Ceci n'est pas une école: Discourses of artificial intelligence in education through the lens of semiotic analytics

Abstract

New ideas and technologies enable new ways of doing as well as new forms of language. The rise of Artificial Intelligence (AI) is no exception. The implications of changing activity and language take on new gravity in certain fields to which AI is applied, such as education (AIEd). Terms like smart, intelligence, and learning, which had certain meanings when describing human cognition, take on new meanings in the context of computational systems, with the potential for polysemy when the human and computational meanings meet. This article unpacks what AIEd companies mean when they use these terms. Drawing on findings from a mixed‐methods study, we first describe how AIEd companies used these terms on two websites. Then, using Natural Language Processing techniques, we quantitatively analyse a corpus of over 65,000 words scraped from 26 AIEd company websites. Our analyses suggest that commonly promoted narratives around student learning and 21st Century skill‐building are not supported by the language on AIEd company websites, which focus instead on mass customization, efficiency, and monitoring—all tasks at which computers excel. Also, notably scarce in the corpus were extensive articulations on ethics. Given these findings we propose that although AIEd companies create powerful new technical possibilities, they must also be evaluated for the powerful ways in which they shape narratives around the use of technology in education and the behaviours and capabilities that constitute education.

The Effect of Interpretable Artificial Intelligence on Repeated Managerial Decision-Making under Uncertainty

“…Mastery learning has been an important starting point for many AIEd systems, and it at least implicitly underpins the common belief that there is great potential to use AI for personalized instruction (Blikstein et al, 2022). AIEd is often contrasted with teacher-led lecturing and K-12 classrooms where information flows from the teacher and the textbook to the head of students.…”

Will Artificial Intelligence Enable Open Universities to Regain their Past Glory in the 21st Century?

“…It is based on the assumptions that we already know how human intelligence works, that learning is what education is all about (Eynon, 2023), that "all significant facets of student activity and the learning process can be captured in data form" (Selwyn, 2022, p. 622), and that such data enables educators to "get insights into students' progress and struggles" (Kizilcec, 2023). Attempts to automate education long predate the genesis of AI, starting from Sidney Pressey's teaching machines in the 1920s and continuing into the 21st century (Blikstein et al, 2022;Watters, 2021).…”

“…For example, in China, "AI has been seen as one of the solutions for the shortage of quality teachers in undeveloped areas" although the reality is far from the intention (Yuan, 2023). Interestingly, the equity discourse is popular in the grand narratives of official documents such as Miao et al (2021), OECD (2023), and the Office of Educational Technology (2023) while equity or ethics-related topics are rarely mentioned by major AIED companies according to Blikstein et al (2022).…”

Imaginaries and problematisations: A heuristic lens in the age of artificial intelligence in education

“…At the same time, AIEd companies such as Cognii offers 'virtual learning assistants' that work much like ChatGPT but that are 'designed and optimised for educational conversation' (https://www.cognii.com/technol-ogy), as well as AI grading tools that can, in turn, solve the 'problem of ChatGPT' (https:// twitter.com/cognii/status/1621547579728359424). Without getting into a full-blown analysis of the imaginaries and problematisations here, just looking at how Cognii's marketing indicate problem-solution configurations, such as 'student success', 'instructor productivity' and 'cost-effective and adaptive online learning systems and formative student feedback', clearly shows that not only the discourses (see Blikstein et al, 2022 for an example) need to be studied as problem-solution configurations but also the actual tools and their sociomaterial underpinnings, operations and effects. This includes the general problem that AI can be trained on data that is biased or flawed, perpetuating discrimination; that it black-boxes and legitimises certain models of reality and existing relations of power and that it may have significant ecological impacts (McQuillan, 2022).…”