A Conference, A Journal, AI, and I
Does the D in PhD stands for… (existential) Dread? On this blog post, I look back at how some less palatable experiences from the beginning of my third doctoral year helped mold my perspective on my own field: AI.
One of the perks of being a PhD student is how flexible your time management and business can be. At times, this is a blessing that allows me to relax and allocate a good amount of time into taking care of my newborn child. Other times, this is a curse that banishes me for months into a hole filled with experiments, reports, papers, and deadlines. Having returned from said hole in the first half of my third year as a Doctoral student, I decided to take some time to reflect on one of the more intense periods of my time in Science Tokyo through a post here on the Student Ambassador Blog.
A Conference
In mid-November, I attended the 34th ACM International Conference on Information and Knowledge Management (CIKM 2025) held in COEX, Seoul, South Korea. This attendance was a bit unexpected since the short paper I submitted to this venue was unfortunately rejected. However, AIST (in which I am currently employed part-time as a Research Assistant) was generous enough to sponsor me in order to nurture my skills as a researcher. This was the first conference I attended as a doctoral student. It also might be the last.
To give you an idea of the scale of recent academic AI conferences, here are some statistics from CIKM 2025: the conference received 2,671 submissions and accepted 810 of them, generating 11,709 individual reviews. Alongside the main and short tracks, it held 10 tutorial and 14 workshop sessions. If my memory serves me right, around 1,000 people attended. Despite these numbers, CIKM is actually on the middle-to-low end for AI conferences in terms of size. AAAI 2026, the biggest conference this year, towers over CIKM with an unprecedented 29,000 submissions. That should give you an idea on how ludicrous the velocity and scale of AI research is right now with all the vigor and capital circulating in the industry. While this means that opportunities for an AI program graduate like me might be opening up, it also means that the competition for them will be equally brutal. I am unsure how to feel about all this.
Among all of the statistics presented at the opening ceremony, one stood out to me as particularly surprising: 70% of the papers submitted to CIKM are produced using some form of assistance from generative AI. The forms of assistance themselves were undisclosed, so in theory they could range from a simple proofread, machine translation, to generating entire paragraphs from prompts. While I am ambivalent to the notion of using generative AI in academia, the number 70% did strike me as being too high. You might have guessed, but my short paper submission was among the remaining 30%, and my reason for not using AI is quite simple: as a student, I consider writing a part of my learning process. Thus, it makes little sense to delegate some of that process to generative AI, maybe with the exception of non-creative tasks (e.g., transcribing an excel table to LaTeX).
My overall experience with CIKM was pleasant. To start with, it was the first time in two years that I was able to spend an extended amount of time listening to presentations on research that has little or no relation to my own. It was honestly refreshing and relieving to see the breadth of topics that exist in my field, reminding me that there is value to all kinds of research, including my own. Attending CIKM was a much-needed boost to my confidence and spirit. Seoul was interesting. More on this later.
As for the presentations themselves, each accepted author on the main track was given 15 minutes to present their work and answer questions on it. However, a non-negligible portion of the presentations I attended were given by a proxy presenter due to the real author being unable to attend. As such, I was unable to discuss some of the more interesting results. Furthermore, on more than one occasion where the author was present, I was unable to get my questions answered since the presenters were unable to communicate or discuss in English beyond their presentation materials. This is not a slight towards researchers that don’t speak English (in fact, I think the world will benefit from more major non-English conferences). I can’t help but wonder, however, if by allowing researchers to skip some steps on the way to the finished paper, generative AI robs them of the opportunity to deepen their understanding of their own work and how to effectively communicate their ideas.
That said, I concur that using whatever tool at hand to get things done faster is very rational when your graduation hinges on you competing against over 29,000 other potential authors.
A Journal
Concurrently with my (rejected) short paper submission, I also submitted an extended version of the same research to a journal for publication around July 2025. This submission is actually my second one after a rejection in another journal (for the same research) after around six months of back-and-forth with the editor. Since the review process for a journal typically takes anywhere between 3 to 18 months, I figured that it’s a good idea to shoot as many shots as quickly as possible. With my expected graduation date of September 2026, not having even one of the two required publications at that time meant I needed to take every chance I could get. To keep this post from being too much of a downer, I am very happy to report that the journal submission was accepted, and I am currently waiting for the reviews for another conference publication that will determine my graduation. Said conference will be held in the exact same venue as CIKM. Life is funny that way sometimes.
Now, back to the recollection. Around August 2025 I received the first round of reviews. There were seven reviewers; my supervisors agree that this is an unexpectedly high number for this particular journal. Fortunately, none of the points raised by the reviewers seemed like a deal-breaker for the submission. And so, after finishing 20 pages of response letter in a week, I re-submitted the paper and waited. The second round of revisions arrived just a week before I fly to Korea for CIKM 2025, complete with a 2-week short deadline. One of the reviewers included the following short response:
… There are also indications of the use of AI in writing the paper and answering the questions. Therefore, this paper cannot be recommended for publication.
To say I was annoyed at the time is a massive understatement. Even discounting the fact that I made the conscious decision to refrain from using any kind of generative AI in my work, after a grueling 12 months of work and one rejected submission to a previous journal, the possibility of another rejection due to something out of my control was honestly, depressing. That being said, I cannot completely put the reviewer at fault. The growing problem of AI-generated junk papers has been surfacing since 2024 (The growing problem of AI-generated research papers), and I guess some of my human writing ticks can come across as AI-generated at times (in particular I used to love using em dashes before it became an obvious AI tell). So I gritted my teeth, politely wrote another 15 pages of response letter on the three of my five nights in Seoul, sent it back to the editors, and finally got accepted a month later.
AI-generated papers are not the only generative AI-induced problem in academia these days, as AI-generated reviews are also a thing. Late last year, third-party analysis done on the reviews for the 2026’s edition of ICLR (another major AI conference) revealed that 21% of them might be written fully by AI. Most infamous among these AI-generated reviews is a particular one with 3000 words and a list of 40 weaknesses/questions, a generational masterpiece that absolutely no reviewer would have the time and commitment for. In a world where low-quality scientific articles are both generated and reviewed by AI in a vicious garbage-in-garbage-out cycle heading towards model collapse, I’d have to say that it’s not easy to be a glass half full-type of a PhD candidate. Despite my eventual acceptance in the journal, the review process has left me a bit lost and skeptical about the future.
A few weeks into December, my supervisor mentioned something interesting about this topic. He was one of the Area Chairs (AC) for AAAI 2026 and oversaw several reviewers during the review process. Due to the massive number of submissions, this year the conference introduced an AI-assisted Peer Review Pilot Program, in which a generative AI powered review is provided to the AC as an additional consideration item. Turns out that he almost always feels that the AI-generated reviews were better-written compared to human-submitted ones. Again, I am unsure how to feel about all this.
AI and I
An AI Doctoral Course student being highly skeptical towards (generative) AI progress might sound like a premise of a bad sitcom, but it is the unfortunate reality that I am living in. Contemplating the experiences I shared above, I do think it’s irresponsible to write off the usefulness of generative AI to help with the problems of today. However, we need to recognize the fact that those problems were partly caused by the existence of the technology itself. My grievances actually lie not with the knowledge and technology, but with the narrative, capital, stakeholders, and institutions (or lack thereof) that scaffold its growth and development. Even confined to the narrow world of academia, I saw firsthand that unchecked and rapid adoption of generative AI potentially caused more trouble than it was worth. It’s easy to work out that this also applies to our everyday lives; a quick glance at our social media feed will most likely be enough to confirm the prevalence of AI-generated content that is innocent at best and downright harmful at worst. To quote contemporary linguist Adam Aleksic: “(the dead internet theory) used to be a lunatic fringe conspiracy theory, but it’s looking a lot more real.
Despite understanding the problems it brings, I can’t fully dissociate myself from generative AI. To start with, maintaining competitiveness without generative AI among the workforce post-graduation seems like a very tall order. It’s much less about being more productive by using them as assistive tools such as Claude Code, but more about the work that is available in the market. As the technology continues to get pushed towards adoption by practically all tech giants, positions and research on more conventional brick-and-mortar AI are starting to fall out of style. As such, I do feel that acquiring generative AI-related skills and know-how is unavoidable. I am only 6 months (hopefully) from my expected graduation and looking at the available job postings, unfortunately it’s time to make peace with my apprehension and dip my toes into the same technology that has been causing me some headaches throughout the year. I guess the old adage is true:, you really do need to know your enemies to survive.
In short, my relationship with generative AI at the moment is probably best summed up by a particularly memetic option (which was removed in 2013) on Facebook’s relationship status feature: “It’s complicated.” It seems like for every useful block of code generated using Claude Code or OpenAI Codex, there are hundreds of bots polluting our social media feeds with false news and engagement bait. It’s easy to forget, but many seemingly innocent technologies we enjoy in our day-to-day life can or already have been utilized in more sinister ways. Do you know that the microwave was invented by the US defense contractor Raytheon, as a shoot-off from their radar technology used in World War II? Or that the technology that enabled the internet was originally created for military purposes? I wonder if the unease I’m living with every day is shared by researchers around the world working on frontier sciences with potentially harmful applications. I sure hope so, because while it might feel like the proverbial curse that comes with the privilege of wisdom, I personally also consider it a call-sign of the moral clarity that’s indispensable for humanity’s progress.
What next?
When I started writing this blog, I wasn’t intending for it to take on this somber of a tone. Initially, I was just thinking of lightly sharing some AI-related anecdotes from the past couple months of me going AWOL due to some PhD-related business. To compensate, I’ll end this blog post on a couple of hopeful blurbs:
- My time attending CIKM 2025 was mostly wonderful. While some of the actual presentations felt lackluster and I spent half of the nights burrowing in my hotel room writing review rebuttals, I actually got a chance to hang out post-conference with some of the attendants and talk about life and research. I feel like I learned a lot from this experience, including things that are impossible to get from talking to an AI. Generative AI can produce texts that feel like they’re written by humans, but these texts don’t come with actual human experience from which they were born. While it’s true that AI conferences now can feel just like a compulsory step towards a degree that you need to get over with, I still think there is value in submitting and attending them to try and find real connections with other researchers.
- Seoul is interesting. Knowing it only from TV shows and dramas, I always thought that the city would feel similar to Tokyo. Surprisingly, while some areas bear a degree of cosmetic resemblance (e.g., Myeongdong and Shinjuku), I think Seoul operates more like my capital city, Jakarta. From the car-centric city design, organic sprawls of back alleys, hustle and bustle of the many markets, delivery and franchise-based food culture, and generally more vocal denizens, Seoul is definitely a different breed of metropolis than Tokyo. Walking down the Han River and slurping down Ramyeon in the dead of night is a memory I won’t forget soon.
- After my journal acceptance, I got a bit of time to slowly organize my thoughts and reflect. This did help to level out my feelings and guide me towards an auspicious frame of mind. As mentioned, I do think it’s almost irresponsible to deny AI wholesale at this juncture. The figurative cat is out of the bag; the pandora’s box has indeed been opened, so to speak. As such, we need to focus on ensuring that it can be a net benefit for humanity as a whole. To that end, we must (responsibly) employ any means, including AI itself. The practical essence of AI has always been its ability to scale certain tasks. In response to the deluge of AI-scaled content generation, AI-scaled content verification is one critical avenue to explore. My supervisor’s anecdote about AI-generated reviews is one good example of this, given some caveats.
My biggest takeaway from all of this is that in times of machine-induced existential crisis, the best thing to do is just to go offline, meet some friends, and touch some grass. Being in an industry where I’m working online in front of a screen by default did make me feel like AI is taking over the world sometimes, but it’s actually very easy to see that it’s not. Will generative AI ever fully displace humans? In some limited scope, maybe. Not completely though, not even in the field of arts and creative media, no matter how often lines such as “Hollywood is dead” or “animators are doomed” are repeated every time a new generative model comes out. This might seem slightly unrelated, but a few weeks ago I went to the 10th Anniversary Exhibition of the mangaka Paru Itagaki (Beastars, Sanda). Endlessly moved and inspired by her journey, visible through all the hand-drawn panels with manually attached screen tones and correction tapes, I was quickly convinced yet again that real value lies inside the human process.
