Panel Discussion: Research in the Age of AI
2026-11-20
(Opening remarks made as part of a panel discussion held at IIT Palakkad on Research in the Age of AI.)
Personally, I don't use LLMs except for search and experimentation. Everything I publish (including code) is manually written. That said, I'm really impressed by what they are capable of. In my area, software verification, there are even companies verifying critical software at scale with the help of LLMs.
Coming to research in general, the situation looks both depressing and hopeful. Let's keep the ethical and legal issues aside for a moment. Even so, the situation is concerning because we are witnessing the formation of a new digital divide. We say that we do research to further human knowledge. But when you are enrolled in a program like PhD, it is a fact that you are competing with other groups in your area. Now, big groups at premier institutes have access to frontier models with a large token budget directly offered by companies like Anthropic and OpenAI. Even if you are okay with using LLMs for research (whether you should be is a different question), this makes the competition tougher. Earlier, digital divide meant those who had access to computers and those who didn't; now it's those who have access to frontier models with an infinite token budget vs those who don't have such a privilege.
Most scholars are wondering what the future holds. One idea is that our role will evolve into the ones who improve LLMs or other AI tools. But that's a bad idea as long as powerful AI is owned and controlled by a handful of companies. We'll be safe in the short-term, but we'll be digging a grave for the entire humanity in the long term. The only solution is regulation or a bubble burst.
What is good about the recent disruption is that LLMs are forcing us to rethink research. LLMs have been shaking the software industry recently by finding a lot of critical vulnerabilities in software that runs the world. But these vulnerabilities stem from the bad programming languages and bad engineering practices that we have been employing all along. In other words, AI is forcing us to pay off the tech debt we have been accumulating for decades. Once the dust settles, we will be left with improved software and increased awareness of best practices. Just like that, AI forces us to rethink research. If what I'm doing can so easily be replicated by a junk generator or an autocomplete engine, what am I really doing? What should I be doing instead? That's good.
I also wish every scholar paid attention to issues like the copyright status of LLM-generated material, but I don't want to spend time discussing that right now.
Nandakumar Edamana
Tags:
Read more from Nandakumar at nandakumar.org/blog/