‘Can’t Help Myself’: on generative AI,the performance of qualitative research and slow scholarship

As the lights of the museum dim, an enigmatic mechanical dance begins. ¹ The solitary performer, an industrial robotic arm, relentlessly strives to accomplish an impossible task. It is Sun Yuan and Peng Yu's renowned installation, ‘Can’t Help Myself (2016–2019)’, an artwork that mesmerises, yet simultaneously disconcerts with its unending Sisyphean struggle. Encased within a transparent plexiglass box, this complex piece of machinery comes to life, ceaselessly interacting with a pool of a deep crimson, blood-like fluid. Sensors guide the robot's movements, diligently detecting the flow and sprawl of the viscous liquid. Its mechanical gestures – swooping, dabbing, guarding – are precisely choreographed in a desperate attempt to contain the spreading liquid, a task that ultimately, yet fascinatingly, reveals itself as an exercise in futility.

The arena in which the robot operates is both its stage and its prison. The work conveys a sense of haunting surveillance, where every motion is tracked, every action a response to the escape attempts of the indefatigable fluid. The seeping liquid teases and defies the robot, presenting a relentless challenge to its predefined boundaries and programmed control.

Within the enigmatic play of movements, a narrative emerges – rooted in the very heart of human existence. ‘Can’t Help Myself’ raises profound questions about control and boundaries, surveillance and futility. It embodies a struggle that seems doomed from the onset, a mechanical representation of the timeless human conflict between order and chaos, control and surrender.

Drawing upon the fundamental dichotomies of containment and freedom, surveillance and oblivion, ‘Can’t Help Myself’ illustrates the perpetual tension between our desire to control our surroundings and the reality that certain elements remain defiantly uncontainable. In the face of the sprawling liquid, the robotic arm continues its task, undeterred by the apparent futility of its mission, and so the dance continues – a mechanical ballet that encapsulates the beauty, complexity and paradox of this unending struggle.

This riveting display of tension between the static and the dynamic, the controlled and the uncontrollable, seems to echo a sentiment much beyond the artistic realm. As the dance of ‘Can’t Help Myself’ unfolds, it mirrors a similar struggle taking place in another arena entirely, one far removed from the silent chambers of a museum – the discipline of qualitative research in the age of artificial intelligence. But that is a parallel yet to be drawn.

Similarities

Just as the robotic arm in ‘Can’t Help Myself’ attempts to control the relentless flow of hydraulic fluid, qualitative researchers today are grappling with a similar challenge. They find themselves in a technological arena, where a torrent of AI tools and services – like ChatGPT, Scite.AI, Jamie, Claude and others – are in constant flux, spreading and transforming the landscape of research. What is possible, the temporality of research production processes, and so on.

The research domain, much like the robotic arm's transparent enclosure, presents boundaries that these tools continuously test. Researchers use these AI-driven technologies to contain and make sense of the expanding universe of information (see, e.g. Gamieldien et al., 2023), just as the robot ceaselessly strives to control the spread of fluid. AI tools offering capabilities such as advanced language modeling, citation generation, source identification, transcription, and more, increasingly shape the contours of this research environment, much like the liquid in the installation.

There is an uncanny resemblance between the futile struggle of the robotic arm and that of researchers trying to keep abreast of the rapid developments in AI. Despite their best efforts, some ‘splatters’ inevitably escape the field of view, a new update, a novel feature, or an entirely new tool springs up, expanding the boundaries and challenging the control exerted by the researchers. How long will it take until something gives way? What about those who can’t keep up?

The observed struggle of the researcher isn’t solitary, but rather meticulously surveilled, echoing the central premise of Shoshana Zuboff's influential work on ‘surveillance capitalism’ (Zuboff, 2019). In ‘Can’t Help Myself’, each action of the robotic arm is closely monitored, a poignant metaphor for the contemporary researcher's interaction with AI tools.

These AI tools, equipped with learning mechanisms, are far from passive; they learn and adapt from each interaction, effectively ‘observing’ the user's behaviour. So when we use tools like ChatGPT to generate ideas, analyse interview transcripts, or write texts, each interaction is stored and fed back into the machine to inform future conversations, coding and articles. Such surveillance and adaptation, as Zuboff (2019) argues, are central to the new form of capitalism, where personal data is the new currency. While these tools serve as research aids, then, they also play a part in this data-driven economy, not only shaping the landscape of research but also contributing to the expanding sphere of surveillance capitalism.

The futility represented in ‘Can’t Help Myself’ resonates strongly in this context. The task of remaining completely up-to-date in this rapidly changing AI-driven landscape may seem daunting, and at times, futile. However, just like the relentless robotic arm, researchers adapt, refine their strategies, and persist in their tasks.

The dance between the researchers and AI tools – much like the robotic arm and the liquid – draws attention to the tensions and challenges of control, surveillance, and boundaries in the era of burgeoning AI technology. As this dance continues, one can’t help but question: what are the implications for the future of research? Just as ‘Can’t Help Myself’ offers no easy resolution, the dance in the technological arena continues, suggesting a complex, uncertain, but undeniably fascinating future.

Anxieties

I’m confident that it is not just me, but I do feel a sense of unease at the thought of deploying generative AI – like ChatGPT – in my research practice. It dawned on me a couple of times over recent weeks, even after the majority of social media posts and discussions with colleagues about what it all means for teaching died down. Rather, the feeling was present when I attended a webinar from a well-known qualitative data software provider on using ChatGPT in qualitative analysis. And when I received a call for papers via email soliciting abstracts on the theme of ‘AI as a collaborator and companion in the social sciences and humanities’.

The feeling is tied to a sense of having to keep up with all the developments, while also keeping up with developments in research, the day-to-day tasks of work, maintaining a social life, etc. If publish or perish was the unifying call in academia for the past however long, what is it now? Adapt to AI or be overwhelmed?

We can’t help ourselves. Or at least, a discussion I’ve had with many colleagues over has it. While many people I speak to are also uneasy or haven’t had the time to fully think through what it means to adopt these tools, there is an implicit understanding that if you aren’t ‘working smarter’ you’ll be left in the wake of those who are.

Our unease isn’t a sign of defeat. Rather, it is a marker of our awareness, our acknowledgment of the shifting sands beneath our feet. Much like the robotic arm in ‘Can’t Help Myself’, we continue our dance with the elusive, shifting boundaries of AI – a dance of apprehension, adaptation, and underlying pressures to evolve.

But as those pressures to evolve multiply or the effects of AI tools become increasingly irresistible (because they have for many tasks and will continue to do so), we mustn’t lose sight of those things that happen in the background but are central to the ongoing refinement of generative AI platforms and so on. For instance, in mid-2024 a deal between Taylor and Francis's parent company Informa and Microsoft means the technology giant will gain ‘non-exclusive’ access to published content from around 3000 academic journals (Palmer, 2024). And while scholars are not unused to the power imbalance that sits at the foundation of commercial academic publishing, this raises serious questions around the ethical implications of these tools in the sector. All the more reason, as I will discuss below, to emphasise a slowness in this rapidly changing landscape.

However, this is not simply a tale of doom and gloom. The robotic arm in ‘Can’t Help Myself’, despite its apparent futility, cannot surrender – it adapts, learns, and persists. Likewise, many researchers will harness the power of AI tools to expand their abilities, turning what might seem a daunting task into an opportunity for growth and exploration. In the dance with AI, the key to flourishing lies not in resistance but in embracing the rhythm of change. The new dictum could then evolve to Adapt with AI and thrive.

Opportunities

Of course, with generative AI comes with what many would consider great opportunities. This article, details of the authorship of which will discussed momentarily, was produced far quicker than anything I’ve written in recent years – including referencing and formatting.

The first section was produced by ChatGPT-4, based on prompts and conceptual guidance provided by myself, curated to tell the narrative I had in mind. Section two was co-authored. Again, ChatGPT produced the majority of the text and I came in with an editorial eye to accentuate the most accurate and relevant material (and remove that which was irrelevant). The third section was more personal and as such was primarily authored by me. However, I used the AI assistant software Jamie to dictate a first draft. Finally, this section was written entirely by me using a keyboard to type each word, which took about as long as it did to produce the rest of the article. Additionally, Scite.ai was used to identify relevant sources that could be cited once a full draft was written. In all, the process took around 45 minutes (ok, I am being slightly hyperbolic, but it certainly took no longer than an hour and a half to produce the first version).

But to what end? The article accurately conveys my thoughts on the adoption and use of generative AI in qualitative research and has demonstrated (at least to me) the potential use case of ‘speeding up’ my ability to produce a linear text. But I’m not sure doing things quickly is how I want to perform my research practice.

In any case, doing things quicker and with more efficiency often comes at a cost; this time is no different. A recent article in Nature highlights the environmental toll of the AI revolution, which will increase its demand for energy and water as the curve continues skyward (Crawford, 2024). Beyond the operational pressures, however, we must also consider what processes other than writing essays, summarising papers and compiling reference lists are being optimised and to what ends. For instance, a recent partnership between Microsoft and ExxonMobil is said to make performance efficiencies that could lead to production increases of up to ‘50,000 oil-equivalent barrels per day’ (Coleman, 2023). I don’t want to get into an argument about engaging in the world we live in, but we cannot be naïve to the flipside of the page as we and our institutions take on and adopt these technologies in our research practices.

Sociologist Les Back advocates for the art of ‘slow scholarship’ in the context of sociological and anthropological research. He emphasises the importance of taking time for deep reflection, careful listening, and thoughtful writing, which he argues are often lost in the rush to publish and the pressures of academic life. We might add to this now, the futile practice of keeping up to date with generative tools. This approach is particularly encapsulated in his book The Art of Listening (Back, 2007), where he calls for researchers to truly listen to their research participants to foster empathy and deep understanding.

Slow scholarship, according to Back, is a deliberate attempt to resist the accelerating pace of academic life and its associated pressures. It invites scholars to pause, to delve deeper, think critically, and to reflect on their research processes. By doing so, they can produce richer, more nuanced analyses and insights, capturing better the ‘liveliness’ of social research (Back, 2012; Back and Puwar, 2012). Sociology and social research is, after all, alive!

Temporalities

But slow scholarship is not, or at least should not only, be considered so literally as about lessening the pace of academic life, which is under increasing pressure in the context of the metrics driven ‘neoliberal university’ (Hartman and Darab, 2012). It is also about creating space for intellectual curiosity, for serendipitous discovery (Busch, 2020), and for the thoughtful interpretation of data. And sometimes, paradoxically, this requires us to move quickly, such as when we are attending to the needs of our collaborators (Mason, 2021) or working on time sensitive topics. This is to say, that qualitative research can neither be slow nor fast per se (Vostal, 2021), but we would do well to resist the urge to do things more quickly, as a matter of course, whenever new tools make it possible to do so.

Perhaps most importantly in the context of this essay, slow scholarship challenges paradigms like publish or perish – or its contemporary avatar, adapt or be overwhelmed – by resurfacing timely questions about the quality and depth of academic work, the very role and responsibilities of a researcher, and the impact of time on knowledge production.

I take this ‘slow’ view of scholarship as a reminder that the researcher's role extends beyond data collection and publication. As grand as it sounds, we are custodians of knowledge, responsible for ensuring that the insights and findings generated from our work are both meaningful and impactful to and for the communities they engage. Attending to the temporal dimension of knowledge production in our practice is therefore crucial, and will be moving forward in the context of increasing automation, for some as acts of resistance and others a mechanism of checks and balances – prioritising AI tool usage as a mechanism for efficiency and depth rather than convenience.

Slowly onward

Back's ideas on slow scholarship represent a call to resist the pressure to constantly adapt and keep up with new technologies, methods, or theories. Instead, he suggests slowing down and taking the time to thoroughly explore, understand, and reflect on one's research practice. Perhaps, following this logic, the robotic arm of ‘Can’t Help Myself’ can be cast in a new light?

‘Can’t Help Myself’, at its core, could also be interpreted as a representation of careful deliberation, reflection, and depth, carrying out the principles of slow scholarship in a subtle but profound way. The artwork's robotic arm, while engaged in a seemingly relentless and futile task, is in fact executing a dance of precision and contemplation. Each of its movements, while swift and ceaseless, is also deliberate and carefully calculated. It mirrors the slow scholar who, while navigating an ocean of knowledge, chooses to develop their practice with, consistency, careful consideration and intentionality. Ironically, despite the hectic nature of the artwork, as an exhibition item it has nothing but time.

Following Back and others, I want to advocate a perspective where these advanced technologies serve not as accelerants to quicker publication or faster analysis as ends in themselves, pushing us towards the edge of overwhelming information flows, but as tools to foster the deliberation, reflection, and depth inherent in slow scholarship and the richness of what rigorous qualitative research has to offer. Many of us are still figuring out what exactly this means or how it will affect our performance as qualitative researchers, but we are where we are. So, onward! Slowly.