In an era where artificial intelligence (AI) is often heralded as the ultimate facilitator of seamless experiences, a groundbreaking discourse challenges this vision of “frictionless AI.” Recent research published by Zohar, Bloom, and Inzlicht in Communications Psychology opens a critical dialogue about the hidden costs embedded within the pursuit of frictionless AI systems. This thought-provoking perspective marks a decisive turn in how society may come to understand, design, and interact with AI technologies. Rather than idolizing smooth and unobstructed interactions, the authors urge us to embrace friction—conceptualized as the cognitive, social, and ethical “speed bumps” AI could intentionally integrate to foster deeper reflection, trust, and resilience.
At its core, the idea of frictionless AI promotes a utopian ideal: technologies that eliminate every delay, confusion, and misunderstanding to deliver instantly gratifying results. However, while this ideal may promise tremendous efficiencies, Zohar and colleagues argue that it risks oversimplifying complex human-machine relationships and understating the consequences of uncritical automation. The allure of fluid digital interactions often obscures the underlying opacity, biases, and control mechanisms that can erode users’ autonomy. Thus, friction emerges not as an annoyance but as a critical safeguard against the dangers of unchecked AI deployment.
Technically, friction can be implemented through design choices that slow down processes or introduce moments of deliberate scrutiny within AI workflows. For example, requiring users to verify the output of generative models or prompting ethical reflection before decision support systems execute recommendations can serve as built-in friction points. These mechanisms compel users to engage actively with AI rather than passively accepting its suggestions, which is paramount in high-stakes domains like healthcare, finance, or criminal justice. Friction thereby transforms AI from a black-box oracle into a collaborative partner prompting human judgment and deliberation.
One of the central technical challenges highlighted by the authors concerns measuring and calibrating friction. Too little friction encourages complacency and blind trust, while too much friction can frustrate users, impair efficiency, and breed resistance. Balancing these dynamics requires multi-disciplinary expertise spanning computer science, human-computer interaction, cognitive psychology, and ethics. Algorithms can be trained to detect moments when friction is warranted based on contextual cues, such as uncertainty in predictions or ambiguity in user goals. Emerging adaptive user interfaces that respond dynamically to such signals exemplify this approach, representing a shift from rigid automation pipelines toward flexible, user-centric frameworks.
The implications of adopting friction-rich AI stretch far beyond interface design. They touch on fundamental debates about transparency and explainability. The research contends that friction facilitates transparency by compelling AI systems to reveal—not conceal—their reasoning pathways and potential limitations. When AI systems pause or engage users for validation, they expose their vulnerabilities and assumptions. This contrasts starkly with frictionless systems that often deliver conclusions with an illusion of infallibility. Friction thus promotes informed skepticism and empowers users to question and challenge AI outputs constructively.
Ethically, friction assumes a vital role in ensuring accountability and fairness. Automated decisions influence life-altering outcomes, raising the stakes of errors and biases embedded within training data or model architectures. By embedding friction, designers can create procedural checkpoints akin to ethical review boards, where decisions undergo critical appraisal before implementation. This procedural rigor can also help mitigate adversarial manipulation or gaming of AI systems. Friction thereby fortifies social trust and legitimizes AI applications by aligning technological capabilities with human values.
Psychologically, the presence of friction taps into well-documented cognitive processes related to attention, reflection, and decision-making. Humans are prone to form habits and heuristics when interacting with technology; seamless interactions risk bypassing reflective cognition, promoting superficial engagement. Introducing moments that disrupt habitual interactions encourages users to slow down, reconsider, and potentially revise their choices. This heightened awareness contributes to better decision quality and reduces the risk of cascading failures stemming from AI reliance. This understanding draws heavily on classical psychological theories about the interplay between automaticity and controlled processing.
The research also grapples with social dynamics mediated by friction in AI-mediated communication. Messaging platforms, recommendation systems, and online marketplaces thrive on rapid, frictionless exchanges. However, the authors argue that some friction could counteract online polarization, misinformation spread, and impulsive behaviors by embedding pauses and prompts for source verification or diverse perspectives. This reimagining challenges business models that prioritize engagement metrics over societal well-being, pushing for ethical design paradigms that consider long-term social consequences.
Counterintuitively, friction may enhance rather than hinder adoption of AI technologies. When users perceive that AI respects their agency and provides opportunities for meaningful involvement, trust deepens. This contrasts with frustration and alienation when users feel overshadowed by opaque, automated decisions. Thus, friction serves as a communicative cue of respect and collaboration rather than an obstacle. Consequently, friction could catalyze more sustainable AI adoption grounded in mutual understanding rather than blind dependence.
The technical architectures facilitating friction arise from innovations in explainable AI (XAI), human-in-the-loop systems, and interactive machine learning. Techniques such as uncertainty quantification, counterfactual explanations, and interactive querying provide methods to implement friction intentionally. Moreover, adaptive friction can be tailored to individual users’ expertise and context, offering personalized interventions that maximize effectiveness without undue burden. These advances signal a significant shift from purely algorithm-centered development toward holistic human-centered designs.
Navigating the regulatory landscape, friction poses new challenges and opportunities. Regulatory frameworks conventionally emphasize compliance and risk mitigation often through prescriptive controls. Integrating friction as a design principle introduces nuanced demands on documentation, auditing, and compliance processes. Yet, it also opens pathways for regulators to encourage innovation that safeguards societal interests proactively. By promoting friction enabling transparency and deliberation, regulatory agencies can better balance innovation with ethical imperatives.
The vision painted by Zohar and colleagues is far removed from oversimplified techno-utopias. It acknowledges the complex socio-technical ecosystems where AI operates—ecosystems rich with tensions among efficiency, ethics, psychology, and social order. Friction-materialized through thoughtful design—invites a more layered and humanistic approach to AI that celebrates uncertainty and critical engagement rather than masking it. This conceptual shift could redefine AI’s role as a catalyst for enhanced human flourishing rather than mechanized replacement.
Such a paradigm also calls for a broader cultural reorientation toward AI literacy and empowerment. If friction introduces moments of pause and questioning, education must prepare users to interpret, critique, and leverage these moments meaningfully. Beyond technical design, this implicates curricula, public discourse, and interdisciplinary collaboration to cultivate AI citizens capable of navigating complexity critically and creatively. Envisioning friction-rich AI thus entails collective responsibility and shared empowerment.
Finally, as the field advances, empirical research is needed to quantify the impact of friction on diverse populations, tasks, and settings. Controlled experiments, longitudinal studies, and real-world deployments can refine the understanding of friction’s benefits and trade-offs. This research agenda promises to generate robust evidence informing designers, policymakers, and users alike. Commitment to such rigorous investigation will ensure that AI systems evolve in harmony with human values, capabilities, and well-being.
In summation, the provocative call for “Against frictionless AI” offers a timely intervention questioning the prevailing obsession with smooth, seamless digital experiences. By illuminating friction as an essential ingredient for ethical, transparent, and psychologically sound AI, Zohar, Bloom, and Inzlicht chart a visionary pathway forward. Their work challenges technologists, designers, and societies to reconceptualize AI not as a frictionless black box but as an interactive space fostering critical collaboration, reflection, and trust—a friction that ultimately advances human and machine flourishing alike.
Subject of Research: The conceptual critique and design implications of friction in artificial intelligence systems to promote ethical, transparent, and human-centered AI interactions.
Article Title: Against frictionless AI
Article References:
Zohar, E., Bloom, P. & Inzlicht, M. Against frictionless AI. Commun Psychol 4, 39 (2026). https://doi.org/10.1038/s44271-026-00402-1
Image Credits: AI Generated
