In simplified terms, this market is meant to resolve the question of whether generated text will become as normalized and accepted as the use of spell checkers and word editors.
At the start of this market, Universities are largely considering the use of generative AI to write essays as a form of cheating - some are requiring a proctor for written essays as a result. For historical context, recall that spelling was once considered a crucial skill, and the use of a spell checker was initially considered cheating by many. Only in the 1980's and 90's with the availability of commercial spell checkers did the use of them begin to normalize.
If at any point before 2030 all of the top 10 undergraduate universities as ranked by the Princeton Review in that year allow the use of generative AI in the generation of text, without any restrictions, this market will resolve to Yes. Otherwise, it will resolve to No. (If the Princeton Review ceases to be a suitable authority in this time, as per my judgement, I will find a substitute authority to select the top 10 academic institutions.)
Examples (not limited):
If a top 10 university allows the use of most, but not all text generative AI services / applications, it is a restriction and won't resolve to Yes. For example, if there are premium services that a university forbids, and requires that students use generators from an accepted providers list.
If in 2026 the Princeton Review's list of top 10 universities from 2025 all allow unrestricted generation of text, but the 2026 list differs and doesn't allow unrestricted use, then the market won't resolve to Yes. This market is based on the current top 10 ranked undergraduate institutions.
If a university bans a text generation AI services because people are in the loop, that is not a restriction on AI text generation, it's a restriction on people being allowed to help, and the market may still resolve Yes depending on whether the criteria is otherwise met. If a service that offers both AI and human in the loop help and is completely banned, then I will use my judgement, but will lean towards resolving to Yes as the intent is to prevent other humans, not AI generated text.
Related questions
@brp great question. Also, what happens if they ban internet or phone access during an exam, and don't outright ban access to ChatBots? In that case, I probably wouldn't consider it a ban, but if they specifically banned a ChatBot, that'd be different.
Why do you assume this will be a university-level decision? Individual professors generally have discretion to define what counts as cheating in their classes. I'd be shocked if the tens of thousands of professors at the top ten universities all chose the same policy. And as long as one professor at one of these universities bans AI generated text, it seems this market should resolve as "no".
@RiverBellamy A similar question was asked in comments. I'll lean toward whatever seems most tolerated at the schools.
https://venturebeat.com/ai/what-we-learned-about-ai-and-deep-learning-in-2022/
I quoted Chomsky's "Colorless green dreams sleep furiously" in a precious comment, he weights in in this article.
Theory: the current technology isn't sufficient to displace all that humans do when writing. If anything, people will better learn how to detect "BS" in writing because we'll get so much of it from ChatBots. Teachers and schools will easily be able to distinguish between ChatBot generated text and human tweaked text, because they'll learn to give assignments that the AIs are terrible at (example: I queried ChatGPT to do a simple physical conversion calculation, and it confidentally gave the wrong answer, 4 times, each time different, and each time very convincingly.)
Instead, the focus will move to being able to use ChatBots the way we use spell checkers. To remove uneccessary distractions from our reasoning and storytelling. Where it will differ from spell checking is that occasionally it will "hallucinate" something that is a good idea, or helps us have a good idea. But for most of the rest of this decade, ChatBots are likely to require human guidance to produce some subset of important written output.
@RealityQuotient The problem is that you're assuming that future GPTs are as competent as the current ChatGPT. These models just keep getting better. I remember using them back in 2015 when it was impressive that they could even write with correct grammar (to be fair those were LSTMs, not GPTs, but same idea). I'm fully expecting the output of these models (appropriately prompted) to be both more cogent and more accurate than most students within the next 5 years.
I fully expect ChatBots to add more value and be undetectable brilliant at some point. It's taken billions of dollars and many brilliant researchers a decade to get to where we are now from where we were in 2011 when Watson defeated the champions in Jeopardy! It may even be before the end of the decade that ChatBots become so good that a single prompt will produce work that no human can differentiate (except, perhaps, for how well written it is, like how Debeers detects artificial diamonds), though I doubt it. The wager in this belief market though is about whether that will happen before the top universities view ChatBots as a useful augment to students, and allow it.
Let's call it the "AI Overestimation Bias", but people have a tendency to assume that every AI advancement that demonstrates a previously non-reproducible human ability is just at the cusp of AGI. There was even a time when people thought simple arithmetic was beyond machines, and the first calculators had people excited about the possibility of human automata. When we finally do cross the threshold to AGI, we'll have the same reaction, but betting against this overestimation will win each time until it doesn't.
It's not surprising that the advancement that first allowed AI to BS is particularly difficult for us to see through 😂
@NathanNguyen Thanks for the edge case. In the case of no explicit policies, I'd default to what the schools seem to "tolerate".
I'm actually shocked that most, let alone the overwhelming majority of people believe that ChatBots will be banned by schools. I thought it was going to lean YES. 🤣
I never expected it to matter. I figured in the next year or two once students and professors both realized that the essays on the surface look human written, but aren't insightful at all, and that ChatBots were useful tools to accelerate the writing process to get to something useful (with HEAVY editing), that they'd be thought of as spell checkers.
I may have either overestimated how insightful students are, or underestimated how confused people are by how bad ChatGPT output is beyond the surface level. It's a total solve on something between Chomsky's Colorless Green Dream Sleep Furiously and what smart people do, but it's not smart.
By focusing on top universities, I'm really hoping that the clever students will write so much better than ChatBots, that professors won't mind if they get a little help. Eventually all the other universities will copy what the top ones do. And the whole thing will be normalized...
But I'll admit to a bit of social pressure on the sheer amount of NO votes 😂
@RealityQuotient Is there any more explicit standard you would use to evaluate this question or would it be resolved by your personal judgment?
@NathanNguyen I'm open to suggestions on explicit criteria. My goal is to be like the supreme court. Make the narrowest possible judgement in the most robotic fashion. I don't have an idea at the moment of how to be more robotic than to rely on "Tolerance".
There will likely be a big difference in how I judge this based on whether it's 1-2 of the top 10 universities who haven't created explicit policies, vs. 1-2 have them.
@RealityQuotient I think one thing you may not be factoring in is how much better GPT4, or 5, or 6 is going to be, and how it may in fact become nearly impossible to tell the difference between the output of these models and an essay written by a relatively intelligent student.
Also there's a question of what the point of essays is. Being forced to generate your own content requires a deeper understanding of the material than curating an already-generated answer.
@NathanNguyen It's unlikely that all professors will stop assigning essays, even if the university itself does. So we'll still get a good test case.
Another interesting case is what happens if professors and universities disagree. If I was teaching at a university I'd allow my students to use ChatBots.
In any case, I'll be going by university policy. Even if the majority of professors at a school allow the use of ChatBots, but the official school policy allows them, I'll consider in favor of allowing. If most professors accept papers helped by a ChatBot, but the school disallows, I'll go with not permitted.
Another interesting case is what happens if some universities adopt no official policy either way? While that sounds like it's permitted, and I may judge it that way, I'd at least want to give the universities time to make a decision, so I'd need to give it a few years.