a working reference — designed to be used alongside your AI tools, not read once and forgotten. Keep it open. Refer back to it. Add your own notes to the margins.
Source: Lisanne Bainbridge (1983); Goddard et al. (2024), Computers in Human Behavior
What it is: Over-reliance on automated systems and under-reliance on human judgement. When AI produces an output, the brain defaults to accepting it — especially when the output is well-formatted and confident in tone.
How AI activates it: Every AI interface is designed to produce fluent, structured, confident-looking output. This triggers the "expert heuristic" — we assume confidence equals competence. Bullet points make it worse (38% more trusted than prose, per research).
Daily corrective: The Junior Colleague Test. Before acting on any AI output, ask: "Would I accept this from a junior colleague without questioning?" Apply the same scrutiny to AI that you'd apply to a human.
Source: Risko & Gilbert (2016), Current Directions in Psychological Science; Kian et al. (2024), Frontiers in Psychology
What it is: Progressive outsourcing of cognitive tasks to external systems — and the accompanying atrophy of the cognitive muscles that those tasks were building.
How AI activates it: AI is frictionless. The threshold for outsourcing has collapsed. We now outsource tasks we used to do because they were "hard" — and hard was where the learning was happening.
Daily corrective: Think First, AI Second. Write your own 3–5 sentence answer before opening any AI tool for a judgement-based task. AI challenges your draft. You don't start from AI's.
Source: Nickerson (1998), Review of General Psychology; Wason (1960) selection task
What it is: The tendency to seek, interpret, and remember information in a way that confirms existing beliefs. AI makes this dramatically easier — you can get a detailed, well-written confirmation of almost any belief if you phrase the prompt correctly.
How AI activates it: Most people prompt AI with their conclusion implicit in the question. "Why is remote work more productive?" already contains the answer. AI obliges.
Daily corrective: The Steel Man Protocol. For every significant belief or decision, ask AI: "Give me the strongest possible argument against this position, from the perspective of a highly intelligent sceptic who has thought deeply about this." Read it seriously.
Source: Alter & Oppenheimer (2009), Psychological Bulletin; processing fluency research
What it is: Mistaking ease of processing for truth. Smooth, well-written text triggers a feeling of rightness — the brain equates fluency with accuracy. AI writes fluently by design.
How AI activates it: AI output is always grammatically correct, well-structured, and confident. These are all surface signals our brain uses as proxies for reliability. The content may be wrong — the format never is.
Daily corrective: The Citation Rule. Any factual claim from AI that you plan to use professionally requires a primary source you found yourself. No source = don't use the claim.
Source: Laibson (1997), The Quarterly Journal of Economics; hyperbolic discounting research
What it is: Overweighting immediate rewards vs future costs. AI tools that feel good to use right now may be building long-term cognitive dependency at a cost you won't feel until later.
How AI activates it: Every AI tool is optimised for immediate satisfaction — fast outputs, smooth UX, instant gratification. The long-term cost (cognitive atrophy, skill regression, dependency) accrues invisibly.
Daily corrective: The Skill Audit. For each AI tool you use regularly, ask: does this build a skill I can use independently, or does it do the skill for me? The second category needs conscious management.
Source: Tversky & Kahneman (1974), Science; anchoring and adjustment heuristic
What it is: Over-weighting the first piece of information encountered when making subsequent judgements. When AI's first output becomes your reference point, all subsequent thinking is adjustment from that anchor — not independent generation.
How AI activates it: You ask AI for a first draft. AI gives you something. Everything you write after that is a response to AI's anchor. Your original frame never forms.
Daily corrective: Draft Zero. Write 5 sentences of your own first. Call it Draft Zero. Then ask AI to respond to your draft, challenge it, or build on it. You set the anchor.
Source: Cialdini (1984), Influence; social proof as heuristic
What it is: Using the popularity or widespread use of an AI tool as evidence of its reliability. "Everyone uses ChatGPT" is not evidence that ChatGPT's outputs are accurate. It is evidence that ChatGPT is popular.
How AI activates it: Mass adoption creates an authority halo. When 200 million people use a tool, the brain assigns it expert status without evaluating individual outputs.
Daily corrective: The "Would I Say This?" test. Before using any AI claim professionally: "Would I say this out loud if someone asked me to back it up?" If no, verify it or don't use it.
Show AI a bad example, explain why it's bad, then ask for something dramatically better. Provide a negative exemplar and explicit failure mode labelling before the actual request.
Force AI to build the strongest possible argument against your position before you act on it. Breaks confirmation bias by making opposition explicit and high-quality.
Write your own 5-sentence rough version before asking AI for anything creative or strategic. Then ask AI to respond to your draft. You set the cognitive anchor.
Use AI as a Socratic interlocutor — ask it to question your reasoning, find gaps, and expose assumptions you haven't noticed. Based on the Socratic method of knowledge through dialogue.
When AI gives you a generic answer, force it down the specificity ladder until the output is actually useful for your exact situation. Generic → Specific → Contextual → Actionable.
Before opening any AI tool in the morning: write 3 sentences about what you're trying to accomplish today and what the most important decision or creative challenge is. Do this in a notebook, not on a screen. This establishes your own cognitive frame for the day before AI can set an anchor.
For any task requiring judgement, creativity, or strategy: write your own Draft Zero first. Minimum 3 sentences. Maximum 10 minutes. Then and only then open AI. AI responds to your thinking — it does not replace it.
Before acting on any significant AI output: "If a junior colleague sent me this, would I accept it without questioning?" Apply the same scrutiny. This takes 10 seconds and catches the majority of Automation Bias failures.
Every Friday, ask yourself: which of the seven biases showed up most clearly this week? Name one specific instance. What would you do differently? Write it down. This is the micro version of the Monthly Audit — same mechanism, lighter touch, higher frequency.
Any factual claim from AI that you intend to use professionally, publish, or present: find the primary source yourself before using it. If you cannot find a credible primary source in 5 minutes of searching, do not use the claim. This is non-negotiable.
This is your quick-reference card. When you're mid-task and something feels off about how you're using AI — check this first. Each bias is defined in one sentence, its AI trigger in one sentence, and its fix in one sentence. No padding.
"Use AI as the second brain that challenges your first one — not the first brain that replaces it."
Every bias on this list is a version of the same mistake: letting AI's process substitute for your own. The corrective for all seven is the same: maintain your own cognitive frame first, then use AI to stress-test, challenge, and extend it. AI is the sparring partner. You are the fighter.