Algorithmic Mangement: beyond the dystopia there are dangers too

5 min read

Over the last few years, the development and deployment of algorithmic management has drawn significant attention from workers and unions, academics and policymakers. The ideal of algorithmic management emerged wrapped around the promise that Big Data and Artificial Intelligence can automate many managerial tasks - if not entirely, at least to a considerable extent.

However, as Hilke Schellmann’s recent book, The Algorithm, shows in detail, the (promised) blessing of a new age of efficient, accurate and neutral managerial decision-making is rapidly showing signs of a curse for workers subject to biased, privacy-wrecking algorithmic tools. One recent, telling example comes from the deployment of so-called AI emotional detection software to monitor and assess workers’ performance at several fast-food chains in the USA. Alongside a vast scholarship from different disciplines, Schellmann’s book eloquently shows how this might be a terrible idea, starting with the fact that this technology is grounded on pseudo-scientific assumptions and has no scientific validity.

Perhaps slower than necessary, Europe Union-level policymakers have been developing regulatory instruments that can offer workers and unions tools to tackle some of these perils (GDPR, AI Act, Platform Work Directive). While these developments may be broadly seen as positive, several voices have highlighted how there is, nonetheless, no room to rest on this front. Drawing from my research, I consider this a call that should be amplified and incorporate not only the flashier algorithmic, quasi-automated decision-making tools but also those dedicated to “merely” tracking workers – even if not necessarily producing more than descriptive statistics for the interpretation of managers.

To better understand this, one must take a step back from the AI hype. As Ulrich Leicht-Deobald and colleagues thoroughly describe, algorithms may be predictive and prescriptive – that is, they may point in some direction or, as we have seen in platform work, directly enforce that direction and, for instance, penalize or even terminate a worker without necessarily running such decision through human supervision. Alarming as this is, algorithms can also be “merely” descriptive – that is, accumulating and sorting data to inform managerial decisions.

Notably, while quantifying workers’ performance has been around for centuries, the novelty of our era is how detailed this quantification can be. More than just rough numbers, like in the recent past, current technologies allow the combination of fine-grained metrics about every step of the execution of a given task with pervasive surveillance techniques, such as geolocation-tracking of workers’ movements. Two aspects of this development deserve special attention.

First, under the still-to-be-proved imperative of increasing companies’ productivity, we observe the gradual demise of workers’ privacy. This is a negative development since privacy is a crucial right of every person. More than that, in the context of work, this offers new opportunities for employers to spy on workers, track associational patterns, and surveil workers in general, as well as union activists and leaders, in particular. In my own research in the food-delivery sector, I found workers feeling the need to ask permission to move away from their motorbikes to go to the toilet, afraid they would appear is the eyes their manager sitting in the office in the other side of town as slacking during work time. Still in this regard, while union busting and discriminating against unionized workers – or any workers engaging with colleagues – is clearly illegal, it is something that still happens and constitutes an obvious source of fear among workers. Therefore, creating more opportunities for it to happen thanks to the refinement of surveilling tools will hardly make life easier for workers and their representatives.

Second, it is well-documented how surveillance curtails workers’ autonomy and how that, in turn, is often associated with occupational health and safety problems due to workers being rushed to fulfil their tasks. What algorithmic quantification and different tracking forms bring new is workers’ internalization of this imperative. Again, in my own research, this time in the grocery delivery sector, I have witnessed how some managers have afforded to take a step back from direct control over workers, only to act as “algorithmic coaches” – that is, mainly focusing on helping workers satisfy the algorithmic metrics’ imperative (for a similar conclusion see, Alex Wood’s report for the Joint Research Committee). What stood out to me was how workers internalized this imperative and, despite the exhaustion and even accidents experienced, would still happily comply with the visibly unrealistic goals imposed by their employer.

Under the current regulations, both these examples can hardly be tackled from a legal perspective – if not for something else, because they take place at a very micro-level, with evidence being challenging to produce. Overall, the two most significant risks are: first, that the increase of workers’ surveillance brought by algorithmic descriptive tools goes under the radar of legislators more focused on predictive and prescriptive applications of algorithms; and second, that these approaches are normalized and seen as inevitable, and therefore remain unchallenged by workers, unions and authorities.

In conclusion, the debate on introducing new technologies in workplaces needs to be open to interrogate the flashiest technologies but also apparently less sophisticated ones. In this vein, the discussion must cover what, beyond maximizing control, is the actual added value of each new technology and, if there is any true added value, the trade-offs that derive from it. At the end of the day, is productivity more important than workers’ well-being and respect for their fundamental rights?