Weber, J., & Prietl, B. (2022). AI in the age of technoscience: On the rise of data-driven AI and its epistem-ontological foundations. In The Routledge Social Science Handbook of AI (pp. 58-73). Routledge.
As I engaged with the meta-theoretical foundations of knowledge-making for scaffolding methodological development in my doctoral thesis, I increasingly began wondering what were the underlying set of assumptions (called inquiry paradigm by Lincoln & Guba 1995, 2005; also research paradigm, Pretorius 2024) which informs knowledge products spat out of a machine in response to a prompt. The dimensions of these ‘sets of assumptions’ are recognized as epistemology (the nature of knowledge), ontology (the nature of reality), axiology (the nature of values) in addition to methodology which together encapsulate a ‘research paradigm’; it lays the foundation for any systematic program of knowledge-making. Here, in the age of generative AI, when pop-ups in academic publishers’ websites offer to ‘make sense’ of the article you clicked on, it raises the question: what are the assumptions on knowledge and reality and values which underlie the knowledge product constructed by a machine? What is it’s research paradigm? How might I navigate any incommensurabilities between it’s assumptions and mine? Or, is the legacy dominance of positivism being imposed on us, regardless of our own paradigmatic preferences (e.g. constructivism or pragmatism) simply as a result of using past data to predict future content?
The book chapter I link to here by Jutta Weber and Bianca Prietl offers me some answers to these questions; and, in a way, confirms my worst case scenario.
Long established standards of knowledge product hegemony will be imposed on us by a ‘black box’ unable to pinpoint or articulate its own assumptions on epistemology, ontology, and axiology. The link above opens their chapter’s pre-print PDF while the entire handbook is also available as such. The authors review both historical critiques of what is called “AI” as well as contemporary critiques.
the concept of big data promotes the idea of a “data-driven rather than knowledge-driven science”. It thereby renews “the primacy of inductive reasoning in the form of a technology-based empiricism” and idealizes a data fundamentalist approach to knowledge production, as critical data scientist Kate Crawford (2013) has called it. (Weber & Prietl 2022:63)
Crawford’s viewpoint of data fundamentalism rests on two premises, both of which are equally controversial according to Weber & Prietl (2022:63):
first, “the belief that life can be captured and modeled by data or even fully transformed into it” and, second, the assumption that objectivity is the result of a subject-free and therefore neutral production of knowledge. Both ideas have been heavily criticized within STS as constituting specifically modern ideals of science (Weber & Prietl 2022:63)
It is the second premise which speaks directly to the concerns I raise above on generative AI’s inability to inform me what set of assumptions and/or which research paradigm underlies it’s knowledge product. I look forward to research on how well the AI summary tools which now pepper academic publisher websites deal with knowledge produced by a variety of research paradigms and their associated assumptions. As Weber & Prietl (2022) point out “scholars engaged in Critical Data Studies have more recently revived the debate on how to understand data as a sociocultural, and highly political, construct” (p. 63). All knowledge, is first of all, local knowledge, say Okere, Njoku & Devisch (2005). Haraway may call this situated knowledge however its situatedness in her estimation is distinctly different from indigenous epistemologies as outlined by Gegeo & Gegeo-Watson (2001). That is, there is also an entire landscape of knowledge systems which originate outside the Euro-Western academy and its dominant paradigms (Chilisa 2019) and continue to remain actively generative in their own societies.
Another interesting observation can be found on p. 64, where Weber & Prietl (2022) point out what gets lost or left behind when easily ‘manageable’ data is the only source of insight/information for a machine-generated knowledge product: Relationity. They call it care and concern, and emotional work done by ‘women and ‘other Others’ (quoting Haraway) but it boils down to the invisible interactions which underpin knowledge-making in human societies who chose to retain the relationship between the mind and body (i.e. where the Cartesian divide does not hold sway; see Kuokkanen 2000). Relational axiology underpins epistemology and ontology in indigenous research paradigms, theorized and constructed separate and apart from the dominant research paradigms of the academy (see Klein 2023:46 on Chilisa 2017, 2019).
the call for numerical representation is considered to favor mechanistic conceptions of the world that focus on singular codifiable components of reality, ignoring their interrelatedness, complexity, variability, and disparity (Weber & Prietl 2022:64)
“Numerical approaches to knowledge production have further been criticized as being less suited to capture relations of power, inequalities, or cultural and symbolic phenomena” (Weber & Prietl 2022:64). To sum up Weber & Prietl’s (2022) in this section, one can say that data-driven machine-generated knowledge products (the chat of the bot) perpetuates every trope and stereotype of coloniality, that which Santiago Castro-Gomez (2021) calls the hubris of the zero-point gaze. That is, my inarticulate concerns, because this domain of study is not my field, are being unfolded and illuminated by Weber & Prietl (2022) lucidly in their chapter.
Remembering the lack of sensitivity toward power asymmetries and social inequalities of data-driven AI, one of the authors (Prietl 2019a, 10–21) argued that the current approaches in AI development run the risk of mistaking the perspective of a privileged view for a universal perspective, rendering those in marginalized positions (once again) invisible. (Weber & Prietl 2022:66)
Interestingly, Weber & Prietl’s (2022) discussion leads them to conclude that a stronger more positivist form of positivism is being perpetuated; where positivism is possibly the very first formal set of assumptions on ontology and epistemology to underpin the ‘scientific method’ (Chilisa 2019:45). At this point, perhaps the best way to conclude my own exploration of this domain is with an observation from management scholars who promote the idea of multi-paradigm inquiry as a means to bring multiple even conflicting viewpoints and perspectives to bear on complex (human/social) phenomena:
…study of phenomena such as sensemaking, meaning construction, power, and conflict becomes very awkward to handle using any immutable objectivist framework (Gioia & Pitre 1990:587)
This confronts directly the limits of today’s machine-generated knowledge products; at least and in so far as until the knowledge produced comes accompanied by its assumptions on ontology, epistemology, and axiology.