Organizational Knowing and AI Research Highlight – Samer Faraj – Organizational Knowing after Large Language Models (LLMs)
At the Canadian AI Advancement Institute (CAIAI), our mission is to help Canadians learn, adapt, and succeed with AI through accessible AI education and training. Part of this initiative is to share the incredible work of AI researchers, thought leaders and knowledge experts to discuss the impacts of AI adoption.
In this blog, we discuss ongoing Canadian AI research and a recent journal article authored by Samer Faraj, a Canada Research Chair in Technology, Innovation & Organizing. Samer is the founding director of the McGill Institute for Transforming Healthcare, an associate member of the Department of Social Studies of Medicine, and head of the research group on Complex Collaboration. He has recently completed 10 years as the Director of the Faculty’s PhD program. His current research focuses on complex collaboration across settings such as healthcare organizations, knowledge teams, and online communities.
His recent publication, A Time for Monsters: Organizational Knowing after Large Language Models, written alongside authors Joël Perez-Torrents, Saku Mantere and Anand Bhardwaj, focuses on “large language models (LLMs) as foundational building blocks for more complex intelligent systems, including agentic and multimodal AI, that reshape organizational knowing” (Faraj et al., 2025).
This blog discusses a recent AI research article, A Time for Monsters: Organizational Knowing after Large Language Models, by Faraj and colleagues, which argues that the rise of LLMs is improving knowledge work while reshaping “how knowledge is created, evaluated, and acted upon in the age of intelligent technologies” (Faraj et al., 2025).
Their journal unpacks how large language models challenge two long-standing views of knowledge: Representationalism-based knowing, which treats knowledge as something that can be codified, stored, and transferred, and Practice-based knowing, which treats it as something earned through practice and engagement with the world (Faraj et al., 2025).
From there, the journal proposes thinking of LLMs as “Haraway-ian monsters”: boundary-crossing systems that destabilize familiar categories while opening new possibilities for inquiry (Faraj et al., 2025). As LLMs do not create, conceptualize, or act on knowledge the way humans do, it is important to understand how they function, what to expect, and how to ensure a human-centred implementation of Artificial Intelligence.
We can view knowledge through many lenses. Faraj and colleagues identify two concepts of ‘knowing’: representationalism-based knowing and practice-based knowing.
Faraj and colleagues argue that, despite their differences, both traditions share the view that knowledge production ultimately resides in humans, with technology serving mainly as a container, channel, or support.
LLMs challenge this. Unlike these traditional ways of knowing or sharing information, LLMs can generate content based on stored knowledge during an interaction without referencing prior experience; thereby transferring knowledge while removing the very essence that makes such knowledge human. This positions LLMs less as tools and more as active participants in knowledge work itself. It tears at the fabric of how knowledge is developed and distributed in an unprecedented way.
To address this, Faraj and his colleagues argue that organizations should not replace human knowledge work with AI systems; instead, they rightfully claim organizations should ensure systems are in place to govern how knowledge is created, validated, and distributed.
This is an important consideration for AI Adoption in any knowledge-intensive environment.
For most organizations and individuals, the advent of AI in their lives is welcome; however, Samer Faraj and his fellow researchers take a different position, critiquing LLMs through a Harawayian approach as monsters. To do so, they draw on Donna Haraway’s concept of entities that don’t fit neatly into categories and therefore force people to reconsider their assumptions.
Considering this approach, it is understandable that LLMs could be framed as “Haraway-ian monsters” because they blur boundaries that organizations rely on (Faraj et al., 2025).
Monsters can be unsettling because they disrupt familiar order; however, there are ways to minimize this unsettling feeling.
In order to minimize the challenges of deploying LLMs within organizations, the authors identify three strategies:
Firstly, organizations must understand the impact of LLMs on organizational knowledge, as LLMs often “overflow their conceptual boundaries [by] destabilizing [said] repositories of codified knowledge” (Faraj et al., 2025). This destabilization is often referred to as hallucinations.
LLMs frequently experience such hallucinations, which, within the organizational environment, are detrimental to stakeholders. It is important that organizations inquire into how the model was trained, its dataset, and its capabilities prior to implementation to establish “when and how machine-generated insights enhance inquiry and when they lead it astray” (Faraj et al., 2025).
Once these questions have been answered and an organization has implemented an AI model, Faraj and his colleagues examine the importance of continuously vetting AI outputs. They recommend a process they have coined “dialogical vetting”:
Through this process, organizations can become confident in an LLM’s outputs and consider using them in their operations. This is why AI literacy and prompt engineering are critical for organizations, as these skillsets align “closely with practice-based approaches to knowing”; without them, verifying AI outputs would be difficult and or impractical at scale.
For an organization that has invested in AI education and seeks to implement vetted AI outputs, it must consider who is responsible for it. “If knowledge is no longer confined to human understanding or situated practice but emerges from ongoing interactions between humans and generative models, then questions of authorship, accountability, and authority must be rethought” (Faraj et al., 2025). Ultimately, stakeholders must take full responsibility for their use of AI outputs and cite their use of AI to ensure they are ethically promoting their collaboration with an AI tool.
In the journal article A Time for Monsters: Organizational Knowing after Large Language Models, Faraj and colleagues illustrate how LLMs have reshaped organizational knowing by changing how knowledge is produced, tested, and trusted.
As organizations continue to explore the benefits of AI, developing AI literacy skills is an essential first step toward responsible AI adoption.
Whether you are beginning your learning journey or preparing your organization for broader AI adoption, CAIAI offers a range of educational pathways to help you learn, adapt, and succeed with AI.
For more information about Samer Faraj and his ongoing research as a Canada Research Chair, view his biography on the Government of Canada’s website.
Works Cited:
Faraj, S., Perez-Torrents, J., Mantere, S., & Bhardwaj, A. “A time for monsters: Organizational knowing after large language models”. Strategic Organization, 0(0), 2025. https://doi.org/10.1177/14761270251410675
