Responsible Engineering in the Age of AI: The Value of Responsible AI Education from Engineering Students’ Perspectives

AbstractBeing a critical enabler of research and development, data-driven systems like Artificial Intelligence (AI) are increasingly relevant to engineers. Due to their generalizability and wide-ranging functionality, they are closely interwoven with social developments. With it comes the responsibility for instilling the right values and the need to gain knowledge of AI and its implications for society. A master’s seminar at RWTH Aachen University trained engineering students in Responsible AI (ethical, social, and legal concerns with respect to AI) within engineering. To complement perspectives from industry and accreditation boards, we investigated students’ reflection papers on the course to determine the relevance that engineering students give to their Responsible AI education. We found that prior to the seminar, students lacked knowledge about AI applications in engineering and assumed that technology (including AI) was neutral and unbiased. Yet after the seminar, students reported having corrected these assumptions. They expressed their positive beliefs about the importance of Responsible AI education in engineering, insisting that future engineers should consider the sociotechnical context of their work. This paper presents the results of the reflection paper analysis to address why engineering students see learning about Responsible AI, including its sociotechnical context, as relevant for their future careers. MethodologyWe conducted a post-hoc qualitative analysis on 20 students' anonymized, ungraded reflection papers, written upon completion of the seminar on Responsible AI. We cyclically defined categories and subcategories for analysis and coded our students' papers. To this effect, we used MAXQDA qualitative data analysis software to help organize and filter our segments, which we manually coded following Kuckartz' (2014, 2019; Kuckartz & Rädiker, 2019) methodology. We did not use any AI to aid in our analysis.The length of segments was not explicitly limited; we agreed that a coded segment should include what was necessary to view it out of context and still understand the student’s argument and why it fits in the assigned category.For more details, please see our paper: Responsible Engineering in the Age of AI: The Value of Responsible AI Education from Engineering Students’ Perspectives, which has been accepted for publication in the SEFI Journal of Engineering Education Advancement.DataThe coded segments are organized by category and subcategory where applicable. (See "Overview - All Categories and Subcategories.xlsx" for an overview of the category structure and number of segments.) All discovered categories and subcategories have been included in this dataset, not only the ones analyzed in this paper. Therefore, all coded segments are included in this dataset.Segments assigned to each category and subcategory can be found in separate .xlsx files. All segments assigned to a subcategory are also assigned to its parent main category. Some subcategories are mutually exclusive while others may overlap. Segments can certainly have overlapping main category assignments, as can be seen in the final column in each uploaded file.These files were generated using MAXQDA and then modified for our purposes. They are available to preview, but legibility may be better when downloading the files.The tables in each .xlsx file are structured as follows:[category] > [subcategory]: the first row acts as a title, stating the (sub)category filter used to extract the segments exported to this fileTable Columns (row 2):Student: the anonymized author of the reflection paper from which the segment originates, randomly numbered 1 to 20, and cited in our paper as (student) s1 to s20Beginning: the reflection paper page number on which the segment beginsEnd: the reflection paper page number on which the segment endsSegment: the full segment, exactly as it appears in the student's reflection paperOther codes assigned to segment: MAXQDA's automatically generated list of (sub)categories that overlap with the segment. That is, at least part of the segment is also assigned to the other (sub)categories listed. The (Weight: 0) can be ignored; we never assigned weights, but these were automatically included in MAXQDA's output.References Cited HereKuckartz, U. (2014). Qualitative Text Analysis: A Guide to Methods, Practice and Using Software. Sage Publications Ltd. https://doi.org/10.4135/9781446288719Kuckartz, U. (2019). Qualitative Text Analysis: A Systematic Approach. In ICME-13 Monographs (pp. 181–197). Springer International Publishing. https://doi.org/10.1007/978-3-030-15636-7_8Kuckartz, U., & Rädiker, S. (2019). Analyzing Qualitative Data with MAXQDA. Springer. https://doi.org/10.1007/978-3-030-15671-8