Confirmation bias

Here’s an academic article titled xm"AI and the problem of knowledge collapse". It’s paywalled though, so DM me if you’d like the pdf.

It looks more at the problem of our collective knowledge being at risk, which I think is a big thing. So much of our institutional knowledge is contained within people, and outsourcing that to AI is just a recipe for disaster on many fronts — not least of all because if an organisation ends up becoming dependent on AI, then it’s just making itself more brittle; if a model is updated, leading to significantly different performance, or it the cost model changes, then that has some big problems.

This next link isn’t an academic study, but hopefully helpful. It’s looking at how many companies are backtracking after the charging model for many AI companies meant costs skyrocketed. If IT gets people to start using AI en masse, are they really willing to be on the hook if the same thing happens with your organisation? AI is still not profitable for the people selling it, so this is unlikely to be the last time that the up the fees

This one is one of my favorites because it directly tackles the issue of ai coding productivity in the real world: https://arxiv.org/abs/2507.09089

Interestingly, the same group tried to repeat the study and found that it was difficult to find ai coders willing to try doing things without AI. Those that did join told the researchers they selectively chose tasks they knew they could handle if AI was taken away, avoiding complexity. In just one year, self confidence has plummeted among most developers who have experience using AI coding tools.

https://metr.org/blog/2026-02-24-uplift-update/#wider-adoption-of-ai-has-made-it-more-difficult-to-measure-task-level-productivity

The blog frames this as saying, without evidence, that maybe the avoided tasks would be even quicker with ai than without, and that avoiding more complex work is hiding an even bigger increase in productivity, but I’m skeptical. The first study showed that simpler tasks were easier to automate, but that the slow down experienced during longer more complex tasks, and the frequency of such tasks occuring in the normal course of software development, meant that ai made developers slower overall.

unfortunately what you need is something that shows short-term negative economic impact. anything else is the next boss’s problem.

It’s cute you think that facts will change their minds…

Try this one:

https://arstechnica.com/ai/2026/04/research-finds-ai-users-scarily-willing-to-surrender-their-cognition-to-llms/

Is this any good?

https://publichealthpolicyjournal.com/mit-study-finds-artificial-intelligence-use-reprograms-the-brain-leading-to-cognitive-decline/

My suggestion is to go through the Computerphile youtube channel: https://inv.nadeko.net/channel/UC9-y-6csu5WGm29I7JiwpnA
And check for the AI videos, it explains in technical details how AI works, and it is often putting it into a harsh, but reasonable light, in a very academic manner.

With it, build your own story of why do you think it doesn’t fit in your company’s IT plan.

I did a similar thing and my points were that automation is great when done correctly and carefully, if your automation produces mistakes, then you will have a big technical debt backlog in a very short period of time. Once we have that big bowl of spaghetti, we will have to eat it whole, and that means stopping everything to go through it as it has already happened in other companies (many articles detailing those failed POCs).

I also reasoned that there are other automation options that can use non-LLM solutions (be it with ML or not) that can be sensible and solve the issues we are aiming at. Finally I concluded with a simple point: LLMs are a nascent technology that hasn’t been proved to be worth the economic, ecologic and implementation costs. It works great for very simple, very small issues, but it doesn’t scale. And there are nowadays better, cheaper solutions for said issues.

(The economic angle is important!)

EDIT: There was an additional angle I touched briefly but it resonated with the team: that those that rely on AI helpers do not strengthen those skills (like a muscle, if you don’t use it, you lose it). I gave an example regarding hackathons, code golf, etc… Being able to do something, doesn’t mean you can be better at it.

RE:RE:RE:FINAL EDIT: In StackOverflow, how many times the most voted answer wasn’t the one that fit my problem? Or there was one that, in context, was better? Maybe a comment below that answer clarified a very important aspect… etc… None of these remarks are taken into account when the LLM mixes everything, and the weighing system won’t care for nuance, when our wold is full of it.

AI is mediocre because it mixes the good with the bad and makes it below average as a generation engine.

Ask the word predictor for a powerpoint presentation arguing against its use in your company.

You can always have a look at Wikipedia. All the articles have a “References” section at the bottom. Those will be filled with studies, conference proceedings and news articles. For example: https://en.wikipedia.org/wiki/Environmental_impact_of_AI

IT should provide you with reasons to adopt AI. It isn’t your job to show them the negatives. The entire thing is a net negative.

If companies have the goal of making money, AI will end up costing them as much as human labour and deskill the workers resulting in lower productivity.

If you find any studies that you don’t have access to, mention me and I’ll see if I have access (I’m a prof so free for me).

From a workplace camaraderie point of view, AI use promotes antisocial behavior, something I didn’t need a study to tell me, but I think the way the tests were designed was interesting. The participants discussed a real interpersonal conflict they are experiencing with different chatbots, and measured whether they were more likely to want to repair the relationship and how much they felt personally at fault. The results were exactly what you’d expect, but I think even more perverse is that participants are also more trusting of sycophantic ai responses, creating a vicious cycle of antisocial behavior. An ai that successfully pushes away your friend becomes closer to you by default.

https://arxiv.org/abs/2510.01395

I believe from personal experience that people heavily using llms are more likely to respond negatively to criticism, and more likely to have public emotional meltdowns from lack of practice dealing with non-sycophants, people that aren’t in love with them.