arXiv subjects

Matthias Uhl

Publications and source records attributed to Matthias Uhl.

15 recordsLinked to original sources

arXiv2026

Operationalizing Fairness in Text-to-Image Models: A Survey of Bias, Fairness Audits and Mitigation Strategies

Text-to-Image (T2I) generation models have been widely adopted across various industries, yet are criticized for frequently exhibiting societal stereotypes. While a growing body of research has emerged to evaluate and mitigate these biases, the field at present contends with conceptual ambiguity, for example terms like "bias" and "fairness" are not always clearly distinguished and often lack clear operational definitions. This paper provides a comprehensive systematic review of T2I fairness literature, organizing existing work into a taxonomy of bias types and fairness notions. We critically assess the gap between "target fairness" (normative ideals in T2I outputs) and "threshold fairness" (normative standards with actionable decision rules). Furthermore, we survey the landscape of mitigation strategies, ranging from prompt engineering to diffusion process manipulation. We conclude by proposing a new framework for operationalizing fairness that moves beyond descriptive metrics towards rigorous, target-based testing, offering an approach for more accountable generative AI development.

cs.CV↗

arXiv2025

The global consensus on the risk management of autonomous driving

Every maneuver of a vehicle redistributes risks between road users. While human drivers do this intuitively, autonomous vehicles allow and require deliberative algorithmic risk management. But how should traffic risks be distributed among road users? In a global experimental study in eight countries with different cultural backgrounds and almost 11,000 participants, we compared risk distribution preferences. It turns out that risk preferences in road traffic are strikingly similar between the cultural zones. The vast majority of participants in all countries deviates from a guiding principle of minimizing accident probabilities in favor of weighing up the probability and severity of accidents. At the national level, the consideration of accident probability and severity hardly differs between countries. The social dilemma of autonomous vehicles detected in deterministic crash scenarios disappears in risk assessments of everyday traffic situations in all countries. In no country do cyclists receive a risk bonus that goes beyond their higher vulnerability. In sum, our results suggest that a global consensus on the risk ethics of autonomous driving is easier to establish than on the ethics of crashing.

cs.CY↗

arXiv2025

ChatGPT's advice drives moral judgments with or without justification

Why do users follow moral advice from chatbots? A chatbot is not an authoritative moral advisor, but it can generate seemingly plausible arguments. Users do not follow reasoned more readily than unreasoned advice, though, we find in an experiment. However, this is also true if we attribute advice to a moral advisor, not a chatbot. Hence, it seems that advice offers users a cheap way to escape from a moral dilemma. This is a concern that chatbots do not raise, but they exacerbate it as they make advice easily accessible. We conclude that it takes ethical in addition to digital literacy to harness users against moral advice from chatbots.

cs.HC↗

arXiv2024

When Two Wrongs Don't Make a Right" -- Examining Confirmation Bias and the Role of Time Pressure During Human-AI Collaboration in Computational Pathology

Artificial intelligence (AI)-based decision support systems hold promise for enhancing diagnostic accuracy and efficiency in computational pathology. However, human-AI collaboration can introduce and amplify cognitive biases, such as confirmation bias caused by false confirmation when erroneous human opinions are reinforced by inaccurate AI output. This bias may worsen when time pressure, ubiquitously present in routine pathology, strains practitioners' cognitive resources. We quantified confirmation bias triggered by AI-induced false confirmation and examined the role of time constraints in a web-based experiment, where trained pathology experts (n=28) estimated tumor cell percentages. Our results suggest that AI integration may fuel confirmation bias, evidenced by a statistically significant positive linear-mixed-effects model coefficient linking AI recommendations mirroring flawed human judgment and alignment with system advice. Conversely, time pressure appeared to weaken this relationship. These findings highlight potential risks of AI use in healthcare and aim to support the safe integration of clinical decision support systems.

cs.HC↗

arXiv2023

The moral authority of ChatGPT

ChatGPT is not only fun to chat with, but it also searches information, answers questions, and gives advice. With consistent moral advice, it might improve the moral judgment and decisions of users, who often hold contradictory moral beliefs. Unfortunately, ChatGPT turns out highly inconsistent as a moral advisor. Nonetheless, it influences users' moral judgment, we find in an experiment, even if they know they are advised by a chatting bot, and they underestimate how much they are influenced. Thus, ChatGPT threatens to corrupt rather than improves users' judgment. These findings raise the question of how to ensure the responsible use of ChatGPT and similar AI. Transparency is often touted but seems ineffective. We propose training to improve digital literacy.

cs.CY↗

arXiv2022

The risk ethics of autonomous vehicles: a continuous trolley problem in regular road traffic

Is the ethics of autonomous vehicles (AVs) restricted to weighing lives in unavoidable accidents? We argue that AVs distribute risks between road users in regular traffic situations, either explicitly or implicitly. This distribution of risks raises ethically relevant questions that cannot be evaded by simple heuristics such as "hitting the brakes." Using an interactive, graphical representation of different traffic situations, we measured participants' preferences on driving maneuvers of AVs in a representative survey in Germany. Our participants' preferences deviated significantly from mere collision avoidance. Interestingly, our participants were willing to take risks themselves for the benefit of other road users suggesting that the social dilemma of AVs may lessen in a context of risk.

cs.CY↗

arXiv2021

Zombies in the Loop? Humans Trust Untrustworthy AI-Advisors for Ethical Decisions

Departing from the claim that AI needs to be trustworthy, we find that ethical advice from an AI-powered algorithm is trusted even when its users know nothing about its training data and when they learn information about it that warrants distrust. We conducted online experiments where the subjects took the role of decision-makers who received advice from an algorithm on how to deal with an ethical dilemma. We manipulated the information about the algorithm and studied its influence. Our findings suggest that AI is overtrusted rather than distrusted. We suggest digital literacy as a potential remedy to ensure the responsible use of AI.

cs.HC↗

arXiv2021

The Behavioral Economics of Intrapersonal Conflict: A Critical Assessment

Preferences often change -- even in short time intervals -- due to either the mere passage of time (present-biased preferences) or changes in environmental conditions (state-dependent preferences). On the basis of the empirical findings in the context of state-dependent preferences, we critically discuss the Aristotelian view of unitary decision makers in economics and urge a more Heraclitean perspective on human decision-making. We illustrate that the conceptualization of preferences as present-biased or state-dependent has very different normative implications under the Aristotelian view, although both concepts are empirically hard to distinguish. This is highly problematic, as it renders almost any paternalistic intervention justifiable.

econ.GN↗

arXiv2021

Hiding Behind Machines: When Blame Is Shifted to Artificial Agents

The transfer of tasks with sometimes far-reaching moral implications to autonomous systems raises a number of ethical questions. In addition to fundamental questions about the moral agency of these systems, behavioral issues arise. This article focuses on the responsibility of agents who decide on our behalf. We investigate the empirically accessible question of whether the production of moral outcomes by an agent is systematically judged differently when the agent is artificial and not human. The results of a laboratory experiment suggest that decision-makers can actually rid themselves of guilt more easily by delegating to machines than by delegating to other people. Our results imply that the availability of artificial agents could provide stronger incentives for decision makers to delegate morally sensitive decisions.

cs.CY↗

arXiv2020

Propagator for a driven Brownian particle in step potentials

Although driven Brownian particles are ubiquitous in stochastic dynamics and often serve as paradigmatic model systems for many aspects of stochastic thermodynamics, fully analytically solvable models are few and far between. In this paper, we introduce an iterative calculation scheme, similar to the method of images in electrostatics, that enables one to obtain the propagator if the potential consists of a finite number of steps. For the special case of a single potential step, this method converges after one iteration, thus providing an expression for the propagator in closed form. In all other cases, the iteration results in an approximation that holds for times smaller than some characteristic timescale that depends on the number of iterations performed. This method can also be applied to a related class of systems like Brownian ratchets, which do not formally contain step potentials in their definition, but impose the same kind of boundary conditions that are caused by potential steps.

cond-mat.stat-mech↗

arXiv2019

Affinity-dependent bound on the spectrum of stochastic matrices

Affinity has proven to be a useful tool for quantifying the non-equilibrium character of time continuous Markov processes since it serves as a measure for the breaking of time reversal symmetry. It has recently been conjectured that the number of coherent oscillations, which is given by the ratio of imaginary and real part of the first non-trivial eigenvalue of the corresponding master matrix, is constrained by the maximum cycle affinity present in the network. In this paper, we conjecture a bound on the whole spectrum of these master matrices that constrains all eigenvalues in a fashion similar to the well known Perron-Frobenius theorem that is valid for any stochastic matrix. As in other studies that are based on affinity-dependent bounds, the limiting process that saturates the bound is given by the asymmetric random walk. For unicyclic networks, we prove that it is not possible to violate the bound by small perturbation of the asymmetric random walk and provide numerical evidence for its validity in randomly generated networks. The results are extended to multicyclic networks, backed up by numerical evidence provided by networks with randomly constructed topology and transition rates.

cond-mat.stat-mech↗

arXiv2018

Force-dependent diffusion coefficient of molecular Brownian ratchets

We study the mean velocity and diffusion constant in three related models of molecular Brownian ratchets. Brownian ratchets can be used to describe translocation of biopolymers like DNA through nanopores in cells in the presence of chaperones on the trans side of the pore. Chaperones can bind to the polymer and prevent it from sliding back through the pore. First, we study a simple model that describes the translocation in terms of an asymmetric random walk. It serves as an introductory example but already captures the main features of a Brownian ratchet. We then provide an analytical expression for the diffusion constant in the classical model of a translocation ratchet that was first proposed by Peskin et al. . This model is based on the assumption that the binding and unbinding of the chaperones is much faster than the diffusion of the DNA strand. To remedy this shortcoming, we propose a modified model that is also applicable if the (un)binding rates are finite. We calculate the force dependent mean velocity and diffusivity for this new model and compare the results to the original one. Our analysis shows that for large pulling forces the predictions of both models can differ strongly even if the (un)binding rates are large in comparison to the diffusion time-scale but still finite. Furthermore, implications of the thermodynamic uncertainty relation on the efficiency of Brownian ratchets are discussed.

cond-mat.stat-mech↗

arXiv2018

Fluctuations of Apparent Entropy Production in Networks with Hidden Slow Degrees of Freedom

The fluctuation theorem for entropy production is a remarkable symmetry of the distribution of produced entropy that holds universally in non-equilibrium steady states with Markovian dynamics. However, in systems with slow degrees of freedom that are hidden from the observer, it is not possible to infer the amount of produced entropy exactly. Previous work suggested that a relation similar to the fluctuation theorem may hold at least approximately for such systems if one considers an apparent entropy production. By extending the notion of apparent entropy production to discrete bipartite systems, we investigate which criteria have to be met for such a modified fluctuation theorem to hold in the large deviation limit. We use asymptotic approximations of the large deviation function to show that the probabilities of extreme events of apparent entropy production always obey a modified fluctuation theorem and, moreover, that it is possible to infer otherwise hidden properties. For the paradigmatic case of two coupled colloidal particles on rings the rate function of the apparent entropy production is calculated to illustrate this asymptotic behavior and to show that the modified fluctuation theorem observed experimentally for short observation times does not persist in the long time limit.

cond-mat.stat-mech↗

arXiv2012

$L^p$-spectral multipliers for some elliptic systems

We show results on $L^p$-spectral multipliers for Maxwell operators with bounded measurable coefficients. We also present similar results for the Stokes operator with Hodge boundary conditions and the Lamé system. Here we rely on resolvent estimates established recently by M.\ Mitrea and S.\ Monniaux.

math.FA↗

arXiv2012

Spectral multiplier theorems of Hörmander type on Hardy and Lebesgue spaces

Let $X$ be a space of homogeneous type and let $L$ be an injective, non-negative, self-adjoint operator on $L^2(X)$ such that the semigroup generated by $-L$ fulfills Davies-Gaffney estimates of arbitrary order. We prove that the operator $F(L)$, initially defined on $H^1_L(X)\cap L^2(X)$, acts as a bounded linear operator on the Hardy space $H^1_L(X)$ associated with $L$ whenever $F$ is a bounded, sufficiently smooth function. Based on this result, together with interpolation, we establish Hörmander type spectral multiplier theorems on Lebesgue spaces for non-negative, self-adjoint operators satisfying generalized Gaussian estimates in which the required differentiability order is relaxed compared to all known spectral multiplier results.

math.FA↗