AI tools have become widespread since news of ChatGPT-3.5 prompted me, in early 2023, to make an OpenAI account. I remember sitting in a coffee shop, sipping on a latte, and holding a half-dozen “conversations” with the newly released model.

I asked it a series of math questions. I asked it about politics and about poetry. I asked it to write me a short story. I even asked it questions about niche topics like Settlers of Catan to see if it “understood” strategy in the context of a game. And, of course, I asked it about economics.

However, as a member of INOMICS’ editorial board, I found that AI tools back then lacked the ability to write a thorough, in-depth, and engaging article about economics topics such as how corruption can affect economic growth or the pros and cons of focusing on “mesoeconomics”. We decided to not use AI tools in our article-writing or editing process.

Similarly, I decided not to use AI tools at all when I was applying for PhD programs.  That’s not to say that I think these models can’t ever be helpful – I’ve found that AI LLM models are very helpful for one thing in particular: generating ideas. However, I’m serious about becoming a PhD economist, and was determined to submit the very best of my own writing for my applications. Further, there’s always a concern that using AI tools would be considered a form of plagiarism.

I think not using AI tools at all for my application was the right choice, and I think it’ll be the right choice for you, too. What’s more, even if AI models become incredibly capable “economists” on their own, I think eschewing them will still be the right choice for your application process.

Without further ado, let’s explore the reasons why.

First: hallucination adds extra work, and relying on AI delays you from building your own expertise

AI models are known to hallucinate. That means, when using them for an important purpose, checking the AI’s output for accuracy is critically important – but that adds work that I didn’t feel was worth it for my application. I felt that the effort I’d have needed to spend studying each aspect of the AI model’s output (to ensure it was accurate) would have been inefficient.

Instead, I spent time reading the economics literature myself, seeing what interested me, and reading about  how other economists in the space structured their research. This was very helpful, especially as I began to outline my own phd research proposal.

Think of it this way: if you want to be a successful PhD student, you’ll need to understand a large portion of the literature in your subfield anyway. You may as well get started during the application process! This will greatly help you with any research proposal you must write, and with any conversations (or admissions interviews!) you have with potential supervisors.

Rather than take an AI model’s potentially erroneous output and start from there, I searched Google Scholar and RePEc myself. I read the literature to see if there was a gap that my ideas could fill. I put together a methodology that I knew was sound, using econometric methods that I was confident I understood. I also found public datasets that I knew were accessible and relevant, to ensure that realistically usable data was available – and to show that on my application. Finally, I wrote my own ideas for future research, and studied the research interests of the faculty at the institutions I was applying to.

Sometimes, doing something the “hard way” really is the best in the long run. After all, there are no free lunches.

Second: personal essays require a personal touch

Some schools require specific essay questions or have topics they want you to broach in your personal statement. For this reason, it’s highly advisable to double-check each institution’s expectations for your personal statement (if they do, this is usually found or referenced in the school’s online portal where you upload the statement). I had to write several different personal statements, as each school had slightly different needs.

Clearly, these statements require a personal touch. If you outsource the work to a machine, you’re preventing the admissions committee from hearing your voice, and from reading your story as you would tell it. This clashes with the entire point of a personal statement: in your application, you must show the admissions committee who you are and why you wish to study for an economics PhD.

This may sound like a “fluffy” or silly concern, but the truth is that admissions committees get many applications from very qualified students. Sometimes, deciding who to give an offer to can come down to less measurable factors, such as how well the committee can picture you succeeding at their institution or fitting into their faculty’s research interests. A good personal statement will help you convince them that you’re a good candidate and pursuing a PhD for the right reasons. 
Apart from this, many admissions officers can, or at least claim that they can, identify work written by an AI. Whether it is an overuse of stock phrases, an inconsistent register, or a “genericness” in the style or argumentation, ultimately if an admissions officer or potential supervisor suspects that a submission is AI written, at best they will treat your application with scepticism, or at worst discard it.

Overall, you ought to give the admissions team as much of a sense of who you are as you can. Express your personality, show them your enthusiasm. AI simply can’t do that as well as you.

Third: AI output is not your work, and this limits your growth

Different people may feel differently about whether or not, or to what extent, using AI tools constitutes plagiarism. Whichever camp you personally fall into, though, the following statement is true: an AI model produces output that you did not write yourself. And this is a problem if you aim to be the best economist you can be.

This is related to the first point about the hallucination problem, but goes deeper. AI tools offer a cognitive shortcut. However, if you’re always taking a shortcut, even if only to kick off ideas and begin your work, you’re robbing yourself of the ability to learn how to start a research process well, which can make future research tasks harder for you. I know this would likely be true for me, because I’d noticed this tendency in myself long before the advent of popular AI models. Allow me to explain.

When learning math in grad school, I often felt very confident that I understood the teacher’s example as they worked it in class. Yet, as soon as they stepped away from the chalkboard and asked me to solve a similar problem on my own, my mind would often blank. Suddenly, recalling the right steps to take – and understanding exactly what they meant – was much harder.

I think this is a relatable experience for many, and I believe that over-relying on AI tools is just the same as relying on a teacher to begin solving a math problem for you. It seems innocuous when you still have the aid of the teacher. A limit or integral — or the beginning of an econometric theorem — all seem very straightforward when your teacher sets them up for you.

But when the aid is removed, suddenly, our brains may panic because they haven’t actually learned how to properly get started. Remember, your brain is an expert in finding shortcuts, and that sometimes means you don’t learn as well as you think you did!

Aside from this, research has shown that if you over-rely on AI tools you may – even subconsciously – become worse at detecting hallucinations, and stop thinking as much about serious issues like plagiarism.

As Zhai, Wibowo, and Li put it in a 2023 paper¹:

“Despite the undeniable advantages of AI dialogue systems in streamlining research processes and enhancing academic efficiency, our analysis reveals a concerning trend: the potential erosion of critical cognitive skills due to ethical challenges such as misinformation, algorithmic biases, plagiarism, privacy breaches, and transparency issues.”

Thus, I’m confident that over-relying on AI tools to begin – say – your research proposal literature review also limits your growth as an economist, even if you don’t realize it.

If you don’t take the time to actually read and process the papers in economics yourself, you are missing out on a great deal of knowledge. Even simple meta-knowledge about what exists in the body of economics literature can be important for your career. It’s potentially embarrassing to not know of a famous or seminal paper in economics, and it certainly wouldn’t look good if you aren’t aware of the literature in your own niche subfield.

Fourth: AI can’t output work up to the same quality as you or I can

Truthfully, the thought of using AI on my PhD application never seriously crossed my mind. That’s because, through my time writing and editing for INOMICS, I’ve continually tested new AI models to see how good they are at writing thoughtful economics articles.

I have yet to find an AI model that can write a helpful, informative, and analytic article the way that Hanna Sakhno, Dr. Sahar Milani, Dr. Tom McKenzie, Dr. Gómez-Pineda, or I can. AI models have yet to demonstrate the ability to go beyond presenting common discourse about a topic and teach the reader something new, or present things in a new light.

And I know that if AI models aren’t up to my standards for what we post on the INOMICS website, it will not be up to an admissions committees’ standards for a PhD application essay or research proposal, either.

I’m confident that someone seriously interested in pursuing an economics PhD is much brighter than even our most advanced LLMs or neural networks today. Perhaps that will change one day in the future – but even if it does, doing the work yourself will develop you into a much better economist than relying on AI output ever would.

Conclusion

I learned a lot throughout the process of applying for my economics PhD. I gained a solid preliminary understanding of my topic area that will help me immensely when I begin to write my first publishable paper. Further, after combing through the literature myself, I’m confident that the research I want to pursue would be advancing the field in a way that hasn’t been done before, which is exciting. I don’t think AI had a place in that process.

And, at the end of the day, my hard work landed me a very generous offer of admission at an excellent university that I’m excited about.

So, my fellow economist hopefuls, take it from me: don’t rely on AI models to write your applications for you. You owe it to yourself to do the hard work. After all, there are no free lunches in economics or in life.

References:

1: Zhai, C., Wibowo, S. & Li, L.D. The effects of over-reliance on AI dialogue systems on students' cognitive abilities: a systematic review. Smart Learn. Environ. 11, 28 (2024). https://doi.org/10.1186/s40561-024-00316-7

---

A note on AI usage at INOMICS

Here at INOMICS, we’re not against using new technology. On the contrary: on our website, you’ll find helpful AI tools – such as a feature that generates questions (and answers) for our Economics Terms A-Z articles, or an AI Summary button for our article pages. But, when it comes to producing articles and resources, quality and accuracy are our highest priority. 

So while we sometimes use AI tools to help us brainstorm ideas, suggest phrasing, translate existing text, or generate interesting header images for articles (like the one on this page!), when we have experimented using AI to produce longer articles, we have always found the results to be lacking, not to mention sharing many of the concerns expressed in this article. As such we do not (yet) publish Advice articles, Blog posts or economics terms/lessons that were partially or wholly written using AI.

---

_{^{Image Credit: INOMICS, using Canva Magic Studio AI Image Generator}}