AI chatbots are not neutral: new paper warns of hidden costs and alternatives

A new academic paper by Sourojit Ghosh, Pranav Narayanan Venkit, Sanjana Gautam, and Avijit Ghosh, accepted at the 2026 ACM Conference on Fairness, Accountability, and Transparency (FAccT), argues that the rapid convergence of AI toward conversational chatbot interfaces is not a neutral design choice but a sociotechnical configuration with far-reaching structural downsides. The authors examine how treating AI primarily as a chatbot reshapes work, learning, and decision-making—often for the worse. They highlight that chatbots frequently fail to meet user needs in complex or high-stakes contexts while projecting unwarranted confidence and authority. This normalization, they claim, contributes to deskilling, homogenization of knowledge, and shifting expectations of expertise, as users increasingly defer to confident but shallow AI responses.

The paper also explores broader societal effects: labor displacement, concentration of economic power among a few large AI companies, and increased environmental costs from sustained investment in large-scale chatbot infrastructure such as massive data centers and model training runs. While acknowledging legitimate benefits of chatbots (e.g., accessibility and convenience), the authors argue the current trajectory reflects specific value choices that prioritize conversational generality over domain specificity, accountability, and long-term sustainability. They conclude by proposing alternative directions for AI development and governance, including pluralistic system design, task-specific tools rather than general-purpose chatbots, and institutional safeguards to mitigate social and economic harm. The paper is available on arXiv (arXiv:2605.07896) and will be presented at FAccT 2026 in Montreal.