Scott Aaronson who is quite outspoken on his blog , coined the term “certified organic theorems” and mentioned that the time for such things are over. It is clear that in these days it has become hard to distinguish mathematics that is AI generated (the analog of growing genetically engineered, hormone driven food) and human generated math. In the climbing analogy, it is climbing a wall by pulling yourself up ropes or more close in analogy now in 2026, to just take the cable car or helicopter. There had been lots of discussions about the difficulty we face today with the new reality. Is the time of “organic human generated theorems” over?
I myself still maintain that “certified organic mathematics” still can be done, and it should be honored as such, similarly as a red point climb is honored more than a climb with technical tools like ladders. The problem is the certification. In sports, we have doping tests, in the food industry, there are inspections. Maybe we need a system of certification also in mathematics. And similarly as in sports or in the food metaphor, a culprit needs to be banned or at least exposed. There are motors now in bikes which can not be seen, there are blood doping methods which are very hard to detect. In climbing, everybody can just brag about having climbed a route free and we live in a time, where even video evidence needs to be checked for AI tempering. I myself have little appreciation for somebody who proves a theorem with a prompt. Granted, it needs patience, money and a bit of luck, but doing things organically with your own brain and no help is something which should be honored as such.
In mathematics, we might just ask authors of papers to present freely (of course not reading from a script) and answer questions about a problem or be able to discuss the topic without any electronic help. In rock climbing the analog is to see somebody live climbing a route. If you claim to have done Dawn Wall free solo, then you better can also free climb much simpler routes under observation.
In other words, “reputation” will become more and more important. While it will be impossible to check whether somebody has done something in every instance, it would just look bad, if some mathematician can not talk and discuss a subject that had been worked on. This means answering questions in person and not electronically. I can imagine that in the future, PhD candidates would be enclosed for a few days into a building that is a Faradey cage without any electronic contact and metal detector checked for electronic devices and that they have to answer without any tools questions about the PhD thesis.
This is not different in other areas. I personally find any video an insult that is not declared to be AI generated (if it is not obvious). Any email sent that is AI generated (and not declared as such) is an insult to the recipient. Any paper that is written AI generated and not declared as such, should be an insult to the math community. An assignment in a course written by AI and not declared as such is an insult to the student. Having papers graded by a machine also. Or administrative tasks done by the machine should be banned too, except if it is made announced like “we are using AI tools to grade assignments and AI tools to evaluate students”. I think it should be obvious that this is essentially the end of teaching. Teachers who do it (reportedly, many teachers already have their assignments written by AI), should have to make this transparent. Otherwise, it is just an insult.
On the student side, having assignments done with the machine also is fraud. We can of course not any more police it, but I think what will happen with time is that folks who rely on machines doing stuff will temporarily gain an advantage as we are still fresh with realizing what kind of new reality has hit us: you put yourself out of business much faster if you delegate your work away. And what happens if in a few years, the AI bubble bursts or data centers go up in flames due to riots, or the money for supporting these kind of operations is no more available? Humans who have delegated their thinking away will be completely stranded.
Consultants, lawers, doctors etc all still enjoy now the “power” of using machines for their tasks. But once it hits the “decision makers” ( machines soon?) that this “prompting work” can actually also be done even cheaper by machine, these folks will be put out of the market faster than anticipated. The reason is that the worst thing which can happen to you is to lose the ability to process information without help, to think without help and to communicate without help. The ability to guide a machine doing stuff is trivial in comparison to actually do the stuff. We should still be able to do all what we do by ourselves, just as a safe guard for the case when all these tools should disappear, fail or being taken away from us (without paying insane amount of money to get access). The drug dealers still give all this stuff away for free now. The enshitification of AI will come like the Amen in the church.
The difficulty now is to shield away the machines from a topic. Is asking google a question already too much help? What about having your paper spell checked? What about having it proof read? The problem with the later is that the machines happily offer now to rewrite the paper. But that essentially makes the machine already a coauthor. Maybe the threshold should be similar as we did in the past with coauthor ship. Acknowledge discussions with a machine like with a person, spell checking probably does not need to be declared. But having a paper rewritten crosses already a line and should be declared “non-organic”.
In the our current climate, the number of papers, things like h-index or citations will become less and less relevant and that personal reputation will start to become more important. The tricky question will be how to distribute jobs. It looks as if the result itself is no more important but how the result is achieved. It is no more the output but the quality and possibly the degree of “certified organic content” which could become more relevant.
P.S. The cartoon displayed above is certainly not organic. I make a habit asking the language model to sign a picture (if it is not obvious that the picture is AI generated). In that case I myself gave only the idea of the joke. I wanted a cucumber asking a strange looking apple whether it is certified organic. The rest was done by the machine, also the idea of the worm finding a bug. I like also the QED certification label and the farmers market setup, both of which had not been prompted.