Background

Applying machine learning techniques to large datasets (“big data”) opens up new possibilities for predicting future events. Such techniques are already employed by the police (predictive policing), in medicine (precision medicine) and in the insurance industry (personalised insurance). Algorithmic prediction is advertised as a promising means to distribute scarce resources more efficiently, thereby benefiting the community as a whole. However, the downsides of algorithmic prediction have become more and more apparent and cannot be ignored. For instance, it can reinforce existing prejudices and power imbalances.

Project content

As value networks are becoming more complex, companies use algorithmic prediction tools more often (so-called “predictive risk intelligence” or PRI) to predict the probability for a malfunctioning of machines and infrastructure and to then take pre-emptive measures (predictive maintenance). The latest generation of PRI software even goes a step further and claims to be able to predict even social events such as demonstrations, rallies or strikes. This touches on questions concerning co-determination within and across companies. Currently, relatively little is known about how companies use PRI to influence and steer processes of co-determination – whether they try to undermine it or whether (taking an optimistic view) they use PRI to reach out before an imminent conflict escalates. It is also unclear which stance trade unions and employee representatives should take on PRI and how they might use it to their own advantage. In this project we intend to close these knowledge gaps.

Goals and research questions

Algorithmic prediction plays an ever more important role for co-determination in value networks. However, there are many unknowns that lead to the following research questions:

• In what ways do PRI and other algorithmic prediction techniques affect co-determination in value networks within and across companies?
• How are PRI technologies developed? Which actors are involved in the development and which perspectives are considered?
• For what purposes do companies use PRI and similar technologies? How do or should trade unions and employee representatives react when companies use or plan to use them?
• Is it possible for employees to avoid (at least in part) being tracked and monitored by such a PRI system?
• Is it possible to apply PRI and similar techniques to social media data in ways that strengthen the position of employees and support their interests?

Methods

To deepen our knowledge on PRI techniques, we not only consider scientific databases, but also media reports and software provider documents (e.g., webpages of companies, press releases, interviews with founders of companies, sector reports, videos, tweets, etc.). The latter can help to understand how PRI technologies are changing and evolving.

Another important component of data collection are qualitative, semi-structured interviews with the following groups: software developers, key account managers, management of software companies, supply chain managers in group of companies (e. g. logistics, automotive industry), workers' councils, representatives of labour unions, scientists and employees.

Throughout our project, we pursue different strategies of knowledge transfer. Specifically, we conduct two multi-stakeholder workshops in which we discuss research findings and possible uses of PRI technologies. In addition, the progress made and the insights gained are published throughout the project's duration via blog posts and newspaper articles.

Results

A better understanding of the ways in which algorithmic prediction is used in business and industry is a major prerequisite for further developing concepts as well as the organisation and the use of new models of co-determination. It is therefore expected that the outcomes of this research project are not only of interest for scientists in organisational and technology research, but also relevant for different stakeholders involved in processes of operational co-determination.

The results obtained not only give employees, employee representatives, the media and NGOs better and deeper insight into the workings of algorithmic prediction techniques; they also help them recognise their potential for applying them in the interest of employees. In addition, we expect the findings to be beneficial for developers of PRI technologies, since they make it easier for them to identify ethical problems and potential violations of regulations at an early stage, thus allowing them to look for and find solutions that are in line with labour rights, for instance.