Are We Training Humanity Into System 1?

Google wants Gemini to think for you. Perplexity has replaced searching with asking. Meta AI is sitting inside your WhatsApp conversations. The question nobody is asking: what happens to the human capacity for deliberate thought?

Last month, Google announced its vision to transform Gemini into a “Universal AI Agent”, a system that doesn’t just answer questions but takes actions, makes decisions, and navigates the digital world on your behalf. It books your flights. It summarises your meetings. It drafts your responses. It does your research, your synthesis, your first-pass analysis.

It does your thinking.

Google isn’t alone. Perplexity has fundamentally reframed how people access knowledge, not by helping you search but by eliminating the need to search at all. You ask a question; it gives you the answer, synthesised from dozens of sources you never have to visit, evaluate, or compare. The entire cognitive process of gathering evidence, weighing credibility, and forming your own synthesis, the slow, deliberate work of System 2, has been compressed into a single conversational exchange. Microsoft has embedded Copilot so deeply into Office, Teams, and Windows that opting out requires more effort than opting in. Apple Intelligence rewrites your messages, summarises your notifications, and prioritises your email. Meta has placed its AI directly inside WhatsApp and Instagram, not as a separate tool you go to, but as a participant in your existing conversations, ready to answer questions, generate ideas, and settle arguments in the flow of your most informal, System 1-native communication space. OpenAI’s ChatGPT has become the default thinking partner for hundreds of millions of people, the first place they go when they encounter a problem that requires more than a Google search.

Every major technology company on the planet is making the same bet: that you would rather have the AI think than think yourself. And they are almost certainly right.

But nobody is asking what that costs.

The Cognitive Subsidy

In 2002, the psychologist Daniel Kahneman formalised something that behavioural scientists had observed for decades: the human brain operates on two distinct processing systems.

System 1 is fast, automatic, and intuitive. It’s the snap judgement, the gut feeling, the pattern recognition that happens without conscious effort. It’s what makes you flinch before you’ve consciously registered the threat, what lets you read a face in milliseconds, what drives the majority of your daily decisions without you ever noticing.

System 2 is slow, deliberate, and analytical. It’s the careful calculation, the logical argument, the effortful reasoning you deploy when the stakes are high and the problem is complex. It’s what makes you check your working, question your assumptions, and arrive at a conclusion you can actually defend.

Here’s the critical detail that most popular summaries of Kahneman’s work miss: System 2 is expensive. It consumes significant metabolic energy. It requires sustained conscious focus. The brain, operating under a strict energy budget, is what psychologists call a “cognitive miser”, it will always prefer to delegate to the cheaper, faster System 1 when it can get away with it.

What AI is doing, across every major platform, at unprecedented speed, is making it possible to get away with System 1 for almost everything.

Need a research summary? Perplexity synthesises it from twenty sources you’ll never read. Need a strategic analysis? Copilot produces the framework before you’ve finished defining the problem. Need to understand a complex document? Gemini extracts the key points in seconds. Need to settle a factual disagreement? Meta AI resolves it mid-conversation in WhatsApp without anyone leaving the chat. Need to make a decision? ChatGPT lays out the options, weighs the trade-offs, and recommends the optimal path.

The cognitive work that used to require System 2, the effortful, sequential, deliberate processing, is being automated. What remains for the human is System 1: recognise the output, evaluate it intuitively, accept or reject it on feel.

We are not just adopting AI tools. We are being trained into a cognitive mode.

The Psychological Dividend

Not everyone experiences this the same way, and the differences are not random. They follow predictable patterns rooted in what psychologists call motivational orientation, the fundamental needs that drive how individuals process information and make decisions.

Some people are energised by analysis itself. For them, the effort of careful thinking is not a cost to be minimised but a reward to be sought. The process of working through a problem step by step, stress-testing assumptions, and arriving at a well-defended conclusion produces what researchers call a “Psychological Dividend”, a sense of competence and mastery that comes from the rigour of the method, not just the quality of the output.

These are the people for whom AI poses the most subtle and insidious threat. Not because AI produces bad work, but because it produces good enough work that bypasses the process they value. When the analysis is handed to you pre-formed, the dividend disappears. The competence signal, I did this well because I thought it through carefully, is replaced by something flatter: I accepted this because it looked right.

For others, the calculation is different. Some people are optimised for speed and social fluency. Their cognitive strength is rapid pattern recognition in dynamic environments, particularly human environments. For them, AI doesn’t threaten a valued process; it extends an existing capability. The AI is another fast-moving system to navigate, another relationship to manage. The cognitive mode AI encourages, quick evaluation, intuitive acceptance, feels natural rather than diminishing.

And then there are those whose primary need is security. For them, the opacity of AI, the inability to see how a conclusion was reached, is not a minor inconvenience but a fundamental violation of how they build trust. They need the audit trail. They need the checklist. They need to verify that every step was followed. When AI says “trust me”, these are the people who instinctively ask “why should I?”

The Velocity Collapse

The danger is not that AI makes us stupid. The danger is that AI makes fast thinking good enough that we lose the capacity, and eventually the desire, for slow thinking.

In organisational psychology, there is a well-documented failure mode called “Velocity Collapse”: a cycle where decisions are made increasingly fast, with decreasing scrutiny, producing errors that are only visible downstream. Each fast decision that appears to work reinforces the pattern. The speed feels productive. The errors, when they eventually surface, are attributed to execution failure rather than to the absence of the deliberative process that would have caught the problem earlier.

AI accelerates this cycle dramatically. When an AI-generated strategy document reads better than what most humans would produce in a week, the incentive to spend a week thinking about it approaches zero. When an AI-synthesised research summary captures 90% of the insight in 10% of the time, the remaining 10% of insight has to be extraordinarily valuable to justify the cognitive investment.

And here’s the trap: the 10% you miss is almost always the part that matters most. It’s the assumption nobody questioned because the AI presented the consensus view confidently. It’s the edge case that didn’t appear in the training data. It’s the strategic insight that requires connecting two domains that have never been connected before, something that only happens when a human mind is doing the effortful, associative, slow work of synthesis.

The companies building these systems are not trying to make us worse at thinking. They are trying to make us faster. But speed and rigour exist in tension, and every cognitive subsidy shifts the balance toward speed. Over time, the muscle for deliberate analysis atrophies. Not because it was taken away, but because it was never exercised.

The Rationalisation Engine

There is a darker dimension to this that deserves explicit attention.

One of the most well-established findings in cognitive psychology is that System 2 often functions not as an independent scrutineer but as a post-hoc justifier. System 1 reaches a rapid conclusion based on intuition, bias, or emotional preference, and then System 2 constructs a plausible logical narrative to defend it. The person believes they are being rational. They are actually being rationalised.

AI supercharges this dynamic.

Now, when your System 1 decides it likes a particular option, you don’t even need to construct the justification yourself. You can ask the AI to build the case. It will produce a beautifully structured argument, complete with evidence, logical sequencing, and appropriate caveats, for whatever position you already hold. The rationalisation is no longer internal and effortful. It is outsourced, automated, and professional-grade.

This is not hypothetical. It is already happening in boardrooms, in marketing departments, in strategy meetings, in political campaigns. The AI does not know what you should think. It knows how to argue persuasively for what you already think. And because the output looks like rigorous System 2 analysis, it passes unchallenged.

What Google Isn’t Telling You

When Google describes Gemini as a Universal AI Agent, the framing is about empowerment. More capability. Less friction. More time for what matters. When Perplexity’s CEO describes the product, the pitch is that you should never have to visit a website again. When Meta positions AI inside WhatsApp, the value proposition is seamless convenience, the answer without the detour.

But “less friction” in cognitive terms means “less System 2 engagement.” And “more time for what matters” often means “less time spent on the thinking that used to be considered the whole point.”

Consider what Perplexity specifically eliminates from the cognitive process. Traditionally, searching for information required you to formulate a query, evaluate which sources looked credible, visit multiple pages, compare conflicting claims, synthesise the findings, and form a conclusion. That is sequential, effortful, System 2 processing. Every step required a judgement call. Perplexity collapses all of it into a single interaction: you ask, it answers, you move on. The synthesis happened, but you didn’t do it. And because the output is fluent, well-sourced, and formatted with citations that signal rigour, your System 1 accepts it as trustworthy. The System 2 verification that you would have performed if you’d done the research yourself never happens.

The companies building these systems understand something they are not advertising: the most successful AI products will be the ones that most effectively reduce the user’s cognitive load. That is the product metric. That is what drives engagement, retention, and revenue. Every design decision optimises for the same thing: making it easier to accept the AI’s output than to think independently.

This is not a conspiracy. It is a market. The demand for cognitive offloading is enormous, and the supply is improving at a rate that has no historical precedent. The companies responding to that demand are doing exactly what companies do: giving people what they want.

The question is whether what people want is what they need.

The Speed vs Rigour Divide

We are entering a period where the most significant human differentiation will not be between those who use AI and those who don’t. It will be between those who can think alongside AI and those who have outsourced thinking to AI.

The difference sounds subtle. It is not.

Thinking alongside AI means maintaining the capacity for deliberate, sequential, effortful analysis while leveraging AI’s speed for data processing, pattern recognition, and first-draft generation. It means treating the AI’s output as a starting point for System 2 scrutiny, not as a finished product to be accepted on System 1 intuition. It means preserving the Psychological Dividend, the reward that comes from the rigour of the process, not just the efficiency of the outcome.

Outsourcing thinking to AI means accepting the cognitive subsidy at face value. It means allowing the brain’s natural preference for System 1 efficiency to go unchallenged. It means mistaking the fluency of the AI’s output for the quality of the underlying reasoning.

The first group will be amplified by AI. The second group will be dependent on it. And the gap between them will widen faster than most people expect.

The Alienation Problem

For a significant minority, this transition will feel not just uncomfortable but threatening. These are the people whose identity is built on the quality of their thinking, whose sense of competence comes from the rigour of their analysis, whose sense of security comes from understanding how conclusions were reached.

For them, AI does not feel like a tool. It feels like an erasure.

When a machine can produce in seconds what took you hours of careful thought, the implicit message is that the careful thought was never the valuable part. The output was. And if the output is what matters, then the process you valued, the slow, deliberate, effortful work of System 2, is just overhead.

This is not a message that technology companies are eager to deliver. But it is the message that is being received, and it is driving a form of cognitive alienation that does not have a name yet but is already visible in the resistance to AI adoption among precisely the people whose analytical capabilities are most needed in an AI-augmented world.

What Happens Next

The trajectory is set. AI will continue to absorb cognitive tasks that were previously the domain of human System 2 processing. The companies building these systems will continue to optimise for reduced cognitive load because that is what drives adoption. The majority of users will continue to accept the cognitive subsidy because the immediate benefits are real and tangible.

But the long-term consequences are worth naming now, before they become invisible:

The atrophy of deliberative capacity. Muscles that are not exercised weaken. The ability to engage in sustained, effortful, sequential reasoning is not a fixed trait. It is a skill that requires practice. If AI removes the need for that practice, the skill will decline, not just individually but culturally.

The erosion of epistemic standards. When AI-generated analysis is indistinguishable from human analysis in quality and fluency, the social signals that distinguish rigorous thinking from plausible-sounding thinking break down. The ability to tell the difference, in yourself and in others, diminishes.

The widening of the cognitive divide. Those who maintain the capacity for deliberate thought alongside AI augmentation will operate at a level that pure AI-dependents cannot match. This will not be visible in routine tasks. It will be visible in the moments that matter: the strategic pivot, the ethical judgement, the creative breakthrough that requires connecting ideas from domains that have never been connected.

The companies building the future of AI are not thinking about these consequences. They are thinking about market share, engagement metrics, and the next capability milestone.

Someone needs to be thinking about what happens to human thinking.

That is a System 2 problem. And we are running out of people who still know how to do it.

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.