Why Is There So Much Stigma Around Using AI in Scientific Work?

October 10, 2026 · Research Tools & AI · 18 min read

By Dr.-Ing. Faisal Qayyum. Facing this yourself? See how I can help.

The email said the draft read too much like AI. Politely, and in more words than that.

I read it twice, then closed the laptop and sat with it. There was nothing in the sentence to argue with and nothing in the paper to fix. The simulations were mine. The data were mine. The figures, the argument, the interpretation, every decision that made the work worth reading, all mine. The only thing the reader had reacted to was clean English arriving under my name. And clean English under my name, to them, meant a machine had done the thinking.

It had not. But even if it had, tell me what would have been wrong. That is the question nobody will answer directly, and it is the question this whole post is about.

I have used generative AI on my research for close to three years. Not quietly, not in the gaps, but on the actual work: setting up and checking simulations, processing results, building figures, drafting, revising, answering reviewers, the whole path from raw data to a submitted paper. I have written it into the papers themselves, in the disclosure. There is nothing in it I am ashamed of, and nothing in it that is not true.

Here is what I am finished with. The same field that has adopted these tools faster than almost any other on earth has decided that admitting it out loud is a mistake. Use it in private, deny it in public. That is the deal on offer, and I am declining it. I will say what I did, in the paper and here. This is not a how-to. I already wrote that one, for the people who want the workflow itself. This is the argument underneath it, and I am going to make it as bluntly as I can.

I Use AI Openly, and I Will Not Pretend Otherwise

Start with the thing everyone does and nobody says. You give the tool access, and the first reaction from the people around you is fear. You have handed it your mail, your files, your whole working life. Especially the messaging. People treat that as reckless, and they are not entirely wrong, because access is exactly the thing you have to think about. But you cannot write good rules for a tool you have never pushed against. Every model behaves differently, every workflow reveals a different edge, and the edge keeps moving as the models change. The only way to know where the line sits is to walk up to it on your own work and see what happens. That is not carelessness. It is the only honest way to learn what you are actually responsible for.

So when the work leaves my desk, I want one thing to be unmistakable. There is a human being on the other end of it, and I am not hiding the machine to protect myself. When I sign my name, the paper carries a disclosure that I used AI for the simulations, for the processing, for the reading, for the writing, for the whole chain. That sentence is not a confession. It is a statement of fact, and it costs me nothing, because everything true about the work is still true with it in.

I could dress it up. I could soften the disclosure, or drop it and let people assume the English came out of me fully formed at three in the morning. Many people do. I will not, because the moment you start maintaining a second version of how the work was made, you have taken on a job that never ends and earns you nothing. You have to remember what you said to whom, and hope the next person does not compare notes with the last. That is a worse position than simply saying what is true. And what is true is not embarrassing.

I am still finding the limit of all this. This is a learning phase for me, not a settled article of faith. There is a genuine risk on the other side that I take seriously: that it all comes to feel like pure machine work, work no human did. I do not have a clean answer to that, and I am not sure anyone does. But the answer will not come from people refusing to look at the tool. Worst case, someone decides a paper leaned too hard on the machine and rejects it. I can live with that far more easily than I can live with not knowing.

Why Is There a Stigma in the First Place?

Here is what I genuinely do not understand, and I have asked it of colleagues in three countries. There is a stigma attached to using these tools in science, and I cannot find where it comes from. The same people who hold it will admit, over coffee, that they too have typed a paragraph into a chat window and taken a cleaner version back out. Nobody objects to the computer, the internet, the statistics software, or the reference manager. This tool alone arrives with a moral charge that does not match the crime.

I have a personal stake in this. I am not a native English speaker, and for most of my career the distance between what I could see and what I could say was the bottleneck. The idea was clear in my head and precise in its feeling, and I could not get it onto the page, because the words that fit were not words I owned. Now the machine closes that gap: it takes the thing I think and gives me the sentence that matches it. And the reaction from some quarters is, we want to hear your opinion, as a scientist. Why do you think you are a better scientist than the model.

The idea is mine. I designed the study, I ran the simulations, I read the results, I decided what they mean and what they do not. The question is only whether the scientific meaning survives because the grammar was cleaned up on the way out. It does. The English is the vehicle; the science is the cargo. Nobody objected when a native speaker hired a copy editor or sent a manuscript through a language service. This is that, with the same tool available to everyone.

The stigma implies a small, guilty minority. It is not a minority. In Wiley’s 2025 survey of more than 2,400 researchers across 70-plus countries, 84 percent said they used AI tools for some part of their work, up from 57 percent the year before, and 62 percent had used them specifically for research or publication tasks, up from 45 percent (Wiley, ExplanAItions 2025). Four out of five reach for a general-purpose tool like ChatGPT or Claude rather than anything built for science. That is not a technical fact about the tools. It is a social fact about us.

If something about the output is wrong, the community that studies how knowledge is made should be the first to sit down with it, not the last. We know better than this, and we are behaving as though we forgot.

AI Is a Tool, Like the Computer and the Internet

Every revolution arrives the same way, and I have watched this film before. It starts with resistance. No, this is not right, this is not acceptable, this changes something we should not change. Then a pause, then the same people quietly using it, and the objection is never mentioned again. The computer went through it. The internet went through it. This is going through it faster, because the change now arrives inside the tools we already use.

I have been on the wrong side of this transition myself. I used to draw my plots by hand, not as a historian of the practice but because it was how I learned the physics, and I was proud of it. There was a rule between me and the students: if you hand in a printed figure you did not make yourself, if you cannot reconstruct the curve from the data, you did not do the work, you borrowed it. I still understand exactly why someone wants to protect that. Craft matters.

When the tool can produce the plot instantly, what it removes is not the understanding. It removes the typing. The understanding is still yours to build or to skip, and the danger is unchanged: the student who hands in a figure they cannot explain. That student exists with or without AI. The tool did not create them.

The boundary between human and machine is getting more diffuse, and the honest thing is to say so. A trained eye could once tell whether a poem, a song, or an abstract came from a person. That certainty is fading, and it is not coming back. You cannot argue a technology back into a world that has already seen it.

And there is a real danger the stigma buries. The risk is not that people use AI. It is that they use it without checking, and the errors spread. One analysis in Scientometrics counted 3,540 retractions tied to machine-generated or randomised content between 2010 and May 2024, climbing from 3 in 2010 to 2,302 in 2023 (Lei et al., 2024). That is a problem of undisclosed, unverified use, not of the tool’s existence. The cure is not abstinence. It is the discipline the stigma discourages: use it in the open, check it, and stand behind the result. Hiding the use is what makes the abuse possible.

What the Reviewer’s “What’s New Here?” Actually Shows

Now the reviewer, standing in for all of this in the world of papers. We sent out a revision, and the first comment came back exactly where I expected. What is new in this work? You published two papers last year, these four this year, and it is all the same thing. So what did you actually add.

I have sympathy for that complaint, because it often lands. Incremental publication is a real disease, and repackaging the same result four times is a strategy, not a contribution. But there is a version of the complaint that has nothing to do with the science and everything to do with suspicion, and that version shows where the stigma really bites.

Here is the pattern I have watched for a decade and a half, long before any of this. A professor republishes the same paper twice, a hundred percent overlap, and nobody says a word, because it is common and the person has enough standing to get away with it. I have seen it since I joined the field in 2016. So when the same generation points at a young author and says, we cannot tell what is new here, I want to ask what standard is being applied. Not because the young author is always right, but because the standard seems to depend on who is holding the pen.

The honest version of the question is not policing, it is infrastructure. If you publish the same line of work across several papers, give the next person everything they need to build on top of you. Say what the new result is, what was already published, and where. Publish it if it is worth publishing, and do not if it is not. But do not bury the key number so that a reader who wants to lift the whole thing up cannot tell what the paper found. That is not rigour. That is hoarding, and it slows the literature for everyone downstream.

When a paper is written well and read easily, suspicion should not be the first reflex. The reflex should be to check the content, which is the reviewer’s actual job. Is the method sound. Are the numbers traceable. Does the claim follow. If those hold, the polish is beside the point, and if they do not, polish will not save it. The game is about ideas, and the reviewer’s work is to test the idea, not to guess at how the sentences were made.

This is the part where I would say: if you are sitting on a revision and the objection you keep getting is about how the writing was made rather than what it says, the problem is usually addressable, and it is worth getting help to separate the two. That is exactly the work I do with people on scientific writing and reviewer responses. It is rarely about the tool at all. It is about making the contribution so clear that the suspicion has nowhere to land.

Every Researcher Now Has a Personal Assistant

Zoom out from the paper and look at what this gives a working scientist, because the stigma hides how large the gift is. For the first time, almost every researcher can have something close to a research assistant: available at any hour, does not tire, does not leave for industry, and never forgets the paper you read two years ago. It can take a rough idea and tell you what already exists that overlaps with it, which is the single most valuable thing a junior researcher never gets.

Think about what a student actually wants from a supervisor. Not a signature, not a deadline: an idea, or validation of their own, or an honest disproof, so they can run faster. The tragedy of a PhD is that three or four years is not much time, and most of it goes into checking whether a thought has already been had, whether a method will work, whether a result is real. A good advisor does that for you, and it has always been rationed, because there are never enough people to give it.

The tool does not ration it. It will go and search, find the work, and tell you the truth about it, including the parts you did not want to hear. That does not make a supervisor unnecessary; it frees a good one to spend their attention on judgement, taste, the argument, the strategy of a career. And for the researcher without a good supervisor, it changes what is survivable. Some of the people I help today would once have stalled permanently, not because their work was bad, but because nobody was around to give them one honest piece of feedback at the right moment.

I want to be careful about the boundary here. It does not replace the supervisor, and it does not replace the researcher. It replaces the mechanical parts: fetching, checking, comparing, formatting. The human parts of science stay human; the clerical parts stop consuming a career. That trade is the whole reason I am writing this.

That is the mentoring conversation I have most often now, with people at the stage where the tool is a fork in the road rather than a convenience. The relevant help is on the PhD and research mentoring side rather than the tool side, because the tool is the easy part.

What This Actually Buys You

I want to be concrete, because the argument so far could sound like philosophy and I am not a philosopher. I am a researcher with a job, a family, and a finite number of hours, and this is what the tools have given me back.

Two weeks ago I helped bring three people’s PhD theses together. Every one was, in substance, finished: the experiments done, the data there, the outline accepted. What was missing was the time to make it straight, to turn three years of scattered work into one coherent document. Their jobs had shifted, life had intervened, and the thesis sat there, a real contribution going stale on a hard drive.

I gave each of them a document and the same words. This is version zero, it is not your submission, but everything you wanted is now inside these pages, in order, and you can see the shape of the whole thing for the first time. Work forward from here. That is real output that would otherwise have stayed buried, returned to the people who earned it. Not that the tools write the science. That they stop good science from dying of incompleteness.

The other thing they buy is more personal. I can now get the large majority of my day’s work done without being tied to my desk. The parts that need a person, the conversations, the reading of a room, the decisions that carry weight, those stay mine and I want them to. But the document grind, the drafting, the formatting, the checking, the small tasks that used to eat a day, a machine does a great deal of that now, and it does it while I am somewhere else entirely.

That is what the stigma cannot see. While it argues about authorship, the practical people are getting their lives back. Eighty percent of the value of a working day is human interaction: the conversation that unblocks a collaborator, the meeting where a decision gets made, the ten minutes that save a student a month. The other twenty percent is documents, and documents are now largely negotiable. If you have children, or a long commute, or a second responsibility that never appears on a CV, you already understand why that trade is not a betrayal of the work. It is the reason to keep doing it.

Adopt From the Front

So where does that leave the scientist who is still hesitant, who reads all of this and feels the pull of the old caution. I will give you the blunt version. It is not meant to be cruel; someone should have said it to me earlier. If you are not using it, you die, and I do not care who finds that uncomfortable.

Industry settled this already, whether or not academia admits it. The companies that build the components, the ones you will consult for and the ones that will hire your students, are not holding a debate. They are using the tools and getting the gains. In a narrow niche, and materials simulation is a narrow niche, you have to be extremely good with the tool, not merely tolerant of it. Standing out of the race on principle is a choice, and it has a cost.

That is the stronger reason to adopt it, and it is not self-interest. If the scientists who know the material do not take the tools seriously, nobody else will build the guardrails for our fields. If we do not check it, the wrong answer gets generated, nobody catches it, everyone downstream inherits it, and it is never corrected, because the people who could have corrected it never engaged. That is the field the cautious path quietly chooses: not a clean, principled one, but one where the tools were run by people who do not know what a correct answer looks like, while the expertise sat outside, tutting.

I keep coming back to a phrase I use with people who ask why I am so open about all of this. I want to ride it from the front, not sit safely at the back. That means accepting that some of my assumptions will be wrong, that I will try things that do not work, that I may get some of this badly wrong in public. I would rather be wrong in the open, where it can be corrected, than right in private, where it changes nothing. And if there are consequences, professional or reputational, let them land on me, because I chose to test the boundary rather than repeat the standard line.

This is why the tool matters to me beyond convenience. I use it on crystal plasticity, on additive manufacturing, on the models that describe how real components are built, so that the boundary of what it can and cannot do becomes visible in the fields I know best. Someone has to do that. If not the people who understand the physics, then who. Standing aside does not protect the field; it hands it to whoever will engage with the tool uncritically.

If you want your group or your students to build this discipline properly, rather than inherit it second-hand from a vendor demo, that is the kind of ground I cover in the hands-on workshops I run for research groups. Not theory. Your own projects, your own files, your own boundaries.

Final Thought

I am not asking anyone to be less rigorous. I am asking for the opposite, which is the irony running through this whole argument. The people who use these tools carelessly are the threat to the record. The people who use them openly, check everything, and disclose what they did are the ones holding the line. The stigma rewards the quiet user and shames the honest one, and a field that runs on honesty cannot afford a rule like that.

I have read every page of what I sign. I take ownership of every sentence that carries my name. And yes, a machine helped produce the layout of those sentences, and I say so, in the paper and here. Both are true at once, and the discipline is holding them together rather than picking the one that makes me look better.

So I will keep saying it plainly. Yes, I use it. Yes, it makes my work better and faster. Yes, there are things it cannot do, and finding those things is now part of the job. If the community decides this is the wrong way to work, I can live with the verdict. What I cannot do is pretend otherwise, when the field already uses these tools, and the only thing standing in the way of doing it well is the silence around it. I would rather be early and honest than late and quiet. And if you are sitting on the fence, with the tool open in another tab and a slight guilt about it, you are exactly who this was written for. The guilt is the only part of this that has to go.

If You Want the Workflow, Not the Argument

This post is the case for using these tools openly. The practical side is elsewhere. How to Use AI to Write a Paper (Without the Fake Citations) is the step-by-step workflow I use to get a manuscript from rough notes to a defensible first draft, with the checks that keep every citation real. And for the data half, Codex for Researchers installs an agent on your own machine that pulls real data, builds the figures, and compiles a report. The argument here is why it is worth doing in the open. Those two are how to do it without losing your discipline.