Towards and Understanding of Data-Driven and Model-Based Managerial Logics

Extended Abstract Submission to AI@work conference, division Computer Science 

Towards and Understanding of Data-Driven and Model-Based Managerial Logics 

Gunnar Holmberg1,2 and Nicolette Lakemond1 1Linköping University, 2 Saab AB 

Currently, AI is transforming our world, society and industry. Infrastructures that are dependent on AI are expected to play an increasingly critical role in society. Early applications of AI were often standalone and had limited consequences if the application failed, but failures in current emerging complex intelligent systems in e.g. health care, autonomous driving, industry and banking, and other critical infrastructures may have disastrous consequences. Thus, an important focus in understanding AI and its impact on work and organizations needs to acknowledge the fact that AI is more and more embedded in complex systems. These systems are traditionally developed using a model-based logic. With the implementation of AI, a model-based logic is complemented with a data driven logic which are increasingly operating together. Until now, insights into the managerial consequences and the underlying logic is limited. This paper will explore if and how data- driven and model-based logics differ in relation to management aspects. We aim to identify such differences and their implications for the underlying value creation processes. An increased understanding of these differences will inform organizations, management and workers on how organizations can be managed so that they responsibly can contribute to the development and operation of complex intelligent systems towards and AI-enhanced society. 

In order to understand the implementation of AI in complex systems, a starting point is taken in the evolution, challenges and current state-of-the art in system building firms. These system building firms, i.e. system integrators, provide complex product and systems. They have since long faced distinct challenges related to complexity and systems integration and the simultaneous consideration of properties such as safety, security, reliability and cost. In order to address these challenges, they traditionally rely on relatively rigorous innovation processes guided by regulatory involvement that may stipulate working approaches and certification of work activities in the firm and its supply chain. Our own previous studies on the evolvement of avionics systems since the 1960s have shown that firms use architectural platform strategies to deal with a multitude of criteria and use a combination of management approaches for the implementation of functionality such as safety-critical controlled functionality and lately also with an increasing focus on combining with generativity and enabling data fusion, learning and AI within partitions in the same complex system. This development reflects the emergence of a new type of actor, i.e. data refineries that can perform an ordered set of steps to cleanse, select, shape and enhance data in order to create value from crude data. Part of data refining may be set of data processing elements. These tend to be connected in such a way that the output of one element forms the input to the next. This can provide organizations with access to more reliable and structured data sets that can be used for analytics and that allow the consolidation of data from multiple sources. The new data related actors, including data 

refineries, algorithm providers and training actors represent a new type of data-driven ecosystem. It rises as a complement to existing industrial ecosystems, that in contrast are mainly model-based. The specific aspects of such data-driven organizations and ecosystems are not yet addressed in management research, and consequently the important coexistence of model-based and data-driven organizations and ecosystems for complex intelligent systems is a blind spot in management research. 

In this paper we propose and discuss different premises underlying data-driven and model- based logics in relation to management aspects by making a distinction between the organizational, the organization’s ecosystem, and the individual workers’ (i.e. engineers) level. and Building on current managerial knowledge in system-building industries that is largely based on a model-based managerial logic, we conceptually explore differences in logic when these systems also become data-driven in development and operations. The managerial logics are outlined and discussed in terms of characteristics on different levels: i.e. the organizational, the organization’s ecosystem, the individual engineers’ level. The results will contribute to a theorizing of management of complex systems in the wake of AI. As AI, as a set of methods (e.g. deep learning, data analytics), is driving the intelligent content of future complex systems, an increased understanding of model-based and data- driven logics will facilitate the adoption of AI in these systems and help to create prerequisites to fulfill safety and reliability requirements as well as ensure that these systems have beneficial outcomes for society.