• by Steven Rosenbaum , Featured Contributor, over 1 hour ago

Two of the internet's core institutions made decisions about AI this spring.  And interestingly, they went in totally opposite directions. This is a critical conversation about how information gets on the internet. 

On March 20, Wikipedia's volunteer English editors voted to ban large language models from generating or rewriting article content, as Wikipedia itself reported.

Wow, that’s a big ban. The vote was 44 to 2. And it didn’t come from the top down; it was Wikipedians voting. Kaboom! Wikipedia will be human-powered, even if that means more work for the volunteer-driven organization. 

But, at almost the same time, Google essentially put a death knell to what has been the underlying basis of the web: search. Google is now an AI assistant. Type a question now and you get an answer, synthesized and handed back to you. The ten blue links, the index, the choice, the act of looking,  are replaced by AI. 

It’s a very different approach. Wikipedia drew a line. Google erased one.

The line Wikipedia drew

I first saw this news framed by Josie Wood, an AI system designer who posts as @josieandthe_ai, where she translates fast-moving AI developments for a general audience under a simple banner: She keeps up so you don't have to. 

Her read on the Wikipedia vote was a tonic. She called it a reason she stays a tech optimist: people signing up for what is effectively more work to keep humans in the loop where it counts.

Wikipedia runs on trust. Volunteers write it, source it, and check each other's work. It carries no ads. For a generation taught never to cite it, that's an irony worth sitting with: It has become one of the last large-scale sources of neutral public information.

The editors tried AI and found what everyone finds: articles that read clean and cited sources that did not exist. Hallucinations. Facts asserted with confidence and no basis. The problem was not that the text looked bad. The problem was that it looked way too good. 

One person can generate AI text in seconds. Verifying it by hand takes hours. Wikipedia called it an asymmetry of effort: cheap to produce, expensive to check. At scale, that math breaks Wikipedia, a system built on human verification.

Editors can still use a model to copyedit their own writing or rough out a translation, as long as a human checks the result. The base text of an article must be written by a person.

There is a second reason this really matters, and it should keep you up at night. Wikipedia is among the most important training sets on earth. Hallucinated text that slips into an article gets scraped, folded into the next model, and re-emerges as something the machine now "knows." What began as something that was wrong finds itself somehow a trusted Wikipedia fact. It’s like money laundering, making the AI words appear real to the outside world, and that’s bad.

So Wikipedia drew the line, partly an act of self-defense for the entire information supply chain. This at a time when Wikipedia itself is in real danger. As a Wired article notes: “As the free online encyclopedia turns 25, it’s facing political opposition, AI scraping, dwindling volunteers, and a public that may no longer believe in its ideals.”

The line Google erased

The sharpest account of what Google did comes from Emily Tavoulareas, who teaches technology and public policy at Georgetown's McCourt School and was a senior policy advisor to the White House CTO. She knows the difference between what technology promises and what it does in practice. That's her whole field. On her Instagram account, her warning was blunt: Stop treating Google like search. “Google is no longer a search engine,” she said, but an assistant that hands you its answer.

Her library analogy explains the change.  “Search was walking into the building, browsing the stacks, finding your own way. Now someone meets you at the door, takes your question, and won't let you in. They just hand you a reply.”

Sit with how big a change that is. Search indexed the world and let you choose. The work of judgment stayed with you.

Now, an AI assistant hands you an answer. You don't see what was weighed or what was left out. You get a synthesis, delivered with the smooth confidence of a thing that cannot tell you why it chose this framing over another.

This is not a small upgrade. It is a change in who does the thinking. For anyone still learning how to write -- which is to say, every student -- the muscle that search used to engage and question is no longer engaged..

The question both answer

Strip away the difference in scale and resources, and Wikipedia and Google were answering the same question: When a machine can produce fluent, plausible text on demand, what do you owe the people who will trust it?

Wikipedia determined you owe them a human who can vouch for it. But Google determined instead that we owe them speed.

I'll admit which answer I find more durable. But the more useful point is that the answer is still being written, institution by institution, and most of these have shareholders and employ people with titles and salaries. Wikipedia's editors had none of those -- but they moved first, and they moved clearly.

That should tell us something about where the clear thinking is coming from. And it should tell us that the line can be drawn, that tech is not inevitable. Forty-four Wikipedia volunteers just proved it.