AI Pullback Has Officially Started
People are beginning to confront the reality of AI, and they are not happy.
The gulf between actual AI performance and AI hype is deep and cavernous, and we have the data to prove it. A recent MIT report found that 95% of AI pilots didn’t increase a company’s profit or productivity. A recent METR report also found that AI coding tools actually slow developers down. Why? Well, generative AI models, even the very latest ones, often get things wrong and “hallucinate”, which requires considerable human oversight to correct. IT consultants Gartner attempted to quantify this and found that AI agents fail to complete office tasks around 70% of the time. Simply put, the amount of human oversight necessary, even for simple tasks, almost always undermines whatever productivity gains are made. In other words, in the vast majority of cases, it is more productive not to use AI than to use AI. Yet despite all the evidence, AI is still being shoehorned in everywhere and being praised as the next industrial revolution. Or is it? Because there is also mounting data that the world is beginning to turn its back on this questionable technology.
Possibly the best example of this is a new report from Wiley.
This is their second ExplanAItions study, which aims to assess how AI is being used and perceived in the academic world. The differences between this year’s report and last year’s are shocking.
Back in 2024, 54% of researchers used AI — that figure jumped up to 84% this year. But, while 53% of researchers in 2024 said that AI was already exceeding human capabilities in the use cases Wiley tested, that figure fell to less than a third, which is a considerable drop-off. In 2024, only 51% of researchers were worried about hallucinations (also known as errors); that figure rose to 64% in 2025, despite new supposedly better models coming out.
The report concludes that researchers are “coming to a better understanding of AI’s present limits and future potential” and are seriously reassessing how they use it and plan to use it.
Now, Wiley did acknowledge that the 2025 study contained half of the same respondents as in their 2024 study, suggesting that there may be some selection bias here. But this could also be explained by researchers moving away from using AI, leading to fewer people responding as they have dropped the technology. Regardless, there are plenty of other studies that support the 2025 study’s conclusion.
For example, recent analysis has found that the AI adoption rate in large companies (250 employees or more) has dropped. Earlier this year, it was 14%; now it has dropped to 12%, which is not an insignificant slowdown, considering this isn’t reflected in the stock market and media hype around AI.
This slowdown is made worse by the fact that the number of companies scrapping their AI initiatives has skyrocketed. Recent analysis found that back in 2024, only 17% were cancelled, meaning real-world ongoing use of AI was climbing fast. However, this year, the rate has surged to a staggering 42%. This, combined with the slowdown in AI adoption, means real-world corporate use of AI is witnessing a marked decline.
It isn’t surprising, as we are seeing key players getting hurt by their reckless AI adoption.
Consulting firm Deloitte has been compelled to provide a refund to the Australian government over a $440,000 report, which it used AI to produce, which contained serious errors. These weren’t small errors like grammar mistakes; the AI had ‘hallucinated’ fake data to fill in gaps and misinterpreted data, rendering the report useless and, at best, potentially damaging.
So, from data to anecdotes, we can see the tide is turning across the globe. The shine has tarnished, and everyone can see that this might actually be a bit of a turd after all.
But why are we in this situation? Why did the world rapidly adopt AI? And why are they turning their back on it now?
Well, researchers from Melbourne may have the perfect explanation.
They found that AI only increases productivity in “low-skill” tasks, such as taking meeting notes or providing customer service. Here, they found that AI can help smooth the outputs of workers who may have poor language skills or are learning new tasks.
For higher-skilled jobs where accuracy is essential, AIs (even cutting-edge ones) make errors so frequently that the extensive human oversight required to catch them makes the entire effort less productive than not using AI at all.
The problem here? The workers who would benefit the most from AI (i.e., “low-skill” workers) don’t possess the skill or awareness to oversee AI and identify and correct its frequent mistakes. So, even though it “improves productivity”, potentially critical errors go unnoticed.
This explains why corporations like Deloitte incorrectly thought AI was such a powerful tool. By augmenting “low-skill” workers or blindly automating “low-skill” tasks with AI, they saw an increase in productivity. But this implementation lacks the feedback required to catch the frequent critical errors the AI makes. So they were left majorly blind to its faults.
It’s only when one of these errors makes itself known, such as with Deloitte’s report, that they actually have the information they need to realise that AI isn’t the productivity machine they thought it was. And, because AI doesn’t improve productivity for the majority of use cases (MIT report), this leads to most AI programs being cancelled.
Sadly, it takes time for these errors to be spotted. In the corporate world, a project utilising AI could easily take many months to complete and be implemented. So, because AI adoption peaked at the end of last year, we would expect these errors to be noticed around about now.
But, to bring this article full circle, the situation is slightly different in academia.
It wasn’t just researchers using AI to write their papers; those conducting the peer review also used AI to assess these papers. This meant that AI errors were not being picked up, as most research papers don’t spawn real-world implementations in which the errors become apparent. This has created an enormous scandal, with thousands of obviously fake AI-generated papers being published in journals, and authors hiding AI prompts in their papers to get the peer review AI to give them a false good grade. This is likely why many researchers thought AI was brilliant, as AI also compromised the academic feedback loop that should have caught their AI’s critical mistakes.
To stem the backlash, many journals and universities are starting to resist or have stopped using AI altogether in the peer review process. This has made it much harder for papers which utilise AI in their creation to get peer-reviewed, as the errors are now being picked up. So, just as in the corporate world, they finally have the awareness to see the reality of AI and are not just pessimistic about its potential but are more worried about the risks of ‘hallucinations’. This explains the findings in the Wiley report pretty damn well.
That is the AI pullback, why it is happening, and why it is happening now.
I know it is a tired cliché at this point, but it really is an emperor’s clothes moment, which is deeply worrying considering how dangerous the AI bubble is right now (read more here). Could this alone pop the bubble? Truth be told, I don’t know. But either way, none of this is a good sign for AI.
Thanks for reading! Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!
Sources: The Register, The Register, The Register, The Conversation, TC, Will Lockett, The Guardian, Fortune, Fortune, Nature, The Guardian, CIO Dive
To some extent, AI is just exposing bad practices that have been around for a long time. For example, if someone produces a paper containing AI-hallucinated citations, it shows that they have never been in the habit of checking that their citations say what they are claimed to. The typical way this happened pre-AI was to copy someone else's list of citations.
In a similar fashion, factoids like "you must drink eight glasses of water a day" can be recirculated for years before anyone bothers to trace them to their original source, which just says something like "this is the average fluid intake from all sources, and people will get thirsty if they ingest less"
Given that the output of an LLM is an answer to the question "what would the average Internet user type next after this", no surprises here.
A key point made here is that AI is very useful for automating low skilled and/or repetitive tasks, but risky when applied to more complex tasks where human analysis is essential. AI is also effective in analyzing visual data, such as X-ray images, retinal images, etc., but human oversight is still necessary.