What AI Actually Changes About Research Papers
The research paper has already died several times. Economics papers looked nothing like prose in the 1940s compared to what they became after the 1970s. Then data got cheaper. Then causal identification became everything. Then researchers needed access to Danish Census records to get tenure.
Each shift redefined what counted as a legitimate contribution. Each time, observers could have asked: will this kill the research paper? Each time, the answer was no-the paper simply became something different.
How the criteria keep changing
Before the 1940s, economics research was largely narrative. Samuelson reframed the field through mathematics and thermodynamics. What had been a research paper became a polemic.
By the 1970s, high theory dominated. Then computers arrived with data so rich you couldn't plot every observation in a single figure. That data required theory to be credible-pure theory papers gave way to applied work across every quantified social science.
Researchers got better at telling stories about instrumental variables. Then they told so many weak stories that credibility standards tightened. Better data. Stronger identification strategies. Eventually, researchers either ran their own experiments or cultivated access to the most granular administrative datasets available.
A paper was a paper. Until it wasn't. Until that wasn't a paper either.
What lowers the cost of mediocrity
AI will reshape what counts as a contribution, much like thermodynamics and cheap computing power did before. The mechanism: it lowers the cost of mediocre work.
Mediocre analysis will move into blog posts and journals no one acknowledges. But mediocrity is relative. The quality of those secondary publications will improve dramatically. What takes an afternoon's work now will expand significantly.
The papers in top journals? Probably safe. But not because AI won't touch them.
More researchers means more ideas
The bar rises when more people can clear it. Research improves the same way sports improve when you widen the geographic pool-more talent emerges.
Right now, someone at a directional state school is teaching a 3-3 load and didn't get the placement they deserved. They're not waiting for grad assistants anymore. They don't need an army of support staff. They have AI for writers and research tools that work like one.
That person can now make a major contribution. So can dozens like them. The raw number of researchers with capacity to produce legitimate work is increasing dramatically because everyone gets research assistance and editing support built in.
More ideas reach publication. The threshold for acceptance rises. The papers that get published get better.
The honest answer about the future
Forecasting technology beyond five years is like forecasting weather beyond five days. The honest answer is: nobody knows.
AI probably won't kill the research paper. It will change what a research paper is, just like everything before it did. The criteria will shift. The standards will tighten. New researchers will surprise everyone.
The paper will survive. It just won't look like what you're reading now.
Your membership also unlocks:
