Can robots possess knowledge? Rethinking the DIK(W) pyramid through the lens of employees of an automotive factory

The increasing integration of robots into various industries, particularly automotive manufacturing, necessitates a reassessment of the concepts of data, information, and knowledge in human-robot collaboration. This study focuses on the practical and scientific need to revise these concepts from the perspective of employees working alongside robots. The practical need stems from the desire to enhance the productivity of robotized factories by improving robots' cognitive abilities and transparency, and by ensuring effective task division between humans and robots. The more effectively robots can process and utilize data, information, and knowledge, the more complex tasks they can perform, leading to increased efficiency and higher-quality products. Transparency is crucial for understanding robots' functions and decision-making processes. The scientific motivation concerns the DIK(W) pyramid, a fundamental model in knowledge-related disciplines. This model, however, has been criticized for lacking empirical support and neglecting tacit knowledge. This exploratory study aims to refine the DIK(W) model and apply it as a tool for understanding employee perspectives on robots and knowledge in a human-robot co-creation setting within the highly automated Valmet Automotive factory in Finland. The research questions are: 1) Do employees believe robots possess knowledge? 2) What types of knowledge do employees attribute to robots (or deny)? 3) Are employees who believe robots possess knowledge more likely to trust them and view them as teammates?

Existing research on human-robot interaction has indirectly addressed knowledge-related aspects. While studies have investigated trust and teamwork in human-robot collaborations, limited empirical work directly explores employee beliefs about robots' knowledge possession. Most research focuses on sectors like healthcare and military, with fewer studies examining the automotive industry despite its extensive use of robots. The DIK(W) pyramid, while foundational, requires further empirical validation. The definitions of data, information, and knowledge remain debated, with some scholars emphasizing the role of meaning and interpretation in the transition from data to knowledge. The human-centric view of knowledge in the DIKW pyramid has been critiqued for neglecting tacit knowledge and non-empirically measurable dimensions. This study addresses these gaps by directly investigating employee perceptions of robot knowledge within a robotized factory, using the DIK(W) pyramid as an analytical framework.

An empirical exploratory study was conducted in 2019 at Valmet Automotive, Finland's most robotized factory. A survey was administered to 269 employees, focusing on their views of robots' knowledge capabilities and their trust in robots. The survey included questions about employees' roles, views on teamwork (human-only, human-robot, or other), a set of arguments regarding robots and human performance (Table 1), and views on the future importance of working with robots (Table 2). The quantitative data (Table 2) were analyzed using descriptive statistics, cross-tabulations, and the χ² test to determine the statistical significance of differences between employee groups (those who believe robots can versus cannot possess knowledge). Qualitative data from open-ended survey questions regarding robots' knowledge were analyzed using content analysis, identifying key themes and categorizing responses based on the actor (robot, human, or both) associated with knowledge. This allowed for exploration of employees’ understanding of what constitutes knowledge in the context of human-robot collaboration. The survey was designed to address three research questions: (1) Do employees believe that robots possess knowledge? (2) What kind of knowledge, if any, do employees believe robots can possess? (3) Are employees who believe robots possess knowledge more likely to trust them and view them as teammates?

Regarding the question of whether robots possess knowledge, approximately 54% (n=135) of respondents answered affirmatively, while 46% (n=116) answered negatively. There were no statistically significant differences in responses based on gender, age, work experience at Valmet Automotive, or the specific department. However, there was an almost significant difference between experts (60% believing robots possess knowledge) and workers (52%). Regarding the type of knowledge attributed to robots, responses were categorized into three groups based on the actor possessing the knowledge: (1) knowledge not linked to any specific actor (data), (2) knowledge attributed solely to the robot (information), and (3) knowledge requiring human involvement (knowledge). The first category (n=20) included descriptions of data like codes, parameters, and sensor readings. The second category (n=15) included descriptions of robot activities like memory tracing, location awareness, and error detection, which suggests information processing capability without human interpretation. The third category (n=62) predominantly involved human programming as a means of imparting knowledge to robots. Those who denied robots' knowledge-possession provided justifications falling under three main themes: a hierarchical human-robot relationship where robots are tools controlled by humans; robots' inability to perform novel, creative, or contextually adaptive actions (the limitations of repetition); and the lack of self-awareness and independent cognition in robots. Regarding trust, 55% (153) of respondents trusted robots, with only 13% (35) expressing distrust. A statistically significant relationship (p=0.002) was found between believing robots possess knowledge and trusting them. No significant relationship existed between believing robots possess knowledge and considering them teammates.

The findings challenge the traditional, human-centric view of the DIK(W) pyramid. The survey results indicate that a significant portion of employees view robots as possessing knowledge, albeit in collaboration with humans, challenging the assumption that knowledge is solely a human attribute. The study highlights the importance of acknowledging both human and robot roles in the knowledge creation process. The data-information-knowledge transformations are bidirectional, with knowledge being transformed into instructions for robots and robots' actions providing new information that can shape human understanding and further knowledge creation. The concept of knowledge acts as a divider between employees based on their understanding of the actor independence needed for knowledge possession. Those who see robots as co-creators of knowledge generally show higher trust levels in robots, signifying the role of shared understanding in human-robot collaboration. The limitations of existing robots in terms of creativity, context-awareness, and independent problem-solving are also evident. These insights emphasize the critical role of human oversight and interpretation in the human-robot knowledge co-creation process.

This study offers both theoretical and practical contributions. Theoretically, it provides empirical evidence for revising the DIK(W) pyramid to incorporate human-robot collaboration, moving beyond the traditional human-centric approach. The revised pyramid acknowledges the distinctive roles of humans and robots in data, information, and knowledge processing. Practically, the findings emphasize the need for transparency in robot design and operations to foster trust and collaboration among employees. Understanding employee perceptions of robot capabilities is crucial for successful implementation and optimization of human-robot teams in manufacturing settings. Further research, including qualitative and ethnographic studies, is needed to deepen the understanding of knowledge co-creation in various human-robot collaborative settings.

The study's reliance on a single factory limits the generalizability of the findings. The low response rate (6%) raises concerns about potential sampling bias. While the survey provides valuable insights, a more comprehensive approach with larger sample sizes and diverse industrial settings is needed for more robust conclusions. The use of self-reported data for both quantitative and qualitative analysis introduces potential subjective biases.