How AI co-scientist tools entrench extractive research paradigms while masking systemic inefficiencies in science
Original framing: “Are ‘AI co-scientist’ tools actually useful for scientists?” — STAT News
Scenario modelling suggests that unchecked AI co-scientist adoption could lead to a bifurcated research ecosystem: elite labs with proprietary tools dominate, while underfunded institutions become data colonies for Silicon Valley’s models. A more equitable future might involve federated, open-source AI systems co-designed with Global South researchers, but this requires dismantling the current incentive structures of academic publishing and grant allocation. The rise of 'AI brokers'—intermediaries who monetize access to these tools—could also emerge as a new extractive industry, mirroring the role of academic publishers today.
The AI co-scientist narrative is not a technical debate but a clash of epistemic paradigms, where Silicon Valley’s extractive logic meets the communal, relational knowledge systems of indigenous and Global South traditions.
Historically, automation in science has deepened inequities, from the deskilling of technicians to the consolidation of publishing power in corporate hands, and the current AI wave risks repeating these patterns unless countered by structural reforms. The tools’ promise of efficiency obscures their role as Trojan horses for corporate capture, while marginalized scientists bear the brunt of epistemic extractivism—data mined without consent, credit stolen, and careers devalued. Yet, cross-cultural wisdom and future modelling suggest alternatives: federated, open-source models co-designed with diverse communities, ethics boards that center justice over optimization, and 'slow science' networks that resist the tyranny of speed. The real question is not whether AI can assist science, but who gets to define what science is—and for whose benefit. The trickster’s laughter reminds us that the most 'useful' tools may be those that force us to ask this question anew.