HPAO – A Human Preference Aware Optimization System

An increasing process digitalization  allows companies to gather vast amounts of data about processes, such as throughput time or mean time to failure  . Modern machine learning techniques make it possible to analyze these data in more detail  so as to optimize the underlying processes. AI-based analysis of processes, however, does not just hold the promise of efficiency gains but also will allow employers to assess in detail individual employees’ behavior and productivity.

AI in industrial processes (and in other work environments) thus raises the spectre of massive increases in employer control, eroding the power of workers and employees within the enterprise and in society. Therefore ethical concerns regarding a loss of autonomy in the workplace has been promoted. Among the ethically problematic possible consequences are reductions in occupational health and safety, as constant monitoring increases stress levels, and the risk that excessive transparency about work performance might expose an employee to   direct or indirect pressure. Labor unions, scholars, and various other observers suggest that these ethical concerns can largely be traced back to a key element of (actual or anticipated) use of AI: It aims to determine how employees might adapt to better serve the process.

Our goal is to point out ways to utilize AI in an ethical manner and to move the employee back into the centre of process design. The human preference-aware optimization system supports employees by assigning tasks related to their past preferences and so promotes their strengths. More satisfied by appreciating individual strengths, employees motivation will rise as they work on tasks they prefer. In addition, AI based considerations can be reduced, since employees benefit from.  Such a system would use AI in data analysis with the objective of optimizing the processes via the assignment of tasks to suit the employees, rather than changing the employees workflow to suit the process. The envisaged AI optimization system will be trained with process data and should only be capable to assign suitable tasks but not be able to judge an employee. For example, some people might prefer to sit and wrap small items while others like to stand up and move, which leads to a higher productivity wrapping bigger items. Intelligent, preference-aware optimization systems would assist employees and the management in the assignment of tasks in a way that is respectful of individual differences and can safeguard and maybe even strengthen employee autonomy. More broadly, the results of this research project are guidelines and design rules for applying AI in human centred processes considering the ethical perspective, leading to a development guide. This handbook will support AI-developers by explaining how to access process metrics, what kind of AI is well suited for the process and what challenges the developer is faced with.

TUM School of Governance, Chair of International Relations

The Research Project is funded by the Institute for Ethics in Artificial Intelligence (IEAI) of the Munich Center for Technology in Society.