Lately, like many others, I’ve been grappling with existential questions about what it means to be a “scientist” in the modern world. I’ve been gradually losing faith in the professional and intellectual culture that gave my work meaning when I entered the field over a decade ago, and wondering if the kind of scientist I spent years becoming still has a meaningful place in it.
\nWhen I think back to industry AI research between roughly 2015 and 2020, I remember we all aspired to enter the "Google Brain"s, "early OpenAI"s, and "early DeepMind"s of the world post-PhD. There was a certain type of intellectual "hunger" to be part of a research lab where something mysterious, cutting-edge, and impactful was happening. Back then, the research scientist role had a relatively distinct identity.
\nIf you were a research scientist in industry at the time, you were expected to identify and propose important open questions, understand the context around them by studying prior work, design prototype experiments to de-risk the most ambiguous directions, interpret the evidence, and communicate your findings both internally and with the broader scientific community -- all ways of directionally contributing to our shared scientific progress. Accumulated knowledge and judgment were an essential part of that professional identity: becoming a “better scientist” meant developing your research taste over time by struggling with unsolved problems, collaborating with others, learning from mentors, receiving feedback and critique, and discovering the limits of your own understanding. The main motivating factors were intellectual autonomy, a drive for discovery, recognition by one’s peers, and being part of a community that cares deeply about similar problems.
\nNow we’re seeing this arrangement change. Expectations attached to the research scientist role are shifting more and more towards rapid productionization of ideas, engineering-heavy applied development, and immediately commercially useful outputs. Purely exploratory research, inconclusive results, or spending time developing a deeper understanding of a problem before deciding how to approach it, have been squeezed out.
\n This doesn’t necessarily mean that fundamental research has disappeared, or that every frontier lab has adopted the same transformation. But it does suggest a shift in what institutions recognize as valuable scientific work. \n Industry research was never insulated from commercial pressures, competition, or the pursuit of recognition. But the pace has accelerated dramatically, and the field feels noisier, more crowded, and more cutthroat than ever. There's enormous pressure to produce visible results quickly, and less and less distinction between rigorous work, and work that's simply well-packaged or aggressively promoted. What worries me most is the erosion of intellectual honesty and scientific standards in an environment where speed and visibility are so heavily rewarded. After spending years developing my own sense of what good research looks like, I find these phenomena difficult to reconcile with the profession I thought I was entering.\n **It feels as though the terms of the contract have been changed retroactively.** The changes unfolding in AI research are not unprecedented. One concept that explains how technological progress changes the way people work, but also how their expertise is valued, is **deskilling**.\n Deskilling is the process by which technological and organizational changes reduce the specialized expertise required to perform certain tasks [1]. A worker can still be highly skilled, while the job function narrows around a set of tasks that no longer requires the full extent of their expertise. For example, as research roles become narrower and more standardized, the years spent developing scientific judgment and research taste may no longer translate into better pay, greater scope and responsibility, or more influence over the direction of the work.\n \n\n In Labor and Monopoly Capital (1974), Harry Braverman examined how management separates the conception of work from its execution: planning and knowledge of the overall process become concentrated in management, while workers are assigned narrower execution tasks within it [2]. His argument was more nuanced than “automation replaced skilled labor”: he argued that **the organization of work can change who gets to decide how knowledge is applied, and toward what ends.**designed to maximize managerial control. Capitalism is characterized by
\n“the incessant drive to enlarge and perfect machinery on the one hand, and to diminish the worker on the other.”
\n<sup>24</sup>—
\nLabor and Monopoly Capital, 25th Anniversary Edition [5]
\nBraverman also questioned what it really means for a worker to become “more skilled” on average. The total body of scientific knowledge continues to grow, and new technologies make more of that knowledge accessible and easier to put to use. But that doesn't necessarily mean individual researchers gain more expertise, autonomy, or influence over how it's applied. Organizations that are best positioned to harness new technologies and draw on an expanding body of scientific knowledge also become the places where understanding, resources, and decision-making power are increasingly concentrated [2].
\naveraging,or, on the contrary, toward polarization... The mass of the workers gain nothing from the fact that the decline in their command over the labor process is more than compensated for by the increasing command on the part of managers and engineers. On the contrary, not only does their skill fall in an absolute sense (in that they lose craft and traditional abilities without gaining new abilities adequate to compensate the loss), but it falls even more in a
\nrelativesense.
\nThe more science is incorporated into the labor process, the less the worker understands of the process; the more sophisticated an intellectual product the machine becomes, the less control and comprehension of the machine the worker has.
\n<sup>27</sup>—
\nLabor and Monopoly Capital, 25th Anniversary Edition [5]
\n So, are we witnessing a broader reorganization of scientific labor, in which researchers retain the knowledge and judgment required to do science, but relinquish control over its direction, organization, and outcomes? The arrangement I’m concerned about is one in which a small group of individuals selects problems, sets priorities, and defines success, while researchers execute experiments, produce data, and generate artifacts. This is close to what is known as **proletarianization**: workers losing control over the conditions and organization of their work, and becoming dependent on employers who control the means of production [3]. Automated systems assist with implementation, analysis, and reporting, but the researchers doing the work have progressively less influence over what the work is for. The researcher then risks **becoming an operator within a scientific production pipeline, rather than an autonomous investigator**.\n \n\n Don't get me wrong, I don’t feel blocked on being able to produce good research. I feel that the institutions organizing scientific work increasingly determine which questions are worth pursuing, how the work should be performed, what counts as productivity, and who gets to claim the resulting value. I can still exercise scientific judgment, but **that judgment no longer necessarily gives me influence over scientific direction.** \n That is a form of what Marx called **alienation**: the worker’s estrangement from the products of their labor, the labor process, other people, and their own creative capacities [4].Access to knowledge has become cheaper: with LLMs, one can acquire a working understanding of an unfamiliar field much faster than before. Implementation is also becoming cheaper: reproducing experiments, writing evaluation code, and exploring numerous ablations and variations no longer necessarily require the same degree of specialized programming experience. However, scientific judgment is not automatically becoming cheaper at the same rate.
\nKnowing how to run an experiment is not equivalent to knowing which experiment is informative. Knowing how to produce a plausible explanation is not equivalent to knowing if that explanation is well-supported by the evidence. Generating a paper-shaped artifact is not equivalent to producing a well-rounded and useful scientific contribution. The ability to produce more doesn’t tell us if we’re learning more, or if the conclusions we draw are justified.
\n There’s an uncomfortable complication, however: organizations don’t always reward the most epistemically valuable work. They reward what they can easily observe, measure, sell, and compare, while recognizing the value of a scientific contribution often requires substantial **research maturity**. If one researcher produces twenty impressive demos, while another researcher spends a week discovering that the underlying evaluation is invalid, the first researcher’s work may be more legible and easier to celebrate, even though the latter may have made a contribution far more consequential to our collective understanding. We are creating an incentive system in which producing convincing signals of progress matters more than establishing if progress has actually been made. That’s not evidence that expertise has become worthless, but it **is** evidence that **the market price of expertise, its organizational recognition, and its de facto scientific value, have diverged**.For years, I had a fairly clear picture of the scientist I wanted to become. I wanted to develop the kind of research taste that comes from spending years fighting with difficult problems, learning to recognize which questions are worth pursuing, and becoming confident enough in my own judgment to defend an unpopular idea. I wanted to publish work I was proud of, earn the respect of people whose opinions I valued, and eventually lead a research group where I could set an intellectual direction, mentor younger researchers, and help create the kind of scientific culture I was once so eager to join.
\nResearch has always been a huge part of my intellectual identity, and the collaborators and friends I met through it played an important role in my life. I loved being around people who were obsessed with understanding something, who would spend hours debating an idea, reading each other's work, or brainstorming through a problem. There was a sense of belonging and shared purpose that made all the years of hard work and PhD suffering feel worthwhile. Now I'm less certain that the profession taking shape around me recognizes or rewards that kind of growth, and I'm starting to question if the future I spent years working towards is still viable.
\n@misc{vasileva2026stranger,\n author = {Vasileva, Mariya I.},\n title = {Becoming a stranger in my own field},\n year = {2026},\n month = oct,\n url = {https://mariya.fyi/posts/stranger-in-my-own-field.html},\n note = {Blog post}\n}