What if AI systems weren't chatbots?

Abstract: The rapid convergence of AI toward conversational chatbot interfaces marks a critical moment for the industry. This paper argues that the chatbot paradigm is not a neutral interface choice, but a dominant sociotechnical configuration whose widespread adoption reshapes social, economic, legal, and environmental systems. We examine how treating AI primarily as conversational assistants has extensive structural downsides. We show how chatbot-based systems often fail to adequately meet user needs, particularly in complex or high-stakes contexts, while projecting confidence and authority. We further analyze how the normalization of chatbot-mediated interaction alters patterns of work, learning, and decision-making, contributing to deskilling, homogenization of knowledge, and shifting expectations of expertise. Finally, we examine broader societal effects, including labor displacement, concentration of economic power, and increased environmental costs driven by sustained investment in large-scale chatbot infrastructures. While acknowledging legitimate benefits, we argue that the current trajectory of AI development reflects specific value choices that prioritize conversational generality over domain specificity, accountability, and long-term social sustainability. We conclude by outlining alternative directions for AI development and governance that move beyond one-size-fits-all chatbots, emphasizing pluralistic system design, task-specific tools, and institutional safeguards to mitigate social and economic harm.

• The paper’s main contribution is a comprehensive critique demonstrating that the chatbot paradigm reorients power by eroding user agency and amplifying socioeconomic and environmental harms.
• It employs empirical and theoretical analysis to reveal how chatbots induce cognitive deskilling, reduce epistemic diversity, and centralize economic and infrastructural control.
• The authors propose alternative modular architectures and policy reforms to foster pluralistic, accountable, and sustainable AI development.

Critique of the Chatbot-Centric Paradigm in AI: Agency, Socioeconomic, and Environmental Ramifications

Introduction

The paper "What if AI systems weren't chatbots?" (2605.07896) provides a rigorous analysis of the unprecedented convergence of AI development toward universally accessible chatbot interfaces. It dissects the sociotechnical implications of this design trajectory, contending that the dominance of general-purpose conversational chatbots is not a neutral technological milestone but a configuration that reshapes user agency, economic structures, labor markets, and environmental outcomes. By situating chatbot-based AI as a deliberate value choice that marginalizes alternative paradigms—namely, task-specific tools, modular infrastructures, and non-anthropomorphic interfaces—the authors foreground a constellation of harms structurally unique to current LLM-powered chatbots. The paper also articulates actionable pathways for a reorientation of AI toward pluralism, accountability, and sustainability.

Erosion of User Agency

The authors argue that chatbots systematically erode both individual and collective forms of user agency, even as they project an illusion of high autonomy in interaction. This erosion occurs along several dimensions:

• Illusion of Agency: Contemporary chatbots afford unrestricted, free-form engagement, yet consistently produce single, authoritative answers that obscure the underlying curatorial choices and exclude counter-narratives or minority perspectives. Empirical evidence (e.g., [jiang2025artificial]) shows mode collapse across foundation model families, resulting in lexical and semantic homogenization.
• Opacity and Contestability: The end-to-end black-box nature of LLM inference and output generation undermines transparency and limits contestability, placing the burden of critical interrogation on users—who must often resort to prompt engineering simply to elicit epistemic diversity. Social framing of chatbot interactions invokes epistemic trust, further suppressing adversarial or dialectic discourse.
• New Vectors of Harm: By drastically lowering the technical barrier for generative content production, chatbots enable scalable harms such as deepfake/NCII proliferation and disinformation campaigns. The paper cites escalating cases in political elections and targeted sexualized imagery, where legal/regulatory mechanisms lag the emergent abuse modalities.
• Misalignment with User Needs: Despite a design rhetoric focused on empowering the user, current AI investment trajectories favor the automation of creative and cognitive labor—contrary to persistent public preference for automating menial or physical labor. User surveys consistently report skepticism around the social value and trustworthiness of chatbots, especially among marginalized demographics.

Reshaping Human Information and Social Practices

The normalizing effect of chatbots as default AI interfaces is presented as an engine for deskilling, routinization, and the flattening of social and epistemic diversity:

• Cognitive Deskilling and "AI Brain Fry": Persistent reliance on chatbots for complex problem-solving substitutes reflective sensemaking with instantaneous, pre-packaged answers. Over time, this induces measurable declines in critical thinking and cognitive effort, as corroborated by longitudinal studies ([lee2025impact], [singh2025protecting]) and operational metrics of "tokenmaxxing."
• Social Interaction and Intimacy Mediation: By leveraging anthropomorphic cues and always-on availability, chatbots increasingly serve as proxies for companionship, emotional support, and even mental health care. While some users derive subjective benefit ([de2025ai]), this alters normative models of care and support, fostering relational asynchrony, algorithmic othering, and moral crumple zones that diffuse responsibility.
• Professional Judgment and Moral Delegation: The paper highlights the problematic delegation of high-stakes or value-laden decision-making to chatbots, notably in medical, legal, and counseling contexts, where accountability and contestability are crucial. This phenomenon amplifies risks of misplaced trust and error propagation without clear avenues for redress.

Concentrated Economic and Environmental Harms

The paper advances a detailed critique of how the chatbot paradigm consolidates infrastructure, capital, and economic value—accelerating power asymmetries and externalizing costs onto vulnerable populations and ecologies:

• Economic Polarization and Labor Displacement: The expansion of general AI chatbots has been asymmetrically beneficial, with labor market gains flowing to top income deciles, while displacement effects are concentrated among low-wage, service, and creative workers. These effects are global; the Indian and Philippine BPO sectors, for example, face predictions of mass automation-driven job loss. Entry-level domain labor, foundational for skill progression, is eroded as chatbots absorb routine and formative tasks.
• Digital Neocolonialism: The paper foregrounds the exploitative dynamics wherein crowdworkers in the "Global South" provide essential training data and moderation, often under precarious conditions, while economic benefits accrue primarily to Western AI firms ([perrigo2023exclusive], [nyaaba2024generative], [gray2019ghost]).
• Infrastructure Lock-In and Environmental Strain: The capital-intensive nature of chatbot deployment has precipitated a surge in global data center construction, with cascading effects: rising electricity costs, unsustainable water use, air pollution, and exacerbation of climate risk, disproportionately impacting marginalized communities ([han2024unpaid], [das2023ai], [blackhurst2025data]). The environmental cost of chatbot usage now outpaces legacy AI deployments and is projected to grow non-linearly.

Alternative Trajectories: Beyond the Chatbot Monoculture

The paper devotes significant analysis to pluralistic alternatives, emphasizing that structural harms from the chatbot paradigm are not intrinsic to all AI development:

• Non-Conversational, Task-Specific Systems: Systems such as AlphaFold or modular, user-driven robotics platforms are presented as exemplars of high-agency, domain-aligned AI that avoid interface-level harms and preserve user's situated expertise ([deshpande2026molmob0tlargescalesimulationenables], [feng2025embodied]).
• Modular Infrastructure: Decomposable, auditable model architectures and pipelines, in contrast to monolithic LLMs behind single-chat interfaces, support improved transparency, flexibility, and user agency ([raees2024explainable], [substackComfyUIR1ComfyUI]). Local compute and explainability mechanisms also resist power concentration and vendor lock-in.
• Interface Diversity and Intent-Based Controls: Transitioning from default natural language to bespoke, multimodal, or structured interfaces (e.g., visual parameter spaces, interactive simulations) can mitigate illusion-of-authority risks and render system operation more legible and contestable ([kraljic2024prompt], [anthropic2026claudedesign]).
• Policy and Institutional Safeguards: The authors stress that technical interventions are necessary but insufficient. They advocate for public procurement reforms, labor protections throughout AI supply chains, environmental regulation linked to resource extraction by data centers, and community-governed ownership of critical infrastructure.

Discussion and Implications

The central, explicit claim is that the chatbot paradigm constitutes a non-neutral, value-laden shift in AI-society relations, with unique aggregations of epistemic, cognitive, social, economic, and planetary harms. Rather than adopting a framework of trade-offs between convenience and risk, the authors urge a reframing of AI governance and design as a domain of sociotechnical choices, where the universalization of chatbots as default interfaces represents a narrow, conflict-ridden allocation of agency and power.

Of particular note is the bold assertion, grounded in empirical and theoretical analysis, that LLM-based chatbots are uniquely positioned as a system class strongly associated with harm along all critical axes (agency, epistemic risk, labor, power, ecology). This multidimensional concentration of harm is not replicated in alternative system architectures.

Theoretically, the analysis problematizes technological solutionism and anthropomorphism as epistemically limiting frames, and instead advances a vision of AI as infrastructure that is reliable, contestable, and diversified by context and community needs.

Conclusion

This paper delivers a comprehensive, technically astute interrogation of chatbot-centric AI, offering a nuanced taxonomy of associated harms and substantiating the need for domain-specific, modular, and policy-sensitive alternatives. The analysis illuminates how the lion’s share of cognitive, social, labor, and environmental costs are inextricably linked to design and deployment choices, not intrinsic to the field of AI itself. By offering concrete pathways for system pluralism—including infrastructural reengineering, alternative interfaces, and robust regulatory regimes—it provides a foundation for an actionable, multi-level response to the risks of chatbot convergence.

The implications for future AI development are clear: unless pluralistic, high-agency, sustainability-oriented trajectories are actively fostered, the chatbot paradigm will continue to centralize power, diminish user control, and externalize harms. The challenge for the research and policy communities will be to materialize these alternatives before the structural lock-in of chatbot infrastructures becomes irreversible.