The McDonaldization of Science by LLMs
We’ve all heard of enshittification. I have heard it used to describe what LLMs are doing to science and the peer review system. However, I think people are wrong and science is not being enshittified by LLMs; it is being McDonaldized.
Enshittification relates to the capture by companies of more money by 1) obtaining network effects by proposing a good product, 2) moving to a new segment, creating a worse product for their previous target which is now captured through the network effect, 3) rinse-and-repeat until value is extracted from all segments with none of them getting an actually good product anymore, even though no one can leave through sheer inertia of usage. In McDonaldization nobody is defecting or capturing anything, instead rational decisions taken within a narrow scope lead to irrational outcomes. I believe that today with LLMs, science is hurling itself toward just that: the McDonaldization of Science.
McDonaldization
While millions eat at McDonald’s, it is not our image of a “good” business: the pollution, the deforestation, the bad food, low pay, low quality, just to name a few. Yet, when you look at each decision that leads to a McDonald’s, it all makes sense:
- Costs are optimised for maximum benefit.
- Familiarity and consistency of the product means it’s a known value, and people eat there all of the time.
- The brand is hyper visible and the competition is crushed.
It’s because the harm comes from individually rational, locally sensible optimisation.
It’s the capitalist dream.
Now let’s look at the polar opposite of a McDonald’s: a community garden. Community gardens are stupid when looked at through the prism of capitalism. They make no money, people grow random things, the output is not optimised for profit…
And yet we love community gardens.
They bring people together, connect children with nature, create beauty and produce healthy food in otherwise (grey) urban places. This raises the question: why don’t we have more community gardens and fewer McDonald’s?
You can read the article on McDonaldization linked above, but a TL;DR version of it is that when we have efficiency, calculability, predictability, and control, and when we optimise our processes in a narrow scope, the outcomes can be harmful or irrational. My argument is that this is what we are about to do to science.
Science Has Been Using the Wrong Metrics for a Long Time
For years now we have had some of the conditions for the McDonaldization of science:
- Calculability: h-index, citation counts, publication in prestigious journals and conferences.
- Predictability — a Big Mac is the same everywhere and no surprise = no risk. Same paper formats, same conferences, same publishing process. We use known buzzwords to gather more visibility rather than focus on the science. We put effort into adapting to the journals’ and conferences’ speech and writing style. Enough equations and complexity usually convince reviewers (similarly to the law of triviality, in this case the majority of the review time is spent discussing relatively minor but easy-to-grasp issues). We can optimise for all those factors, forgetting the paper and the science itself.
- Control — where judgement is replaced by a script or a process. Peer review is becoming scripted control, where reviewers increasingly judge papers based on compliance with the form and field expectations (“do you have enough baselines”, “how recent are the baselines”, “is it novel enough” rather than which new knowledge it brings) rather than the substance. Furthermore, we rely on the h-index for funding and hiring, rather than substance and contributions.
Why work on an interesting niche paper that won’t gather citations when we can publish a paper on the latest buzzwords? Why publish one large, complete paper, when I can publish multiple small steps as different papers, gathering more citations?
No one likes this system, yet we compete in it. As said by the OECD, “the over-reliance of research assessment on narrow performance measures tends to undervalue critical research practices and contributions such as collaboration, openness, societal impact or policy support” which I argue are more important than the h-index. This is an irrational outcome fuelled by rational decision making; what’s good in the short term or for one’s career is bad for science and for the scientists in the long term.
However, up until now, chasing those metrics had one key limitation: efficiency. We still needed to implement our methods, test our results, convince people to finance our research, convince students to work with us… This helped regulate the madness, but this system is also coming to an end.
LLMs Make Us Chase Those Metrics even Harder
Today LLMs have changed the landscape. Low-hanging fruit are now even closer to our grasp; one prompt could be enough to implement a working solution and test it. Writing the paper can be done automatically (although I doubt the quality would be high enough on its own yet). In essence, the bar for effort needed to produce what qualified as a scientific paper has been dramatically lowered.
So, what should we do about those gains in efficiency?
We should build a community garden.
We should stop and talk about the future of science. We should invent new ways to evaluate success and decide how to fund our research. We need to look beyond h-index, citation counts, and low-hanging fruit, beyond minimal papers that we throw as hard as we can to the lottery of journals and conferences, all so we can optimise a metric that has never been that great to begin with. Interestingly, this is a problem that LLMs are fundamentally ill-equipped to solve for us; it requires creativity, forward thinking, and more than anything, it requires humanity.
However, what are we doing today? A fucking McDonald’s.
We are running as hard as we can toward the cliffs by making seemingly rational decisions for our careers that are bad for science. We are trying to publish as much as possible (flooding the journals with scientific slop and generated (possibly fake) datasets), as quickly as we can, to inflate even more those now worthless metrics, because those metrics are still the most important factor toward our personal career goals, and to obtain funding. It’s Goodhart’s law.
I cannot blame young researchers and PhD students. They are just starting in a system that doesn’t give them a choice; publish or perish is still true for them, and they don’t have the ability or weight to influence the system.
So I’m looking at everyone who is an established researcher and I ask the question: What are we doing?
I think it’s time we sit down as a community, decide what we value, and give those values the value they deserve. Because we get to choose the metrics (if any) that we judge ourselves by, and right now we are acting like we do not. I think that Slow Science and responsible research assessment practices (e.g. DORA) are good starting points toward our “community garden” of science. However, I somehow only see discussion about how to use LLMs to McDonaldize science as quickly as possible around me, and very few conversations on how not to.
And I personally do not want fast food science.