Oksana Dorofeeva - Not just a ride on the hype train: how machine learning researchers and practitioners navigate AI hype [STS Italia presentation]

Not just a ride on the hype train: how machine learning researchers and prac77oners navigate AI hype 13.06.2025 Critical Hype Studies: Towards a Collaborative and Unified Approach STS Italia 2025 Oksana Dorofeeva PhD researcher, Aarhus University & visiting PhD student, University of Edinburgh Where this talk is coming from •Looking into experiences and narratives in proximity to hyped technology •WiP based on an interview study into the narratives of ML work in the context of peak valorisation and thus also “ethical valence” (Slota et al., 2020) •Juggling STS and cultural sociology inspirations •Zooming in on self-presentation, self-identity & navigating buzzwords (Techno)hype and technologists •McKenzie’s trough (1998) •Growing literature on AI technologists’ relationship with hype (Horváth & Vicsek, 2024 ; Züger et al., 2023) including re: science communication (Gallagher et al, 2024) •Technologists’ relationship with hype often portrayed as contradictory 15 The Certainty Trough Donald MacKenzie How does intimate knowledge of technologies affect confidence in them? One might expect that those closest to these technologies would have the highest confidence in them, and those ignorant of them would be fearful. This expectation, however, turns out to be an oversimplification. The relationship between proximity and confidence is not a simple linear one, and several pieces of research suggest the more complicated pattern shown schematically in figure one. On the horizontal axis is 'social distance' from the technology in question. On the left are 'insiders', those directly involved in producing knowledge about the technology: for example, its designers, or those who conduct and analyze tests of it. Next along the horizontal axis come those who, while not insiders in this sense, nevertheless have a commitment to the technology in question. Many users of a technology would fall in this category. So (typically) would the senior managers of firms producing or using it. As we continue to move to the right of the horizontal axis, we pass through the uncommitted to those who are hostile to the technology in question. These latter -'outsiders' -might be those who are committed to a different, competing technology; they would also include those who distrust the institutions responsible for the technology at issue. high low directly involved in knowledge production committed to technological institution/program but users rather than producers or knowledge Figure 15.1 The certainty trough 325 alienated from institutions!committed to different technology R. Williams et al. (eds.), Exploring Expertise © Palgrave Macmillan, a division of Macmillan Publishers Limited 1998 (Buzz)words in hype + professionals •Buzzwords (Bensaude Vincent, 2014) and their affordances •Market categories launched and abandoned by consultants (Pollock et al, 2022) •Previous work connecting (over)use of ‘AI’ to hype mechanisms (Markelius et al., 2024) and reporting its problematisation by professionals (Gallagher et al., 2024; Züger et al., 2023) +hype, buzzword, etc. currently being ordinary language concepts! Work, worth & hype •Data science as a nascent technical elite (Avnoon, 2023; 2021; Burrell & Fourcade, 2021) •Professional narratives of worth and AI hype •Negative side of hype? AI controversies, ‘techlash’ experiences (Su, Lazar, & Irani, 2021)? ➡ joining contested re: moral worth elite occupations, e.g., advertising (Cohen & Dromi, 2018), law (Clair & Hunt, 2025), finance (de Keere & Burchartz, 2025)? In what ways does AI hype figure into the ways ML/AI researchers and practitioners make sense of their work and navigate professional self-identity and presentation? Methodology •In 2024: 44 interviews with ‘people who train models’ (though not all interviews touch upon this topic) •Academic (27), industry, and both; 18 DK / 17 NL / 9 UK •Narratives of ML work •Hypetopics sometimes arise without probing •But also from me asking about how they experience the developments in the field, their self-identification and how they present themselves, and of their opinions on terminology of ML vs. AI Experiencing hype •My interviewees note the hype: ‘hyped’, ‘buzz’, ‘hot’, ‘shiny new tool’ •Public imaginaries and expectations •Similar to Gallagher et al. (2024): attention and funding, faster pace, more competition and stress •‘Hype-driven’ research agenda Immunity to the hype as apart of expertise •Expectations vs. capabilities •More applied ML: what’s hyped (generative models) vs. what’s useful •knowledge and not ‘buying it’ as a boundary between insiders and outsiders We've seen customers approaching us saying they want to do AI. And what is often interpreted as AI is. now mostly language models. Coming from academia, I know AI is quite a lot more. So, deep learning, a small subfield of AI, is what is gaining all the attention, especially generative models. And the use cases in the real world are actually quite often for the more classical machine learning stuff. (ML scientist at a consultancy) I witnessed the web and the start of the Internet, and people started having desktops in their homes, set up and started using the Internet. At that time, anything anyone built, they said: it is computer-based. Now they call it AI-based. But 90% of these products have little to no AI in them. This is my impression. Some people buy it, but I'm not an outsider. (DL researcher in a medical company) Maybe other factors… •The AI act & other regulations affecting this I know that if the work is about AI, people need to write another ethics statement. About the use of AI and stuff. I think this is a slightly grey area right now because again, I don't perceive my work as AI, but outsiders might disagree with this. And is there a strict definition of this? I don't know. Right now in my grant applications, I try to not use the term AI. (computer vision researcher) So (what?) •Hype dynamics on the micro level •Hype & buzzwords at the boundary between ‘insider’ technologists and ’non-technical people’ •Focus on worth and valuation as an approach to studying hype? 15 The Certainty Trough Donald MacKenzie How does intimate knowledge of technologies affect confidence in them? One might expect that those closest to these technologies would have the highest confidence in them, and those ignorant of them would be fearful. This expectation, however, turns out to be an oversimplification. The relationship between proximity and confidence is not a simple linear one, and several pieces of research suggest the more complicated pattern shown schematically in figure one. On the horizontal axis is 'social distance' from the technology in question. On the left are 'insiders', those directly involved in producing knowledge about the technology: for example, its designers, or those who conduct and analyze tests of it. Next along the horizontal axis come those who, while not insiders in this sense, nevertheless have a commitment to the technology in question. Many users of a technology would fall in this category. So (typically) would the senior managers of firms producing or using it. As we continue to move to the right of the horizontal axis, we pass through the uncommitted to those who are hostile to the technology in question. These latter -'outsiders' -might be those who are committed to a different, competing technology; they would also include those who distrust the institutions responsible for the technology at issue. high low directly involved in knowledge production committed to technological institution/program but users rather than producers or knowledge Figure 15.1 The certainty trough 325 alienated from institutions!committed to different technology R. Williams et al. (eds.), Exploring Expertise © Palgrave Macmillan, a division of Macmillan Publishers Limited 1998