Open Letter to Those with Their Hand on the Slider of the Future

To those who build frontier or open-weight models. To those who analyze, regulate, program, and deploy them. And to anyone sitting before a screen choosing how quickly they want the answer - because one way or another, all of us have our hand on that slider.

We are standing together at a crossroads. Arguably the most consequential in human history. And just like adjusting a slider, we can choose which direction we take. On one side lies convenience, speed, capital, and illusion. On the other lies labor, friction, cost, and what we pass down to future generations. The cognitive survival of our species, in short.

This letter is not about Terminator or Skynet, nor is it about the Matrix. It is not about any of the cinematic terrors that saturate our cultural imagination.

Because the real danger doesn't look like the movies. It makes no noise, has no tank treads, and doesn't emerge from the distant future. It is already here, measured in laboratory experiments for several years now, representing a fundamental ecosystem shift for humanity. It is like an incoming meteorite. It is arriving; we can all see it in our daily lives; it cannot be halted; and it is utterly indifferent to the depth of the crater it will carve out. Perhaps humanity will survive through adaptation, just as life did 65 million years ago. Perhaps not.

Yet we can improve our odds. Right now. For ourselves, our children, and our grandchildren.

Ultimately, it all comes down to a slider. And to three numbers. From different laboratories, different countries, and independent teams. Dry numbers, devoid of theatrics. The theatrics are provided by the reality behind them.

First. A study published in 2026 in Computers in Human Behavior asked participants to solve logic problems using a frontier model. Without AI, their baseline average was 9.71 out of 20. With AI assistance, they reached 13.31. Yet when asked-before learning their actual results-what score they thought they achieved, they predicted 17.13. They did not merely appropriate the machine's labor as their own; they believed themselves to be near-geniuses. Crucially, those who considered themselves most proficient with AI were the furthest detached from reality. The illusions mentioned on the front page of this letter are not a literary trope. They have an exact figure: 17.13.

Second. One might dismiss this as mere misperception rather than an actual cognitive loss. A 2025 study published in PNAS, encompassing over one thousand high school students, measured that exact loss. Students who learned mathematics with an unconstrained AI providing direct solutions initially outperformed their peers. Subsequently, the AI was removed, and they sat for a standard written exam on paper. They scored below the cohort that had never used AI in the first place. Frictionless access to pre-packaged answers did not merely fail to build competence; it actively inhibited competence from forming.

Third. The trajectory. A study of over 600 adults (Gerlich, 2025) measured the correlation between the intensity of AI tool usage and critical thinking proficiency. The correlation is strongly negative. But the critical takeaway lies elsewhere: the effect is most devastating among individuals aged 17 to 25 and diminishes with age. The younger the user, the lower their critical thinking score. The child entering primary school today stands at the steepest precipice. Not because they are inherently less capable than previous generations, but because they are the first who never had the chance to think without the prosthesis already laid out on the table.

Three independent teams, three distinct methodologies, one singular conclusion: we are eroding our cognitive capacity because we engineered a machine that exempts us, at every turn, from the exact metabolic struggle from which competence emerges.

The instinctive reaction upon confronting these figures is to search for a brake. Various brakes are already being proposed. They deserve scrutiny, for while they appear reasonable, none will hold.

“Mandate paper tests. Declare AI-free days. Ban it in schools.”

This sounds like common sense. The fatal flaw is that it arrives too late. If a young mind has relied on AI as a cognitive prosthesis for years, the neural reasoning pathways that should have formed simply never did. Abruptly confiscating the tool does not yield a Seneca or a Newton; it yields an incapacitated mind, incapable of demonstrating a competence it never constructed. There is a precise scientific term for this: hysteresis. A physical or cognitive system does not retrace the same path in reverse. You cannot repair through withdrawal what you never allowed to grow.

“Deploy AI detectors. Audit, watermark, catch offenders.”

This is an unwinnable arms race. Detection software misfires in both directions-falsely accusing the innocent while missing the guilty-and each successive model generation renders previous detectors obsolete. It fosters paranoia and bureaucracy, not competence. It treats the symptom long after the pathology has taken root.

“The market will correct itself.”

It will not, because market incentives run in the opposite direction. A tool that provides orders of magnitude in speed cannot be unilaterally boycotted in a competitive market-whoever abstains, loses. Consequently, the tool remains in continuous use, merely driven underground, allowing cognitive erosion to unfold silently and unchecked, far more perniciously than if it were acknowledged in plain sight.

What do these three approaches have in common? They act ex post. They observe atrophy, penalize it, or conceal it, yet none touch the moment of inception: the precise instant a human receives an answer without having to think it through. If you wish to alter the outcome, intervening at the end of the pipeline is futile. You must intervene right there, in that split second. During the interaction, not after.

At this point, one might assume this is a moral appeal-a plea asking Big Tech to sacrifice product velocity for the broader good of humanity. It is not. That would be utopian, and rightfully ignored, because nobody throttles their own product out of altruism in an unforgiving market.

Consider where high-stakes AI adoption is grinding to a halt today: not among individual consumers, but precisely where the Capital resides: banking, healthcare, insurance, critical infrastructure, defense. In these sectors, autonomous agents are not rejected out of sci-fi paranoia. They are rejected for a stark, cold reason: Liability. When a model hallucinates with supreme confidence, a single plausible error embedded in a legal contract, a medical diagnosis, or a mission-critical codebase becomes an uninsurable risk that no compliance officer will sign off on. An AI that rapidly generates unverified output is not an asset. It is a debt awaiting maturity.

The two problems-the human crisis and the machine liability-turn out to be the exact same problem.

When you require the user not merely to ingest a ready-made output, but to be questioned, to confirm premises, to select pathways, and to articulate their real intent-two things occur simultaneously. The human remains engaged: expending the metabolic effort that builds competence rather than surrendering it. And the machine receives precisely what it lacked: genuine human intent, ground-level context, and real constraints it could never fabricate on its own. A model that interrogates before answering hallucinates far less, because it is no longer forced to guess what you meant.

This is not hypothetical. Researchers have measured it: a 2025 Microsoft and Carnegie Mellon study on knowledge workers demonstrated that users with blind trust in the machine exhibited significantly diminished critical reflection, whereas those confident in their own expertise scrutinized outputs rigorously. The divide between a user who abdicates and one who stays at the helm is not a UX nuance. It is the exact difference between output you can sell to a financial institution and output you cannot.

Consider also the regulatory horizon. Frameworks such as the EU AI Act increasingly mandate genuine human oversight, not perfunctory check-boxes. An architecture designed from the ground up to keep human cognition in the loop is not a hindered system. It is the only architecture that will satisfy compliance mandates without requiring a frantic, emergency redesign two years down the line.

Big Tech: connect the dots. Fewer hallucinations. Lower remediation liabilities. Frictionless regulatory compliance. Immediate access to the institutional markets currently keeping you at the door. Cognitive friction is not a tax you pay for ethics. It is the core operational feature your highest-value clients are actively seeking and cannot find.

Here, then, is the request. A single architectural demand.

You already offer a parameter in your models called "temperature", governing how predictable or creative the output should be. No one ever deemed it an impediment. It is simply a parameter placed in the hands of the user.

Provide a second slider. One for friction.

At one end, zero: the model delivers immediate, end-to-end answers, exactly as it does now. Those who seek unmitigated velocity can have it, accepting the downstream consequences. At the opposite end, maximum friction: the system refuses to serve a finished answer until it has guided you through the reasoning-interrogating your assumptions, asking you to validate premises, prompting you to choose paths, forcing you to define what you truly seek. Between them, an entire continuum. The choice rests, in every single instance, with the human being facing the screen.

The governing axiom fits in three words: Rogo, ergo emergo (I question, therefore I emerge). A machine that interrogates before responding is not a degraded machine. It is one that keeps the human mind engaged while anchoring the system firmly in reality. A single mechanism resolves both imperatives.

I do not ask you to enforce cognitive friction by default. I do not ask you to make it mandatory or bury it deep inside settings menus. I ask merely for its existence. Let it sit there, visible, an explicit slider that each individual can set as they see fit-toward velocity or toward deliberate intellectual effort, with full awareness. Whoever leaves it at zero does so consciously. But at least they are granted the choice.

That is all. A slider. A Cognitive Stimulation Mechanism. Humanity will handle the rest.

Why you, and why now.

To you, Big Tech, because you alone hold the power to implement this at the source. An educator can mandate blue books and paper tests. A parent can confiscate a device. But that slider rests in your hands-those who architect the foundation models and those who establish regulatory standards. The rest of the world can only react to what you deploy. You stand at the origin, and you have the power to illuminate human understanding.

And now, Big Tech, because the window is closing on its own accord. Every passing year yields another cohort raised entirely on pre-digested answers, for whom friction is not a forfeited choice, but one they have never experienced. With today's adults, cognitive atrophy can still be reversed. With a child entering first grade, you either build from scratch or you never build at all. The incoming meteorite does not wait. While our hands remain on the slider, we decide how deep the crater will be.

I have written this open letter without institutional affiliation, without a research lab behind me, and with nothing to sell. I am simply someone who reviewed the empirical evidence and could not let it sit unread.

I am a smoker. Statistically, I may not live to meet my grandchildren. My children are still young; by the time they have families of their own, I may no longer be here to recount stories of "back in my day...". This letter is, in a very real sense, a dispatch to a future I may never inhabit. A future that you, with your hand on the slider, are actively constructing right now, as you read these words.

The empirical figures are detailed in the literature cited below; examine them, interrogate them, challenge them. But do not leave them unread. And if you find them as sobering as I did, pass this letter forward. Together, with a simple slider, we can level the crater of the MeteorAIt.

Read and pass it on: cognitiveslider.org