The American Journal of Bioethics, 2026
Sedlakova et al. (2025) argue that conversational AI (CAI) cannot satisfy the conditions of epistemic trust in therapeutic contexts. I accept their diagnosis but press a further question: given that patients will de facto treat CAI outputs as reasons, how should patients, clinicians, and designers calibrate their epistemic responses? I argue that even where epistemic trust is inappropriate, calibrated epistemic deference remains rational. On a total evidence view, CAI outputs function as defeasible contributory reasons whose weight is proportional to demonstrated reliability and context-sensitive. Against preemptionism, I show that preemptive CAI deference is distinctively objectionable in therapeutic contexts: it undermines precisely the epistemic capacities — self-knowledge, critical reflection, and reason-integration — that psychotherapy aims to develop. An open question is whether calibrated deference is dynamically stable given users' documented tendency to anthropomorphize CAI over extended interaction, which may cause deference to drift toward the quasi-trust that therapeutic contexts warrant resisting.
In B. Steffen (ed.), Bridging the Gap Between AI and Reality, AISoLA 2024. Lecture Notes in Computer Science, vol. 16032, Springer, 2026
When should we defer to AI outputs over human expert judgment? Drawing on recent work in social epistemology, I motivate the idea that some AI systems qualify as Artificial Epistemic Authorities (AEAs) due to their demonstrated reliability and epistemic superiority. I then introduce AI Preemptionism, the view that AEA outputs should replace rather than supplement a user's independent epistemic reasons. I show that classic objections to preemptionism — such as uncritical deference, epistemic entrenchment, and unhinging epistemic bases — apply in amplified form to AEAs, given their opacity, self-reinforcing authority, and lack of epistemic failure markers. Against this, I develop a more promising alternative: a total evidence view of AI deference. According to this view, AEA outputs should function as contributory reasons rather than outright replacements for a user's independent epistemic considerations. This approach has three key advantages: (i) it mitigates expertise atrophy by keeping human users engaged, (ii) it provides an epistemic case for meaningful human oversight and control, and (iii) it explains the justified mistrust of AI when reliability conditions are unmet. While demanding in practice, this account offers a principled way to determine when AI deference is justified, particularly in high-stakes contexts requiring rigorous reliability.
Beziehungen mit KI
In K. Krüger & R. Geiger (eds.), KI und Demokratie: Aktuelle Fragen, Springer, 2026
Mensch-KI Beziehungen nehmen in unserem Alltag eine immer zentralere Rolle ein: etwa in der Interaktion mit Chatbots, KI-Companions, digitalen Assistenten oder virtuellen Avataren. Dieses Kapitel untersucht, ob und inwiefern sich zwischen Menschen und KI-Systemen Beziehungen entwickeln können und welche ethische Relevanz solchen Beziehungen zukommt. Ziel ist es, Orientierungswissen für die ethische Bewertung zukünftiger Mensch-KI-Interaktionen bereitzustellen. Ausgehend von einer Systematisierung momentaner Mensch-KI-Interaktionen wird der Beziehungsbegriff konzeptuell geschärft und auf seine philosophisch normative Bedeutung hin analysiert. Es wird untersucht, ob KI-Systeme als genuine Beziehungspartner gelten können oder ob die dabei entstehenden Bindungen notwendigerweise asymmetrisch und einseitig bleiben. Ein zentrales Argument ist, dass die normative Debatte nach der Plausibilität von genuinen Mensch-KI Beziehungen im Gesamtkontext von unseren normativen Handlungsgründen betrachtet werden sollte. Anschließend werden drei Ansätze zur ethischen Gestaltung von Mensch-KI Beziehungen vorgestellt und auf ihre Anwendbarkeit hin erläutert.
npj Artificial Intelligence 1 (38), 2025 — with G. Keeling, A. Manzini, A. McCroskery
We argue that accountability mechanisms are needed in human–AI agent relationships to ensure alignment with user and societal interests. We propose a framework according to which AI agents' engagement is conditional on appropriate user behaviour. The framework incorporates design strategies such as distancing, disengaging, and discouraging.
Philosophy & Technology 38 (2025): 1–8
Digital duplicates reduce the scarcity of individuals and thus may impact their instrumental and intrinsic value. I here expand upon this idea by introducing the notion of collective scarcity, which pertains to the limitations faced by social groups in maintaining their size, cohesion and function.
Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT '24): 1078–1092 — with K. Lam, B. Blili-Hamelin, J. Davidovic, S. Brown, A. Hasan
An increasing number of regulations propose the notion of AI audits as an enforcement mechanism for achieving transparency and accountability for AI systems. Despite some converging norms around various forms of AI auditing, auditing for the purpose of compliance and assurance currently have little to no agreed upon practices, procedures, taxonomies, and standards. We propose the criterion audit as an operationalizable compliance and assurance external audit framework. We model elements of this approach after financial auditing practices, and argue that AI audits should similarly provide assurance to their stakeholders about AI organizations' ability to govern their algorithms in ways that mitigate harms and uphold human values. We discuss the necessary conditions for the criterion audit, and provide a procedural blueprint for performing an audit engagement in practice. We illustrate how this framework can be adapted to current regulations by deriving the criteria on which bias audits for hiring algorithms can be performed, as required by the recently effective New York City Local Law 144 of 2021. We conclude by offering critical discussion on the benefits, inherent limitations, and implementation challenges of applying practices of the more mature financial auditing industry to AI auditing where robust guardrails against quality assurance issues are only starting to emerge.
International Review of Information Ethics 34(1), 2024 — with G. Keeling, A. McCroskery, K. Pedersen, D. Weinberger, B. Zevenbergen
"Moral imagination" is the capacity to register that one's perspective on a decision-making situation is limited, and to imagine alternative perspectives that reveal new considerations or approaches. We have developed a Moral Imagination approach that aims to drive a culture of responsible innovation, ethical awareness, deliberation, decision-making, and commitment in organizations developing new technologies. We here present a case study that illustrates one key aspect of our approach — the technomoral scenario — as we have applied it in our work with product and engineering teams. Technomoral scenarios are fictional narratives that raise ethical issues surrounding the interaction between emerging technologies and society. Through facilitated role-playing and discussion, participants are prompted to examine their own intentions, articulate justifications for actions, and consider the impact of decisions on various stakeholders. This process helps developers to re-envision their choices and responsibilities, ultimately contributing to a culture of responsible innovation.
Technological Forecasting & Social Change 204 (2024): 123403 — with F. Poszler
With the rise and public accessibility of AI-enabled decision-support systems, individuals outsource increasingly more of their decisions, even those that carry ethical dimensions. Considering this trend, scholars have highlighted that uncritical deference to these systems would be problematic and consequently called for investigations of the impact of pertinent technology on humans' ethical decision-making. This article conducts a systematic review of existing scholarship and derives an integrated framework that demonstrates how intelligent decision-support systems (IDSSs) shape humans' ethical decision-making. We identify resulting consequences on an individual level (deliberation enhancement, motivation enhancement, autonomy enhancement and action enhancement) and on a societal level (moral deskilling, restricted moral progress and moral responsibility gaps). We carve out two distinct operation types — process-oriented and outcome-oriented navigation — that decision-support systems can deploy and postulate that these determine to what extent the previously stated consequences materialize.
AI & Ethics, 2023 — with A. McCroskery, B. Zevenbergen, G. Keeling, S. Blascovich, K. Pedersen, A. Lentz, B. Aguera y Arcas
We propose a 'Moral Imagination' methodology to facilitate a culture of responsible innovation for engineering and product teams in technology companies. Our approach has been operationalized over the past two years at Google, where we have conducted over 40 workshops with teams from across the organization. We argue that our approach is a crucial complement to existing formal and informal initiatives for fostering a culture of ethical awareness, deliberation, and decision-making in technology design such as company principles, ethics and privacy review procedures, and compliance controls. We characterize some distinctive benefits of our methodology for the technology sector in particular.
The Current State of AI Governance
Whitepaper, Algorithmic Bias Lab, 2023 — with J. Davidovic, A. Hasan, K. Lam, M. Regan, S. Brown
As AI, machine learning algorithms, and algorithmic decision systems (ADS) continue to permeate every aspect of our lives and our society, the question of AI governance becomes exceedingly important. This report examines the current state of internal governance structures and tools across organizations, both in the private and public sectors and in large and small organizations. This report provides one of the first robust and broad insights into the state of AI governance in the United States and Europe.
Digital Society 1(2) (2022): 14 — with A. Hasan, S. Brown, J. Davidovic, M. Regan
In this paper, we distinguish between different sorts of assessments of algorithmic systems, describe our process of assessing such systems for ethical risk, and share some key challenges and lessons for future algorithm assessments and audits. Given the distinctive nature and function of a third-party audit, and the uncertain and shifting regulatory landscape, we suggest that second-party assessments are currently the primary mechanisms for analyzing the social impacts of systems that incorporate artificial intelligence. We then discuss two kinds of assessments: an ethical risk assessment and a narrower, technical algorithmic bias assessment. We explain how the two assessments depend on each other, highlight the importance of situating the algorithm within its particular socio-technical context, and discuss a number of lessons and challenges for algorithm assessments and, potentially, for algorithm audits.
2021 IEEE International Symposium on Technology and Society (ISTAS) Proceedings: 1–5 — with T. M. Lechterman
AI-supported methods for identifying and combating disinformation are progressing in their development and application. However, these methods face a litany of epistemic and ethical challenges. These include (1) robustly defining disinformation, (2) reliably classifying data according to this definition, and (3) navigating ethical risks in the deployment of countermeasures, which involve a mixture of harms and benefits. This paper seeks to expose and offer preliminary analysis of these challenges.
