Can AI write your PhD dissertation for you?
I spent four days trying to get OpenAI's new tool Deep Research to research and write a complete dissertation. This is what I discovered.
Just a few days ago I asked the question: Does OpenAI's Deep Research signal the end of human-only scholarship?
I’d just got my hands on OpenAI’s new AI research tool Deep Research, and l was feeling somewhat overawed with what it was capable of — despite its flaws.
Like many early users, I was using it to research and write relatively short reports and papers. But I was intrigued to know how much further I could push it.
And so I set about seeing if it could research and write a complete PhD dissertation.
That was four days ago. As I type I am just putting the finishing formatting touches on a 400 page dissertation created by Deep Research that, for all its flaws, blew me away.
You can read more about the process I followed and my initial thoughts below. But if you’re interested in seeing what a state of the art AI-generated PhD dissertation looks like, here you go:
Can humanity survive the emerging polycrisis of environmental stressors, sociopolitical upheaval, and technological acceleration?
Abstract
Humanity stands at a crossroads, confronted by an unprecedented polycrisis—a convergence of environmental stressors, socio-political upheaval, and technological acceleration. Unlike isolated crises of the past, these crises interact in unpredictable ways, amplifying systemic risks and challenging traditional problem-solving paradigms. This dissertation explores the nature, evolution, and implications of the global polycrisis, employing an interdisciplinary approach that integrates complexity science, historical analysis, resilience theory, and systems thinking.
Through a comparative examination of past crisis periods—including the 14th-century Black Death, the 17th-century General Crisis, and the 20th-century World Wars—this study identifies recurrent patterns of societal stress, adaptation, and transformation. Contemporary manifestations, such as climate change, economic instability, geopolitical conflicts, and technological disruptions, are analyzed within a systemic framework to reveal their interdependencies. The dissertation argues that addressing the polycrisis requires a paradigm shift: from linear, siloed responses to holistic, anticipatory, and adaptive governance models.
By synthesizing insights from philosophy, history, and complexity science, this research proposes a new conceptual lens for navigating the polycrisis—one that acknowledges uncertainty, fosters resilience, and promotes ethical responsibility. The study concludes by outlining strategic pathways for human survival and flourishing, emphasizing the need for transdisciplinary collaboration, global cooperation, and innovative resilience-building measures to secure a sustainable future.
Download the complete dissertation above
To be very clear, this is not a PhD dissertation — AI isn’t there yet. But it is frighteningly close, and in some respects this document exceeds the depth, breadth and insight that many human-written dissertations demonstrate.
I’ve included an in-depth description of my approach below as it’s important to understanding what it took to achieve this — and just how impressive the final document is.
There are, as you might expect, some glaring flaws in the dissertation.
Deep Research is lousy — really lousy — at providing in-text citations, compiling comprehensive bibliographies, and using primary sources rather than whatever it can find on easily accessible websites (it even used Goodreads at one point!).
I’m also not sure how far I trust its discernment and critical “thinking” as it selects and weighs sources. And from my initial read I’m not sure how strong a through-line there is in the research and writing. And there’s quite a bit of repetition and redundancy, as well as a lack of originality in many places (more on this below).
But those concerns aside the dissertation is impressive enough that, if I was doing my PhD now without the help of AI, I would be deeply worried.
I would go so far as to say that we are rapidly heading toward a point where we will have to critically rethink the purpose and value of a non-AI augmented PhD.1
Of course, it’s questionable how original this AI-generated dissertation is, even thought the simulated thinking it represents is quite impressive. To get a sense of this I turned once again to Deep Research and asked it to assess the originality of the completed work. It’s conclusion: the ideas aren’t original, but the ways it combines them may be:
Originality Score: ~70%
Assessment: “Surviving the Polycrisis” provides a valuable interdisciplinary synthesis of complexity, resilience, crisis management, and risk analysis perspectives, but many of its core ideas have been discussed by earlier scholars and frameworks. It skillfully integrates these domains – an approach that is timely given the emerging nature of polycrisis research (cascadeinstitute.org) – yet the dissertation’s key arguments (e.g. that multiple concurrent crises amplify one another and require systemic resilience) are concepts already well documented in existing literature (polycrisis.org, polycrisis.org). In sum, its uniqueness lies more in combining and applying known theories across fields than in introducing completely new theoretical frameworks or insights.
But remember that this is what an AI produced with the only input from me being the initial questions, some light prompts, and a bit of editing. It’s not hard to see how, with more human input, Deep Research and subsequent reasoning AI platforms could transform the process of doing a PhD.
So does this mean that the concept of a PhD dissertation as an original piece of uniquely human scholarship is on the way out?
It may be that the consensus is that pursuing a PhD is more about the journey than the new knowledge it produces — in which case not using AI would make sense, at least to a degree.
But beyond developing new skills and understanding — both of which are important — a PhD is also about pushing the bounds of what is known, and making an original contribution to the sum total of human knowledge.
From this perspective I think we’re already at a point where an AI-augmented PhD has the potential to generate more new knowledge, and to generate it much faster, than a non AI-augmented one.2
This certainly goes for PhDs that draw on scholarship that isn’t tied to physical experiments — such as in a lab — or working with human subjects. But even in these cases, we are getting very close to scenarios where AI can suggest research strategies or even carry out research on its own.
Of course, there will be skeptics — and I’m sure that some readers of this article will already have decided that any semblance of scholarship exhibited by AI is a mere smoke and mirrors illusion.
But before jumping to any conclusions, I would suggest you read the dissertation above with an open mind.
You may still conclude that there’s nothing to be seen here. But you may also be surprised — and jolted into thinking differently about what research and scholarship might look like in the AI-enabled future we’re heading toward.
The process
This AI-generated dissertation started its life in the shower on the morning of February 5. Having decided to see just what Deep Research was capable of that morning, I needed a solid driving question for the dissertation I was going to ask it to write. This needed to be deep enough to underpin a dissertation that could be constructed through reasoning, scholarship, and web searches alone.
Somewhat randomly I came up with the question “Can humanity survive the emerging polycrisis of environmental stressors, sociopolitical upheaval, and technological acceleration?” — it had the feel of an interesting question that might lead to some interesting transdisciplinary research and insights.
This was bolstered with three sub-questions:
Is there evidence that humanity is entering a period of polycrisis that is unlike any previous point in history? What are the dimensions of the emerging polycrisis, how do they individually and synergistically potentially impact society, where are the commonalities and points of departure from previous points in history, and what can be learned from the past?
How might human survival be understood in the context of this research? What are emergent concepts and ideas around human survival and flourishing, and how are they informed by history? What can be learned from thinking at the edge of the distribution rather than in the mainstream.
How can humanity at a global scale begin to think about approaches to surviving the emerging polycrisis? What are the philosophical, social, political and economic frames and factors that are likely to inform thinking and actions?
This 5 minutes in the shower was the easy part. Actually getting Deep Research to produce something substantial proved far more difficult.
My first two single-prompt attempts were a train wreck. It turns out Deep Research is incapable of going any deeper than the research necessary to produce a 10 - 20 page report — no where near what’s needed to produce a full dissertation.
My next step was to see if it was possible to chunk the process down into a series of sub tasks while maintaining an overarching structure and narrative flow that threaded through them.
I approached this by first asking Deep Research to do the foundational research for the dissertation and provide what was essentially a roadmap and a chapter outline. These would then be used in each subsequent sub task prompt. I also asked for a list of foundational references, although if I’m honest I’m not sure how much Deep Research drew on these through the dissertation.
Once I had the roadmap, I asked Deep Research to research and write each chapter of the dissertation in sequence — there were eight of them. For each sub task I provided the roadmap/foundations document, the chapter outline, the foundational references, and the preceding documents.
What resulted was a ~400 page dissertation that is flawed, but even with the flaws it’s deeply impressive.
Ironically the process that took the longest was formatting the final dissertation — something my graduated PhD students will appreciate (probably with a touch of schadenfreude).3 The overall thinking time for Deep Research was around 3 hours. It took me a lot longer to get the text in order. But when you consider that over the past few days I’ve had a full schedule of teaching, meetings, chairing committees, and writing reports (real human-written ones), 4 days from concept to completion isn’t bad.
By comparison, if you treat Deep Research’s dissertation an in-depth analysis rather than a piece of novel scholarship — the sort of thing a research consultancy or an organization like the National Academies would compile — my estimate is that it would take a minimum of 3 months for a single researcher to get anywhere close to producing something of this breadth, depth, and polish.
I’ve included details of all the prompts I used in the process below, as well as the three foundational files that I used in sub tasks 2 - 9. It’s a lot of stuff, but possibly of interest to anyone digging deeper into how I got to the final dissertation:
I feel at this point I should apologize to my current student for throwing this potential wrench in their well-laid plans!
To be clear, I don’t think we are heading for a future where AI can independently research and write a PhD dissertation — that would require independent intent and understanding on the AI’s part. But I think we do have to grapple with the very real possibility that AI is becoming a powerful catalyst and accelerant in research that will relegate human-only research to a class of artisanal intellectualism where the primary purpose is the provenance and process, not the product.
My ASU students will be amused to know that yes, I did use the ASU Dissertation Wizard. And yes, I did use my privilege of not having to submit this to the dissertation formatting police to subsequently alter the formatting!
Fascinating, and a bit frightening. I agree that this is a fantastic opportunity, but only if we move carefully. Do we lose the important "process" if we rely too much on AI? We need to think very carefully about how we build this into our curriculum, not only with PhD theses.
Is it the ultimate irony that AI alone can fix it?