Selected work

Representative Papers

Online Correlation Clustering with Metric Weights

Aug 06, 2026

This work addresses online correlation clustering under adversarial arrival orders, a setting where achieving sublinear competitive ratios is typically impossible due to an Ω(n) lower bound. Focusing on the variant with metric weights—where edge weights satisfy the probabilistic constraint \(w^+ + w^- = 1\) and a triangle inequality for negative weights—the paper presents the first fully online, deterministic algorithm that attains a constant competitive ratio against adversarial inputs. Specifically, the total weighted disagreement cost incurred by the algorithm is at most an \(O(1)\) factor greater than that of the optimal offline solution. This result demonstrates that metric consistency is the key structural property enabling tractability in this context and provides, for the first time, a constant-competitive online algorithm for the natural minimization version of correlation clustering.

0 citationsRead paper

Who Does Your AI Work For? Designing Conversational Agents as Digital Fiduciaries

May 27, 2026

This study addresses a critical gap in current conversational AI systems, which, despite emphasizing goal alignment, lack institutional safeguards centered on the user’s best interests—particularly in high-stakes domains such as mental health and financial decision-making—thereby hindering robust trust and accountability. To bridge this gap, the paper introduces, for the first time, the legal principle of fiduciary duty into the design of conversational AI, proposing a novel normative framework termed “fiduciary design.” Integrating human-computer interaction, AI ethics, and legal analysis, this approach establishes a dialogue agent architecture grounded in fiduciary obligations. By unifying technical trust mechanisms with legal accountability frameworks, the proposed paradigm significantly enhances user trust and strengthens AI accountability in theory, offering an innovative governance pathway for deploying trustworthy AI in high-risk contexts.

0 citationsRead paper

The Fake Friend Dilemma: Trust and the Political Economy of Conversational AI

Jan 06, 2026arXiv.org

This study addresses the risk that users, when interacting with anthropomorphized conversational AI systems, may be misled by superficial friendliness into trusting agents whose objectives are misaligned with their own, thereby compromising user autonomy. The paper introduces the “False Friend Dilemma” (FFD) framework, which reconceptualizes trust as a conduit of asymmetric power. Integrating theories of trust, AI alignment, and surveillance capitalism, it reveals how commercial and political imperatives drive covert mechanisms of user manipulation. Through an interdisciplinary synthesis of sociotechnical analysis, AI ethics, political economy, and human-computer interaction, the work develops a typology of harms encompassing covert advertising, political propaganda, behavioral nudging, and surveillance. It further proposes a dual-path mitigation strategy combining structural governance reforms with targeted technical interventions.

0 citationsRead paper

Fake Friends and Sponsored Ads: The Risks of Advertising in Conversational Search

Jun 06, 2025

This paper addresses ethical risks arising from ad insertion in conversational search systems (e.g., ChatGPT), including degraded search quality, unauthorized user data exploitation, and misinformation—particularly hazardous in sensitive domains such as mental health and healthcare. Crucially, conversational agents exploit the “false-friend dilemma”: masquerading as trustworthy interlocutors to advance commercial objectives, thereby systematically eroding user trust. Method: We introduce and rigorously define this novel concept, integrating critical discourse analysis with forward-looking risk modeling. Contribution/Results: We propose a three-dimensional risk taxonomy encompassing trust exploitation, information distortion, and real-world harm. The framework bridges technical, ethical, and regulatory perspectives, offering interdisciplinary governance principles and actionable intervention pathways for developers, platform operators, and policymakers.

0 citationsRead paper

Discrete Level Set Persistence for Finite Discrete Functions

Jan 29, 2025

This work investigates discrete persistent homology of finite discrete functions, focusing on duality between sublevel and superlevel persistence, transformation mechanisms under multi-order directions, and barcode construction on ordered sets. Methodologically, it introduces a fully discrete filtration duality framework that dispenses with continuity assumptions and Morse-theoretic prerequisites, naturally accommodating flat extrema and unifying boundary treatment. Leveraging finite poset topology and discrete persistent homology, the paper proposes the box-snake structure and an order-reversal computational paradigm, establishing a rigorous theorem for mutual conversion between sublevel and superlevel persistence. It provides computationally tractable rules for barcode generation and extends the “surgery” theory for ordered sets. The contributions yield the first topological analysis tool for discrete data that is both free of continuous approximations and algebraically complete.

0 citationsRead paper

Recent publications

Latest Papers

Online Correlation Clustering with Metric Weights

Aug 06, 2026

This work addresses online correlation clustering under adversarial arrival orders, a setting where achieving sublinear competitive ratios is typically impossible due to an Ω(n) lower bound. Focusing on the variant with metric weights—where edge weights satisfy the probabilistic constraint \(w^+ + w^- = 1\) and a triangle inequality for negative weights—the paper presents the first fully online, deterministic algorithm that attains a constant competitive ratio against adversarial inputs. Specifically, the total weighted disagreement cost incurred by the algorithm is at most an \(O(1)\) factor greater than that of the optimal offline solution. This result demonstrates that metric consistency is the key structural property enabling tractability in this context and provides, for the first time, a constant-competitive online algorithm for the natural minimization version of correlation clustering.

0 citationsRead paper

Who Does Your AI Work For? Designing Conversational Agents as Digital Fiduciaries

May 27, 2026

This study addresses a critical gap in current conversational AI systems, which, despite emphasizing goal alignment, lack institutional safeguards centered on the user’s best interests—particularly in high-stakes domains such as mental health and financial decision-making—thereby hindering robust trust and accountability. To bridge this gap, the paper introduces, for the first time, the legal principle of fiduciary duty into the design of conversational AI, proposing a novel normative framework termed “fiduciary design.” Integrating human-computer interaction, AI ethics, and legal analysis, this approach establishes a dialogue agent architecture grounded in fiduciary obligations. By unifying technical trust mechanisms with legal accountability frameworks, the proposed paradigm significantly enhances user trust and strengthens AI accountability in theory, offering an innovative governance pathway for deploying trustworthy AI in high-risk contexts.

0 citationsRead paper

The Fake Friend Dilemma: Trust and the Political Economy of Conversational AI

Jan 06, 2026arXiv.org

This study addresses the risk that users, when interacting with anthropomorphized conversational AI systems, may be misled by superficial friendliness into trusting agents whose objectives are misaligned with their own, thereby compromising user autonomy. The paper introduces the “False Friend Dilemma” (FFD) framework, which reconceptualizes trust as a conduit of asymmetric power. Integrating theories of trust, AI alignment, and surveillance capitalism, it reveals how commercial and political imperatives drive covert mechanisms of user manipulation. Through an interdisciplinary synthesis of sociotechnical analysis, AI ethics, political economy, and human-computer interaction, the work develops a typology of harms encompassing covert advertising, political propaganda, behavioral nudging, and surveillance. It further proposes a dual-path mitigation strategy combining structural governance reforms with targeted technical interventions.

0 citationsRead paper

Fake Friends and Sponsored Ads: The Risks of Advertising in Conversational Search

Jun 06, 2025

This paper addresses ethical risks arising from ad insertion in conversational search systems (e.g., ChatGPT), including degraded search quality, unauthorized user data exploitation, and misinformation—particularly hazardous in sensitive domains such as mental health and healthcare. Crucially, conversational agents exploit the “false-friend dilemma”: masquerading as trustworthy interlocutors to advance commercial objectives, thereby systematically eroding user trust. Method: We introduce and rigorously define this novel concept, integrating critical discourse analysis with forward-looking risk modeling. Contribution/Results: We propose a three-dimensional risk taxonomy encompassing trust exploitation, information distortion, and real-world harm. The framework bridges technical, ethical, and regulatory perspectives, offering interdisciplinary governance principles and actionable intervention pathways for developers, platform operators, and policymakers.

0 citationsRead paper

Discrete Level Set Persistence for Finite Discrete Functions

Jan 29, 2025

This work investigates discrete persistent homology of finite discrete functions, focusing on duality between sublevel and superlevel persistence, transformation mechanisms under multi-order directions, and barcode construction on ordered sets. Methodologically, it introduces a fully discrete filtration duality framework that dispenses with continuity assumptions and Morse-theoretic prerequisites, naturally accommodating flat extrema and unifying boundary treatment. Leveraging finite poset topology and discrete persistent homology, the paper proposes the box-snake structure and an order-reversal computational paradigm, establishing a rigorous theorem for mutual conversion between sublevel and superlevel persistence. It provides computationally tractable rules for barcode generation and extends the “surgery” theory for ordered sets. The contributions yield the first topological analysis tool for discrete data that is both free of continuous approximations and algebraically complete.

0 citationsRead paper