It is now routine to talk about human-AI collaboration and the possibility of AI systems acting as co-workers or partners for us. It’s claimed that great advances in human-AI collaboration will increase productivity and / or augment the capabilities of humans. Such ferment is not just limited to tech companies’ public statements or a media that largely repackages them. It’s also an unstated assumption of much technology-oriented research. The Computer-Supported Cooperative Work and Social Computing (CSCW) conference that our research is published in has joined in too, albeit more critically. For example, a 2024 panel “Is Human-AI Interaction CSCW?” debated the “controversial and timely question of whether the purview of CSCW should be expanded to include research on ‘collaborations’ involving a single person and one or more AI models or agents”.
We think talking about ‘collaboration’ like this obscures the richly detailed, diverse investigations and conceptualisations of joint, coordinative human action around technologies and their infrastructures that CSCW research has revealed over time. Furthermore, if we act as though collaboration with machines is simply an extension of CSCW’s interests, we might risk detracting from deepening our understandings of how people and their social worlds actually intersect with (AI) technologies.
Our CSCW 2025 paper, “Opening Up Human-Robot Collaboration”, focusses on the robot-specific version of human-AI collaboration. We note that research on human-robot collaboration (HRC) is increasingly taking inspiration from and contributing to CSCW as it searches for ways to conceptualise what happens between people and robots, reflecting many of these wider trends in human-AI collaboration. But what our research finds is that people and robots don’t really ‘collaborate’ in any normal sense of the word. In fact, people are actually doing a load of work to bring about something that — from a distance — might just look like ‘collaboration’.
To make this concrete we can consider an example.
Let’s look at a moment of encounter between someone pushing a buggy (stroller) on a public street in the UK, and a delivery robot which is autonomously navigating these same streets to deliver goods to a customer. We captured this during our fieldwork, where we followed and video recorded robots as they went about fulfilling orders. We saw numerous situations where robot trajectories intersected with people on the street, forming a massive, pervasive class of fleeting encounters where robots and humans must ‘get on’.
We can break what happens down using the figure below.
The robot’s trajectory on the pavement (sidewalk) diverts it around a bin that has several extra binbags placed against it (see figure, A). The robot is thus driving towards the centre of the pavement space towards a buggy-pushing pedestrian (P8) approaching from the opposite direction. P8 seems to anticipate trouble by moving the buggy (and walking trajectory) to the hug the shop fronts (A-B). But the robot — reaching the middle of the pavement — continues, driving further towards the pedestrian (B). P8 moves even closer to the shop front edge, stopping completely to avoid collision (B-C). The robot straightens, drives immediately forward (C), and steers leftwards slightly (D), during which the buggy-pushing pedestrian then visibly blows out their cheeks in an exasperated manner (highly noticeable in person). P8 resumes pushing the buggy once the robot is clear.
We are not interested in focussing on how the robot’s design could take into account buggy-pushers and yield space for them. Nor are we interested in specious arguments that “technology will get better” and therefore dismiss such troubles. By all accounts the delivery robot service operates effectively and does the job. The point of the fragment is to instead show the irremediability of human and robot perspectives and the immensely intricate practical work people do — in this case—to accommodate them. CSCW’s interest in the inherent sociality of technology use in situ allows us a glimpse into the rich world of mundane reasoning that such robots are being deployed. The buggy-pusher visibly anticipates the need for managing space, making way for the robot and reconfiguring their use of the pavement as a ‘proffer’ for how passing could proceed. The buggy-pusher is generous with space (clear in figure). Yet, this does not work. They and the robot end up in a momentary turn-taking system where someone (or something) must go first.
Yet there is even more complexity here than first glance. While the robot seems to optimise pavement position on the basis of obstacles, there is a social, moral order of passing someone pushing a buggy that is grounded in how turn-taking may be done on this street, in this place, at this time. For the place we are investigating (UK), a buggy-pusher is a visual category that normatively would be seen to have rights to go first where space is tight. Yet it is they who yield for the robot. And they provide a bodily account (blowing out their cheeks) of the (social) objectivity of this norm in doing so. Such categories and their associated moral implications are simply irrelevant to robots.
To return to our opening arguments, our fieldwork teaches us that operational robots like these are unavoidably intertwined with the inherent sociality of everyday worlds, worlds that are replete with locally-organised, normative ways of doing things and ways of being held to account for such public actions. If we want to understand phenomena that purport to be human-AI collaboration we need to take the constituent details of this mundane action seriously — just as CSCW has shown us time and time again.
It’s not that we can’t call it ‘collaboration’, but rather that stopping at that point glosses over the immense complexity of everyday life with and around technology. This means actively talking about ways of being around robots that emphatically position human social practices at the centre, i.e., rendering what is often background ‘invisible work’, visible.