Yesterday, I had one of those moments that feels small in real time and larger a few hours later. My five-year-old was talking to ChatGPT in voice mode. At first, it sounded like what you would expect from a child who has discovered that a machine talks back: Japan, then dinosaurs, then planets. But the more I listened, the more I realized that this was not just a cute parenting anecdote. It was a very clear signal that The End of Information Advantage is not some abstract business trend. It is already here, in the living room, speaking out loud.
He asked how many bridges there are in Japan. That is the kind of question that used to break the old information model. Not because the answer was impossible, but because the answer was awkward. Google would likely give you scattered pages, tourism results, infrastructure reports, and a lot of low-confidence noise. Wikipedia would probably not have a neat page called Number of bridges in Japan. A child asking a broad, fuzzy, innocent question would often hit the limits of search very quickly, not because curiosity had failed, but because the question did not fit neatly into the old retrieval machine.
This time, the question itself mattered less. The system could respond contextually. It could explain that there is no clean universal number, that Japan has an enormous amount of infrastructure, that the answer depends on what counts as a bridge, and that the country likely has hundreds of thousands when road, rail, pedestrian, local, and small structures are included. In other words, it could do something that old information systems often failed to do well: help a curious person move forward even when the question was incomplete, oddly shaped, or not indexed in exactly the right way.
Then he asked what its favorite planet was. Again, this is not a question designed for search. Nobody wants to Google what is the favorite planet of most people in our solar system just to satisfy a five-year-old’s curiosity, unless they enjoy turning childhood wonder into a procurement process. A conversational model can simply say Saturn because of the rings, and the interaction continues. The conversation does not collapse just because the question does not fit the structure of a search engine or encyclopedia.
That was the moment that stayed with me. We already had the internet, Wikipedia, and search, so in one sense, information had already been democratized. Knowing things had been losing scarcity value for years. But something is different now. The difference is not only that answers are easier to access. The difference is that exploration itself has changed shape.
I have written before about how AI is changing the middle of my work, not just the beginning or the end, and how AI has lowered the cost of starting but raised the standard for finishing. This feels like another piece of the same larger shift. We are moving from a world where advantage often came from stored knowledge and better retrieval, toward a world where advantage comes increasingly from how well you can frame, explore, verify, and continue thinking. That shift is much more personal than the abstract statement that AI will change the economy, because it changes what it means to not know something.
The End of Information Advantage Did Not Start with AI
To be clear, AI did not suddenly destroy a perfect old world where experts alone had access to knowledge. That world was already gone. Once search engines matured and Wikipedia became normal, most basic information asymmetries had already started collapsing. If one person knew the capital of Japan and another did not, that was no longer a durable advantage. If one person knew a historical date, a scientific fact, or a broad concept, that was useful, but increasingly easy for others to recover. The internet had already weakened the value of possessing isolated facts.
But the internet still had a shape, and that shape rewarded a certain kind of user behavior. You had to know how to search. You had to reformulate. You had to distinguish between good and bad results. You had to infer when the answer did not exist in the exact form you wanted. You had to stitch together fragments across pages, forums, manuals, Reddit threads, documentation, and whatever suspicious blog post happened to rank well that day. Search gave you access to information, but it still demanded that you do a lot of navigation, translation, and synthesis yourself.
That is why I do not think Wikipedia ended information advantage in the way people sometimes imply. Wikipedia flattened reference knowledge, while generative AI flattens exploration friction. Those are not the same thing. One gives you access to documented knowledge. The other helps you move through uncertainty when the question is still messy, the language is incomplete, and the problem has not yet been named properly.
This means people can ask worse questions and still get somewhere useful. They can ask follow-up questions without starting over. They can stay in the conversation instead of learning the rituals of search. A child can ask about bridges in Japan, and a professional can paste in a broken Excel formula or upload a screenshot of a confusing interface and continue from there instead of first becoming a librarian, a search strategist, and a detective. Somewhere, a search engine optimization consultant just felt a disturbance in the Force.
This changes the cognitive experience of missing information, and in many cases it changes the emotional experience too. Not knowing something used to create a larger gap between intention and progress. Now, very often, it creates a conversation. That conversation may be flawed, incomplete, or overconfident, but the starting point has changed from go find the answer to let us explore the problem together.
What The End of Information Advantage Actually Changes
The deepest shift is not that people can get answers faster. The deeper shift is that the boundary between I do not know and I can move forward anyway has become dramatically thinner, and that matters in both personal and professional life.
A funny example from my own work is ice-breakers for workshops. This is not exactly the stuff of civilizational history, but it is a good example of how knowledge friction used to work. You might want recommendations for an ice-breaker tailored to a specific audience, a specific energy level, a specific room setup, and a specific workshop objective. Search could give you lists, generic facilitation advice, and an impressive collection of activities that make adults pretend they enjoy passing imaginary balls around the room. But it was much harder to ask for something that actually matched the situation in front of you.
Now you can say that you need an ice-breaker for a skeptical leadership audience, with low patience, a 15-minute window, a hybrid setup, and a connection to collaboration instead of small talk. That is a different category of help. The system does not merely retrieve facilitation content. It adapts the shape of the recommendation to the situation you are dealing with.
The same is true for something more serious, like fixing a broken Excel formula. Previously, solving that kind of problem often required knowing what you were looking for in the first place. You needed the right vocabulary, the right search terms, and quite a bit of luck. Maybe someone on Stack Overflow had the same issue. Maybe someone on Reddit had written about it. Maybe the problem was buried in a forum thread from 2018 with one reply and no conclusion, because naturally the internet enjoys dangling partial solutions in front of desperate people like a tiny productivity horror movie.
The burden was not just on finding the answer. The burden was on locating the problem inside the right conceptual box. Generative AI changes this because you can now show the mess directly. You can paste the formula, upload the screenshot, describe the intended result badly, and be imprecise at first. The system can often still help you move toward the right diagnosis, which means information advantage is no longer only about having prior knowledge. It is increasingly about how fast you can externalize a problem, interrogate it, and refine your way toward a solution.
In other words, the new leverage is not always in memory. It is in interactive problem framing. This is especially obvious in smaller ecosystems. I use Coda heavily, and Coda has its own formula language. It does not have Excel-scale community support. So the old model breaks more easily. Search becomes weaker when the ecosystem is smaller, the community thinner, and the documentation less complete. In that kind of environment, generative AI becomes much more valuable because it can interpret intent, notice missing parentheses, reason through syntax, and help close the gap between this seems trivial and why is this not working.
That is not just speed. It is a different problem-solving posture. You are no longer limited to whether someone else has asked your exact question, in your exact context, using the exact vocabulary that the search engine will reward. You can start with the actual mess in front of you.
From Searching to Steering
I think this is the real personal shift. The old information world rewarded search skill and patience, while the new one increasingly rewards steering. You ask, inspect the answer, follow up, correct the framing, push deeper, and keep going until you are satisfied. The loop matters more than the first question.
Your first question can be clumsy. It can be incomplete. It can be slightly wrong. In the old model, that often meant you would get bad results or no results. In the new model, the system can help you improve the question while answering it. That is a subtle but massive change, because it lowers the penalty for not yet knowing the shape of what you are trying to find.
My child captured this better than most strategy decks could: the questions became less important. For a long time, one hidden privilege in knowledge work was being good at turning messy human uncertainty into something search engines or experts could process. If you knew how to ask the right question, knew the right person, or knew the right terminology, you could move faster than others. There was real advantage in knowing how to begin the search.
Now the beginning matters less. You can start with the wrong framing and still recover. That changes who gets to participate confidently. It lowers the penalty for imprecision, unfamiliarity, missing jargon, and being early in your learning curve. That is one reason I suspect AI will matter so much for children, junior employees, career switchers, and people operating across unfamiliar domains. Not because they suddenly become experts, because they do not. But because the cost of entering the conversation drops.
Once entry becomes easier, persistence matters more. This is also why multimodality matters more than people sometimes think. Screenshots are a perfect example. Saying that screenshots of broken things are a game-changer sounds like a practical observation, but it points to a bigger truth. When you can show the problem instead of first translating it into the exact right textual description, you remove one more layer of friction between confusion and progress.
That is where the interface tax starts to become visible. The system is no longer just waiting for perfect inputs. It helps absorb messy ones. And once systems can absorb messier inputs, the old advantage of being the person who knows how to navigate information systems manually begins to shrink.
Why This Does Not Mean Expertise Is Dead
This is the part where people get silly, because if access to answers improves, some immediately jump to the conclusion that expertise stops mattering. I think the opposite is happening. Information advantage is shrinking, but interpretive advantage is not.
Pattern recognition still matters enormously. You described this well in organizational terms: walking into a company, observing how people behave, noticing what they struggle with, hearing what the coffee gossip is really about, and recognizing a pattern that no one has named cleanly yet. That is not the same as knowing information. That is sensing structure inside a messy environment, and a master’s class can teach frameworks without fully teaching that kind of live pattern recognition.
This is why I do not think generative AI eliminates the value of experience. It changes where experience pays off. The person who simply knows facts becomes less distinctive, while the person who can ask sharper follow-ups becomes more valuable. The person who can tell whether an answer fits the real situation becomes more valuable. The person who can spot when the system is answering the wrong question becomes more valuable. The person who can connect weak signals across human behavior, constraints, incentives, and informal power becomes much more valuable.
This is especially important in professional settings because many real problems are not information problems at all. They are interpretation problems, diagnosis problems, coordination problems, political problems, and trust problems. Generative AI can help with all of these at the edges, but it does not dissolve them into pure information retrieval. It just makes it harder to hide behind the idea that the information was inaccessible.
That is the uncomfortable exposure. Once everyone can retrieve, generate, and synthesize at a decent baseline level, the remaining gap becomes harder to fake. People who used to rely on information possession may suddenly discover that the harder part was never the possession. It was judgment.
The Danger Is Not Ignorance. It Is Synthetic Confidence
There is, of course, a darker side to all this. If the penalty for asking bad questions goes down, the penalty for accepting bad answers can also go down. This is where your neighbor story is unexpectedly useful, because humanity has apparently decided that even street lamps can become a case study in epistemology.
A dispute over street lamps and electricity costs turned nonconstructive because one person used AI to calculate the yearly cost badly, then treated the output as decisive proof that the other person was being cheap. The information looked legitimate enough to support outrage, even though the underlying calculation was wrong. That is funny in a neighborhood-drama way, but much less funny as a societal pattern.
This is what happens when information friction drops faster than epistemic discipline rises. People feel more informed without necessarily becoming more accurate. They feel more capable of argument without necessarily becoming more grounded in reality. They can produce numbers, interpretations, and explanations more easily, but they do not automatically become better judges of whether those outputs deserve trust.
This is where some of the research gets uncomfortable in useful ways. Microsoft researchers reported in a 2025 study of 319 knowledge workers that higher confidence in generative AI was associated with less critical thinking, while higher self-confidence was associated with more critical thinking. They also found that critical thinking shifts toward verification, integration, and task stewardship rather than disappearing altogether (Microsoft Research, 2025). Their broader Tools for Thought framing is useful too: the challenge is not only getting AI to answer, but getting humans to stay cognitively engaged enough to challenge, refine, and verify what they receive (Microsoft Research blog, 2025).
That feels exactly right. The End of Information Advantage does not produce universal wisdom. It produces a new split between people who treat AI as a collaborator in thinking and people who treat it as a confidence vending machine. Those are not the same users, and over time, they will not get the same results.
The New Advantage Is Follow-Through
If I had to name the new advantage in one phrase, I would say this: the new advantage is not knowing more, but staying with the question longer. That begins with better questions, but it also includes better follow-ups, better skepticism, better synthesis, and better stopping conditions.
You keep asking until you are satisfied. You keep refining until the answer survives contact with reality. You keep testing until the explanation matches the actual problem. You keep checking until sounds plausible becomes I trust this enough to use it. This is a different skill set than traditional information privilege, because traditional information advantage often rewarded possession while this new environment rewards interaction.
I think that has interesting consequences for both adults and children. A child in voice mode does not need to master the rituals of search before being allowed into curiosity. That is powerful, but it also means the surrounding adults matter more, not less. If the child grows up in an environment where AI answers are treated as conversational starting points, curiosity may deepen. If they grow up treating fluent outputs as truth by default, then we are raising a generation with easier access to language and weaker habits of verification. Because apparently every miracle must arrive bundled with a maintenance contract.
The same is true at work. One thing the field evidence on generative AI suggests is that tools often help less experienced workers the most in concrete tasks. The well-known field study by Brynjolfsson, Li, and Raymond found productivity gains in customer support, with especially strong effects for lower-skilled and less experienced workers, partly because the system helped transfer patterns that previously lived mostly in stronger performers (Quarterly Journal of Economics, 2025). That is important because it means AI can compress the gap between novice and competent performance in some contexts, but compressed access to patterns is not the same as fully internalized judgment.
That distinction matters because the true advantage is shifting from access to use. The people who will gain the most are not necessarily the ones who receive the first answer fastest, but the ones who know what to do with the answer once it appears.
What My Five-Year-Old Was Actually Showing Me
The surprising thing about yesterday was not that a child could ask an AI about Japan, dinosaurs, and planets. The surprising thing was that the interaction made the old model look clumsy. Not wrong, not useless, but clumsy.
The old model asked the user to adapt to the information system, while the new model increasingly adapts the information system to the user. That shift sounds technical, but it is actually philosophical. It changes who gets to feel capable, how quickly curiosity can stay alive, how much missing vocabulary blocks progress, what the emotional cost of not knowing feels like, and what counts as being good with information. It also changes where expertise still holds and where it gets exposed as mostly jargon plus retrieval.
Most importantly, it changes what advantage now looks like. I do not think the future belongs to the person who merely knows more facts than everyone else. I think it belongs more to the person who can frame a problem, ask a better follow-up, notice when the answer is shallow, connect it to reality, and keep going long enough to get somewhere useful.
That is true in parenting, Excel, Coda, workshop design, organizational diagnosis, and probably most of knowledge work. The End of Information Advantage does not mean information stops mattering. It means that information alone matters less than it used to, and what matters more now is what you do when the first answer arrives.
That is why I do not think this is really a story about AI replacing knowledge. I think it is a story about AI changing the shape of curiosity, lowering the friction of exploration, and exposing a new kind of human advantage in the process.
Not who can retrieve the answer fastest, but who can think with the answer best.
