Wikipedia’s volunteer editors recently banned the use of large-scale language types to reproduce or rewrite articles. Gartner reported that 53% of US consumers do not trust AI-powered search results, and 61% want to turn off abbreviations. Add the “Human Made” badges sprouting from Substack, and a consensus seems to be forming: humans are rejecting AI content.
I get it.
In 2024, I founded TrueMedia.org to fight political deepfakes; as a professor, I’m thinking hard about the possible decline of AI-generated content and cognitive surrender (the practice of letting the model do the thinking for you). However, my concern is with the output and the downstream effect, not with the input process. A deepfake is harmful because it is deceptive; the polished paragraph is not tarnished because the model was reinforcing it.
Therefore, the anti-AI-content movement, like the anti-GMO used movement, is missing the boat.
Back in 2017, researchers warned that AI risks a “GM-style backlash.” They had the right to be matched. They just bet the wrong half. I foresee the anti-AI-content movement going where the anti-GMO movement went: a very open action, a long taper, and a quiet end where the product is everywhere.
‘Frankenfood’ Lessons
In 1992, an English professor named Paul Lewis coined the term “Frankenfood” in a letter to the New York Times. In the late 1990s, Greenpeace was building an entire campaign around the metaphor, Prince Charles was lobbying Tony Blair, and the European Union was imposing a new moratorium on GMO approvals that lasted from 1998 to 2004. American consumers were told that they were eating animals. Labeling battles at the federal level have consumed a decade.
Look how far we’ve come. By 2025, herbicide-resistant soybeans will account for 96% of US soybean acres, up from 17% in 1997. Maize is 92% resistant to weeds. Cotton is 93%. The National Bioengineered Food Disclosure Standard went into effect in January 2022, and Cornell researchers, analyzing data from Nielsen’s scanner, found that it did not actually bring about a change in behavior. Shoulders together. Mandatory labeling, a central demand of the activist movement for two decades, proved futile when it arrived.
European attitudes follow a similar arc, albeit more slowly. Eurobarometer concern about GMOs in food decreased from 63% in 2005 to 27% in 2019. It ended with people losing interest. Today, most people have never heard of Frankenfood.
Why have GMOs won the long game? Three reasons map almost directly to AI-generated content.
First, the product is invisible. No one can tell that the corn syrup in their soda came from bioengineered corn, and after a while they stop wondering. AI-written prose has already passed the Turing limit of readability. Most students can’t tell the LLM draft from a competent person.
Second, economics is decisive. GMO seeds offered higher yields and lower input costs, so farmers adopted them and grocery chains stocked the resulting products. AI-generated content is almost free to produce. The supply curve has shifted to such an extent that purist restraint is no longer a market option; it’s a hobby.
Third, a minority of concern is acquired through voluntary labeling. The Non-GMO Project certifies more than 50,000 products for consumers who care. The mandatory organization label was no longer valid when it arrived. AI equivalents are already emerging: C2PA proofs, “human-written” proofs, stack verification marks. A dedicated few will have their own channels. Everyone else won’t bother checking.
Market solutions for real injuries
GMO crops cross-pollinated neighboring fields whether the neighbor wanted them or not. An AI script can do the same to the training data of the following model: today’s output becomes tomorrow’s input, without human consent and no clear way out. This is the downfall of the model: the worry that the supply will get worse over time rather than better, as synthetic text crowds out the human-written corpus.
The market is already solving this problem. Every major lab now pays for human-written content because they recognize the risks.
The GMO scare has also produced its share of catastrophic situations: a gene escaping into the wild, an accidentally created HIV virus, a collapse in food supplies. None of them happened. Markets were fixed, regulators were educated, shelters were planted, pollution was controlled.
Not all worries were overblown. The consolidation of seed markets became a reality, the Roundup case continues, and the overuse of herbicides is a living agricultural problem. None of this alarmed the Frankenfood campaign.
A common AI fear is that synthetic text will overwhelm the human corpus and that we will drown in a sea of AI slop. It’s in the same category: it’s clear, sounds mechanical at first, and eventually succumbs to the same boring energy. Students read the selected text. Publishers go into their archives. Moral values are emerging. The state of civilization is the part that cannot communicate with the real market.
Not all AI concerns are overblown either. NewsGuard has identified more than 3,000 AI content farm sites that churn out fake local news and propaganda for ad revenue, in 16 languages. Deepfakes deceive voters in real elections. The output damage is real. Thus the solution: authentication and gatekeeping. The same tools we already use against malicious content of any origin.
The Wikipedia ban, in this case, is a Greenpeace moment rather than a market decision. It’s the strongest sign of a place that cares, and it’s also the least representative of how the other 99 percent of students behave. The encyclopedia has already drawn the exception, as all internet policy eventually does. Translations from Wikipedias in other languages and basic copying of the editor’s prose are permitted by policy on day one. The carve-outs will expand from there: accessibility rewrites, citation formatting, draft scaffolding for new editors in poorly maintained languages.
The concerns are legitimate but also typical of early concerns with many powerful technologies. Five years from now, Gartner’s question will probably read differently because the product will be better and the innovation will have worn off. Watermarking will be more important when the stakes are high (elections, courts, financial disclosures), and less important for everyday reading. The slop will be filtered out, good AI writing will come together, and most people who say they won’t read it will read it without thinking too much about it.
Frankenfood became corn syrup. Residents will put down their torches when the lights come on.
