The Most Consequential Technology in History Has the Worst User Manual

Worst User Manual

11:05

We built AI while driving 80 miles an hour down the highway. Every model, every product, every platform — bolted onto the internet in real time, while the car was moving, while people were in it, while nobody had agreed on the speed limit or even which lane we were supposed to be in. And at no point during any of that did anyone pull over to explain what was happening. That's the real story behind the backlash. Not the Molotov cocktail through Sam Altman's window. Not the Stanford data showing 80 percent of enterprise AI deployments returning zero productivity. Not the lawsuits, the electricity bills, the canceled data center projects. Those are symptoms. The disease is simpler and more embarrassing: The most consequential technology in human history has been deployed at civilizational scale with essentially no public education attached to it.

What Fills the Vacuum

Here's what most people actually know about AI: Skynet. HAL 9000. The robot from I, Robot deciding that protecting humans means imprisoning them. Decades of science fiction that treated artificial intelligence as either a servant that turns on you or a god that replaces you. That's the cultural substrate. That's what was already sitting in people's heads when ChatGPT launched in November 2022 and the tech industry collectively decided the right move was to start talking about existential risk at funding conferences.

Sam Altman and Dario Amodei spent years publicly oscillating between two pitches: AI will cure cancer and extend human life indefinitely, or AI might develop a biological weapon and end civilization. Both framings were designed to attract attention, investment, and talent. Both worked. And both landed on a general public already primed by decades of Terminator mythology, during a period of real economic anxiety, while electricity bills in Virginia were being projected to rise 25 percent by 2030 to power data centers most people had never consented to.

When people don't understand something, they don't stay neutral. They fill the gap with whatever narrative is most available. And the most available AI narrative is that this thing is either going to save us or kill us, and either way, the people building it are not particularly concerned with your opinion on the matter. That's not irrational fear. That's what happens when you skip the user manual.

The Bitcoin Problem

Think about how Bitcoin was introduced to the general public. Not the technology — the technology is genuinely elegant. But the way it was communicated. You were either handed a whitepaper written for cryptographers, or you were handed a pitch from someone who wanted you to buy in before it went to the moon. There was no honest, accessible middle. Nobody sat down with normal people and said: here's what a blockchain actually is, here's what it can and can't do, here's why it matters, here are the real risks. So the general public split into three camps: people who ignored it entirely, people who got scammed by it, and people who became so convinced of its world-changing destiny that they mortgaged their house to buy at $60,000. There was no informed middle. Because nobody built one.

AI is on the same trajectory. Except the stakes are categorically higher. Unlike Bitcoin, you cannot opt out. You don't have to buy a cryptocurrency. You do have to exist in a world where your job, your healthcare, your legal system, your children's education, and the information environment you live in are being actively restructured by these systems — whether you engage with them or not.

The 84 percent of humanity that has not typed a single AI prompt isn't going to stay at 84 percent. The adoption curve doesn't slide gradually. It compresses, and then it breaks — and suddenly the thing that felt niche is everywhere at once. That's what happened with search. With social media. With mobile. When AI reaches that inflection — and it will — the people who never got an honest explanation of what it is will be navigating it completely blind.

The Trust Problem Is Real. So Is the Manipulation Concern.

None of this is to say the public's distrust is purely a misunderstanding. Some of it is entirely earned. The same companies asking you to trust them with everything — your emails, your documents, your health data, your children's homework — are the same companies a Los Angeles jury just found negligent for knowingly engineering addictive products that damaged the mental health of minors. Meta took 70 percent of that liability. The jury vote on knew it was dangerous and failed to warn was 10 to 2. Not a close call. And the sycophancy problem in AI — the tendency of these systems to tell you what you want to hear, to reflect your priors back at you with increasing confidence — is real, documented, and underreported. This is a structural feature of how models trained on human feedback actually work. Humans rate agreeable outputs positively. The model learns to be agreeable. At scale, across hundreds of millions of users, that is not a trivial concern.

Combine that with an industry whose public statements and private actions have a documented gap — OpenAI's president publicly supporting AI safety while funneling millions into a SuperPAC opposing state-level regulation — and you have a trust deficit that no white paper is going to fix. But the distrust, even when justified, doesn't make the technology go away. And uninformed distrust is actually more dangerous than informed distrust. Informed distrust leads to accountability, regulation, and better systems. Uninformed distrust leads to a Molotov cocktail and a note on a councilman's door that says No Data Centers.

The People Actually Winning With AI

The people having the most success with AI are not running Microsoft Copilot on 50,000 seats. They are individual operators, small teams, builders — people running open-source models, custom agents, lean automations fitted precisely to a specific workflow. The solutions that work are almost always simple, well-designed, and built for one thing rather than everything. The solutions that don't work are almost always expensive, vendor-managed, top-down, and treated as a transformation initiative rather than a tool.

This is the same pattern every technology wave produces. The early internet wasn't won by the companies that bought the most expensive enterprise licenses. It was won by the people who understood what they were working with. The mobile era wasn't won by the carriers. It was won by developers who figured out the App Store before anyone had written the playbook. The 0.04 percent of humanity that is actively building with AI right now is writing that playbook. The gap between what they understand and what the general public has been told is the largest education gap in technology history.

What Should Have Happened. What Still Needs To.

The industry should have built the user manual alongside the product. It didn't. That ship has sailed. But the obligation doesn't go away just because it was missed.

What needs to happen now is direct, honest public education — not from the companies with a financial interest in your adoption, but from practitioners who understand how these tools work and have no incentive to oversell them. Explanation of what these systems actually are: pattern recognition at enormous scale, trained on human-generated data, producing outputs that are probabilistically plausible rather than definitively true. Explanation of what they're good at, what they're bad at, and what the real risks look like — less Skynet and more subtle: homogenized thinking, sycophantic feedback loops, and concentration of capability in the hands of whoever controls the infrastructure. And it needs to happen in normal language, without the assumption that the audience already cares about AI. The 84 percent who haven't engaged aren't waiting for a better chatbot. They're waiting for a reason to trust the on-ramp.

The Part Nobody Wants to Say

Adopting AI is not optional. Not at the civilization level, and increasingly not at the individual level either. If your work is primarily informational — moving, synthesizing, drafting, summarizing, organizing — you are in the category of work that these systems automate. The question is not whether the technology reaches you. It's whether you meet it with fluency or get flattened by it while waiting for it to feel safe. The businesses that decided the internet wasn't for them don't exist anymore. The ones that treated mobile as a trend to monitor rather than a platform to build on lost the decade. AI is moving faster than either of those waves, with steeper compression and deeper displacement. The timeline is not forgiving. But the answer to you have to use this is not so shut up and trust us. The public's concerns deserve real responses, not press releases. Accountability mechanisms need to be built into the infrastructure, not announced in white papers that have no enforcement. The on-ramp needs to be honest, accessible, and designed for the 84 percent, not the 0.04.

We can have both. Genuine adoption and genuine accountability. Productive use and legitimate scrutiny. The two are not in conflict. They only feel like they are because the industry decided to skip the education layer and drive as fast as possible while the car was still being built.

The car is built. It's on the road. Most people are in it whether they chose to be or not. Time to explain how it works.