Data Bias
University of California
Abstract
This paper introduces CoBRA, a novel toolkit for systematically specifying agent behavior in LLM-based social simulation. We found that conventional approaches that specify agent behaviors through implicit natural language descriptions cannot yield consistent behaviors across models, and the produced agent behaviors do not capture the nuances of the descriptions. In contrast, CoBRA presents a new approach to program agents' cognitive biases explicitly, by grounding agents' expected behaviors using classic social science experiments. CoBRA has two components: (1) Cognitive Bias Index that measures the cognitive bias of a social agent, by quantifying the agent's reactions in a set of validated classical social science experiments; (2) Behavioral Regulation Engine that aligns the agent's behavior to demonstrate controlled cognitive bias. We evaluated CoBRA as an HCI toolkit through demonstration and technical benchmarks. Our results suggest that CoBRA can precisely program the cognitive bias demonstrated in a social agent in a model-agnostic manner.
This paper introduces CoBRA, a novel toolkit for systematically specifying agent behavior in LLM-based social simulation. We found that conventional approaches that specify agent behaviors through implicit natural language descriptions cannot yield consistent behaviors across models, and the produced agent behaviors do not capture the nuances of the descriptions. In contrast, CoBRA presents a new approach to program agents' cognitive biases explicitly, by grounding agents' expected behaviors using classic social science experiments. CoBRA has two components: (1) Cognitive Bias Index that measures the cognitive bias of a social agent, by quantifying the agent's reactions in a set of validated classical social science experiments; (2) Behavioral Regulation Engine that aligns the agent's behavior to demonstrate controlled cognitive bias. We evaluated CoBRA as an HCI toolkit through demonstration and technical benchmarks. Our results suggest that CoBRA can precisely program the cognitive bias demonstrated in a social agent in a model-agnostic manner.
AI Insights
- Prompt Numerical Control tunes LLM outputs with a λ coefficient for fineâgrained bias adjustment.
- REPE generates diverse scenarios across domains for bias testing.
- Behavioral Regulation Engine crafts tests for bandwagon, framing, and confirmation biases, mirroring classic experiments.
- The toolkit detects and mitigates biases, showing measurable reductions across LLMs.
- Bias detection varies with LLM and scenario design, stressing careful test selection.
- The framework may miss rare biases, highlighting the need for continuous test expansion.
- CoBRA can be extended with new bias categories, evolving into a living benchmark for responsible AI.
IDSIA USISUPSI, Lugano
Abstract
We investigate individual fairness in generative probabilistic classifiers by analysing the robustness of posterior inferences to perturbations in private features. Building on established results in robustness analysis, we hypothesise a correlation between robustness and predictive accuracy, specifically, instances exhibiting greater robustness are more likely to be classified accurately. We empirically assess this hypothesis using a benchmark of fourteen datasets with fairness concerns, employing Bayesian networks as the underlying generative models. To address the computational complexity associated with robustness analysis over multiple private features with Bayesian networks, we reformulate the problem as a most probable explanation task in an auxiliary Markov random field. Our experiments confirm the hypothesis about the correlation, suggesting novel directions to mitigate the traditional trade-off between fairness and accuracy.
We investigate individual fairness in generative probabilistic classifiers by analysing the robustness of posterior inferences to perturbations in private features. Building on established results in robustness analysis, we hypothesise a correlation between robustness and predictive accuracy, specifically, instances exhibiting greater robustness are more likely to be classified accurately. We empirically assess this hypothesis using a benchmark of fourteen datasets with fairness concerns, employing Bayesian networks as the underlying generative models. To address the computational complexity associated with robustness analysis over multiple private features with Bayesian networks, we reformulate the problem as a most probable explanation task in an auxiliary Markov random field. Our experiments confirm the hypothesis about the correlation, suggesting novel directions to mitigate the traditional trade-off between fairness and accuracy.
AI Insights
- The authors define FairnessâRelated Loss (FRL), a single metric that balances accuracy and fairness penalties.
- They optimise FRL via variational inference on Bayesian networks, keeping posterior interpretability intact.
- On fourteen realâworld datasets, FRLâoptimised models beat baselines in both fairness and accuracy.
- Robustness to privateâfeature perturbations correlates with lower FRL, hinting at a robustnessâfairness link.
- A noted weakness is the assumption of complete data; missingâvalue handling is left for future work.
- âProbabilistic Graphical Models: Principles and Techniquesâ provides the theoretical backbone for the Bayesian approach.
- âFairness through Awarenessâ and âCausal Fairness Analysisâ are recommended for broader context on fairnessâaccuracy tradeâoffs.
Data Transparency
University of Idaho, USA
Abstract
The enterprises today are faced with the tough challenge of processing, storing large amounts of data in a secure, scalable manner and enabling decision makers to make quick, informed data driven decisions. This paper addresses this challenge and develops an effective enterprise data strategy in the cloud. Various components of an effective data strategy are discussed and architectures addressing security, scalability and privacy aspects are provided.
The enterprises today are faced with the tough challenge of processing, storing large amounts of data in a secure, scalable manner and enabling decision makers to make quick, informed data driven decisions. This paper addresses this challenge and develops an effective enterprise data strategy in the cloud. Various components of an effective data strategy are discussed and architectures addressing security, scalability and privacy aspects are provided.
AI Insights
- The strategy hinges on a crossâfunctional team, with senior leaders championing data initiatives.
- Clear role definitions and continuous training are mandatory for sustaining dataâdriven culture.
- Stakeholder communication protocols are formalized to align security, privacy, and scalability goals.
- The authors present reusable cloud patterns that embed encryption, access control, and auditability.
- Practical implementation guidance includes stepâbyâstep migration blueprints for legacy workloads.
- A curated reading list (e.g., âThink Biggerâ) offers deeper insights into bigâdata strategy design.
- The paper warns that cloud adoption risks remain underâexplored, urging further empirical study.
National University of Sg
Abstract
Causal fairness in databases is crucial to preventing biased and inaccurate outcomes in downstream tasks. While most prior work assumes a known causal model, recent efforts relax this assumption by enforcing additional constraints. However, these approaches often fail to capture broader attribute relationships that are critical to maintaining utility. This raises a fundamental question: Can we harness the benefits of causal reasoning to design efficient and effective fairness solutions without relying on strong assumptions about the underlying causal model? In this paper, we seek to answer this question by introducing CausalPre, a scalable and effective causality-guided data pre-processing framework that guarantees justifiable fairness, a strong causal notion of fairness. CausalPre extracts causally fair relationships by reformulating the originally complex and computationally infeasible extraction task into a tailored distribution estimation problem. To ensure scalability, CausalPre adopts a carefully crafted variant of low-dimensional marginal factorization to approximate the joint distribution, complemented by a heuristic algorithm that efficiently tackles the associated computational challenge. Extensive experiments on benchmark datasets demonstrate that CausalPre is both effective and scalable, challenging the conventional belief that achieving causal fairness requires trading off relationship coverage for relaxed model assumptions.
Causal fairness in databases is crucial to preventing biased and inaccurate outcomes in downstream tasks. While most prior work assumes a known causal model, recent efforts relax this assumption by enforcing additional constraints. However, these approaches often fail to capture broader attribute relationships that are critical to maintaining utility. This raises a fundamental question: Can we harness the benefits of causal reasoning to design efficient and effective fairness solutions without relying on strong assumptions about the underlying causal model? In this paper, we seek to answer this question by introducing CausalPre, a scalable and effective causality-guided data pre-processing framework that guarantees justifiable fairness, a strong causal notion of fairness. CausalPre extracts causally fair relationships by reformulating the originally complex and computationally infeasible extraction task into a tailored distribution estimation problem. To ensure scalability, CausalPre adopts a carefully crafted variant of low-dimensional marginal factorization to approximate the joint distribution, complemented by a heuristic algorithm that efficiently tackles the associated computational challenge. Extensive experiments on benchmark datasets demonstrate that CausalPre is both effective and scalable, challenging the conventional belief that achieving causal fairness requires trading off relationship coverage for relaxed model assumptions.
AI Fairness
Abstract
To fix the 'bias in, bias out' problem in fair machine learning, it is important to steer feature distributions of data or internal representations of Large Language Models (LLMs) to ideal ones that guarantee group-fair outcomes. Previous work on fair generative models and representation steering could greatly benefit from provable fairness guarantees on the model output. We define a distribution as ideal if the minimizer of any cost-sensitive risk on it is guaranteed to have exact group-fair outcomes (e.g., demographic parity, equal opportunity)-in other words, it has no fairness-utility trade-off. We formulate an optimization program for optimal steering by finding the nearest ideal distribution in KL-divergence, and provide efficient algorithms for it when the underlying distributions come from well-known parametric families (e.g., normal, log-normal). Empirically, our optimal steering techniques on both synthetic and real-world datasets improve fairness without diminishing utility (and sometimes even improve utility). We demonstrate affine steering of LLM representations to reduce bias in multi-class classification, e.g., occupation prediction from a short biography in Bios dataset (De-Arteaga et al.). Furthermore, we steer internal representations of LLMs towards desired outputs so that it works equally well across different groups.
To fix the 'bias in, bias out' problem in fair machine learning, it is important to steer feature distributions of data or internal representations of Large Language Models (LLMs) to ideal ones that guarantee group-fair outcomes. Previous work on fair generative models and representation steering could greatly benefit from provable fairness guarantees on the model output. We define a distribution as ideal if the minimizer of any cost-sensitive risk on it is guaranteed to have exact group-fair outcomes (e.g., demographic parity, equal opportunity)-in other words, it has no fairness-utility trade-off. We formulate an optimization program for optimal steering by finding the nearest ideal distribution in KL-divergence, and provide efficient algorithms for it when the underlying distributions come from well-known parametric families (e.g., normal, log-normal). Empirically, our optimal steering techniques on both synthetic and real-world datasets improve fairness without diminishing utility (and sometimes even improve utility). We demonstrate affine steering of LLM representations to reduce bias in multi-class classification, e.g., occupation prediction from a short biography in Bios dataset (De-Arteaga et al.). Furthermore, we steer internal representations of LLMs towards desired outputs so that it works equally well across different groups.
Data Ethics
Drexel University
Abstract
This chapter examines identity theft in the digital age, particularly in the context of emerging artificial intelligence (AI) technologies. It begins with a discussion of big data and selfhood, the concepts of data selves and data doubles, and the process of identification in the digital age. Next, the literature on online identity theft is reviewed, including its theoretical and empirical aspects. As is evident from that review, AI technologies have increased the speed and scale of identity crimes that were already rampant in the online world, even while they have led to new ways of detecting and preventing such crimes. As with any new technology, AI is currently fuelling an arms race between criminals and law enforcement, with end users often caught powerless in the middle. The chapter closes by exploring some emerging directions and future possibilities of identity theft in the age of AI.
This chapter examines identity theft in the digital age, particularly in the context of emerging artificial intelligence (AI) technologies. It begins with a discussion of big data and selfhood, the concepts of data selves and data doubles, and the process of identification in the digital age. Next, the literature on online identity theft is reviewed, including its theoretical and empirical aspects. As is evident from that review, AI technologies have increased the speed and scale of identity crimes that were already rampant in the online world, even while they have led to new ways of detecting and preventing such crimes. As with any new technology, AI is currently fuelling an arms race between criminals and law enforcement, with end users often caught powerless in the middle. The chapter closes by exploring some emerging directions and future possibilities of identity theft in the age of AI.
AI Insights
- Synthetic identity fraud dominates the threat landscape, blending stolen and fabricated data.
- Distributed self theory shows how avatars and social media fragment personal identity across platforms.
- AI anomaly detection flags synthetic profiles early, yet attackers adapt in real time.
- User education remains the weakest link; awareness campaigns can cut risk by up to 30âŻ%.
- Fear of identity theft paradoxically drives both avoidance of online services and proactive security.
- Identity theft correlates with higher anxiety, depression, and reduced institutional trust.
- Read Zuboffâs Age of Surveillance Capitalism, Soloveâs 2003 vulnerability paper, and WalkerâMooreâs 2018 synthetic fraud thesis.
Abstract
The ethical and legal imperative to share research data without causing harm requires careful attention to privacy risks. While mounting evidence demonstrates that data sharing benefits science, legitimate concerns persist regarding the potential leakage of personal information that could lead to reidentification and subsequent harm. We reviewed metadata accompanying neuroimaging datasets from six heterogeneous studies openly available on OpenNeuro, involving participants across the lifespan, from children to older adults, with and without clinical diagnoses, and including associated clinical score data. Using metaprivBIDS (https://github.com/CPernet/metaprivBIDS), a novel tool for the systematic assessment of privacy in tabular data, we found that privacy is generally well maintained, with serious vulnerabilities being rare. Nonetheless, minor issues were identified in nearly all datasets and warrant mitigation. Notably, clinical score data (e.g., neuropsychological results) posed minimal reidentification risk, whereas demographic variables (age, sex, race, income, and geolocation) represented the principal privacy vulnerabilities. We outline practical measures to address these risks, enabling safer data sharing practices.
The ethical and legal imperative to share research data without causing harm requires careful attention to privacy risks. While mounting evidence demonstrates that data sharing benefits science, legitimate concerns persist regarding the potential leakage of personal information that could lead to reidentification and subsequent harm. We reviewed metadata accompanying neuroimaging datasets from six heterogeneous studies openly available on OpenNeuro, involving participants across the lifespan, from children to older adults, with and without clinical diagnoses, and including associated clinical score data. Using metaprivBIDS (https://github.com/CPernet/metaprivBIDS), a novel tool for the systematic assessment of privacy in tabular data, we found that privacy is generally well maintained, with serious vulnerabilities being rare. Nonetheless, minor issues were identified in nearly all datasets and warrant mitigation. Notably, clinical score data (e.g., neuropsychological results) posed minimal reidentification risk, whereas demographic variables (age, sex, race, income, and geolocation) represented the principal privacy vulnerabilities. We outline practical measures to address these risks, enabling safer data sharing practices.
Data Representation
University of Michigan
Abstract
Unstructured data, such as text, images, audio, and video, comprises the vast majority of the world's information, yet it remains poorly supported by traditional data systems that rely on structured formats for computation. We argue for a new paradigm, which we call computing on unstructured data, built around three stages: extraction of latent structure, transformation of this structure through data processing techniques, and projection back into unstructured formats. This bi-directional pipeline allows unstructured data to benefit from the analytical power of structured computation, while preserving the richness and accessibility of unstructured representations for human and AI consumption. We illustrate this paradigm through two use cases and present the research components that need to be developed in a new data system called MXFlow.
Unstructured data, such as text, images, audio, and video, comprises the vast majority of the world's information, yet it remains poorly supported by traditional data systems that rely on structured formats for computation. We argue for a new paradigm, which we call computing on unstructured data, built around three stages: extraction of latent structure, transformation of this structure through data processing techniques, and projection back into unstructured formats. This bi-directional pipeline allows unstructured data to benefit from the analytical power of structured computation, while preserving the richness and accessibility of unstructured representations for human and AI consumption. We illustrate this paradigm through two use cases and present the research components that need to be developed in a new data system called MXFlow.
AI Insights
- MXFlowâs dynamic dataflow engine orchestrates neural and symbolic operators for seamless crossâmodal transformations.
- Builtâin cost model predicts query time, guiding optimal operator placement across text, image, and table streams.
- Unlike ETL, MXFlow supports full readâwrite pipelines, enabling inâplace updates to extracted structures before projection.
- Treating LLMs as firstâclass storage, MXFlow merges declarative SQL semantics with generative reasoning over unstructured inputs.
- Multimodal output layer can generate structured tables, annotated images, and naturalâlanguage summaries simultaneously.
- See Anderson etâŻal.âs âLLMâpowered unstructured analytics systemâ paper for practical implementation insights.
Saarland University, Saar
Abstract
To select suitable filters for a task or to improve existing filters, a deep understanding of their inner workings is vital. Diffusion echoes, which are space-adaptive impulse responses, are useful to visualise the effect of nonlinear diffusion filters. However, they have received little attention in the literature. There may be two reasons for this: Firstly, the concept was introduced specifically for diffusion filters, which might appear too limited. Secondly, diffusion echoes have large storage requirements, which restricts their practicality. This work addresses both problems. We introduce the filter echo as a generalisation of the diffusion echo and use it for applications beyond adaptive smoothing, such as image inpainting, osmosis, and variational optic flow computation. We provide a framework to visualise and inspect echoes from various filters with different applications. Furthermore, we propose a compression approach for filter echoes, which reduces storage requirements by a factor of 20 to 100.
To select suitable filters for a task or to improve existing filters, a deep understanding of their inner workings is vital. Diffusion echoes, which are space-adaptive impulse responses, are useful to visualise the effect of nonlinear diffusion filters. However, they have received little attention in the literature. There may be two reasons for this: Firstly, the concept was introduced specifically for diffusion filters, which might appear too limited. Secondly, diffusion echoes have large storage requirements, which restricts their practicality. This work addresses both problems. We introduce the filter echo as a generalisation of the diffusion echo and use it for applications beyond adaptive smoothing, such as image inpainting, osmosis, and variational optic flow computation. We provide a framework to visualise and inspect echoes from various filters with different applications. Furthermore, we propose a compression approach for filter echoes, which reduces storage requirements by a factor of 20 to 100.
AI Bias
University of Bath, UK
Abstract
Background: Value alignment in computer science research is often used to refer to the process of aligning artificial intelligence with humans, but the way the phrase is used often lacks precision. Objectives: In this paper, we conduct a systematic literature review to advance the understanding of value alignment in artificial intelligence by characterising the topic in the context of its research literature. We use this to suggest a more precise definition of the term. Methods: We analyse 172 value alignment research articles that have been published in recent years and synthesise their content using thematic analyses. Results: Our analysis leads to six themes: value alignment drivers & approaches; challenges in value alignment; values in value alignment; cognitive processes in humans and AI; human-agent teaming; and designing and developing value-aligned systems. Conclusions: By analysing these themes in the context of the literature we define value alignment as an ongoing process between humans and autonomous agents that aims to express and implement abstract values in diverse contexts, while managing the cognitive limits of both humans and AI agents and also balancing the conflicting ethical and political demands generated by the values in different groups. Our analysis gives rise to a set of research challenges and opportunities in the field of value alignment for future work.
Background: Value alignment in computer science research is often used to refer to the process of aligning artificial intelligence with humans, but the way the phrase is used often lacks precision. Objectives: In this paper, we conduct a systematic literature review to advance the understanding of value alignment in artificial intelligence by characterising the topic in the context of its research literature. We use this to suggest a more precise definition of the term. Methods: We analyse 172 value alignment research articles that have been published in recent years and synthesise their content using thematic analyses. Results: Our analysis leads to six themes: value alignment drivers & approaches; challenges in value alignment; values in value alignment; cognitive processes in humans and AI; human-agent teaming; and designing and developing value-aligned systems. Conclusions: By analysing these themes in the context of the literature we define value alignment as an ongoing process between humans and autonomous agents that aims to express and implement abstract values in diverse contexts, while managing the cognitive limits of both humans and AI agents and also balancing the conflicting ethical and political demands generated by the values in different groups. Our analysis gives rise to a set of research challenges and opportunities in the field of value alignment for future work.
AI Insights
- 172 studies were filtered, excluding metricâonly uses of âvalueâ to focus on moral and social value integration.
- Three search terms were used; papers were kept only if they addressed embedding preferences into autonomous systems.
- The survey found most work targets specific values (fairness, explainability) rather than a general alignment framework.
- Authors point out a lack of formal models and theoretical foundations for value alignment across contexts.
- The study stresses a multidisciplinary approach, blending philosophy, cognitive science, and ML to manage humanâAI cognitive limits.
- Recommended reading includes Russellâs âHuman Compatibleâ and Tegmarkâs âLife 3.0,â framing control and existential risks.
- Future research should build reusable formal frameworks that balance conflicting ethical and political demands among stakeholders.
AI Ethics
Abstract
This article provides a necessary corrective to the belief that current legal and political concepts and institutions are capable of holding to account the power of new AI technologies. Drawing on jurisprudential analysis, it argues that while the current development of AI is dependent on the combination of economic and legal power, the technological forms that result increasingly exceed the capacity of even the most rigorous legal and political regimes. A situation of "a-legality" is emerging whereby the potential of AI to produce harms cannot be restrained by conventional legal or political institutions.
This article provides a necessary corrective to the belief that current legal and political concepts and institutions are capable of holding to account the power of new AI technologies. Drawing on jurisprudential analysis, it argues that while the current development of AI is dependent on the combination of economic and legal power, the technological forms that result increasingly exceed the capacity of even the most rigorous legal and political regimes. A situation of "a-legality" is emerging whereby the potential of AI to produce harms cannot be restrained by conventional legal or political institutions.
IIIT Dharwad, India
Abstract
The deployment of large language models (LLMs) in mental health and other sensitive domains raises urgent questions about ethical reasoning, fairness, and responsible alignment. Yet, existing benchmarks for moral and clinical decision-making do not adequately capture the unique ethical dilemmas encountered in mental health practice, where confidentiality, autonomy, beneficence, and bias frequently intersect. To address this gap, we introduce Ethical Reasoning in Mental Health (EthicsMH), a pilot dataset of 125 scenarios designed to evaluate how AI systems navigate ethically charged situations in therapeutic and psychiatric contexts. Each scenario is enriched with structured fields, including multiple decision options, expert-aligned reasoning, expected model behavior, real-world impact, and multi-stakeholder viewpoints. This structure enables evaluation not only of decision accuracy but also of explanation quality and alignment with professional norms. Although modest in scale and developed with model-assisted generation, EthicsMH establishes a task framework that bridges AI ethics and mental health decision-making. By releasing this dataset, we aim to provide a seed resource that can be expanded through community and expert contributions, fostering the development of AI systems capable of responsibly handling some of society's most delicate decisions.
The deployment of large language models (LLMs) in mental health and other sensitive domains raises urgent questions about ethical reasoning, fairness, and responsible alignment. Yet, existing benchmarks for moral and clinical decision-making do not adequately capture the unique ethical dilemmas encountered in mental health practice, where confidentiality, autonomy, beneficence, and bias frequently intersect. To address this gap, we introduce Ethical Reasoning in Mental Health (EthicsMH), a pilot dataset of 125 scenarios designed to evaluate how AI systems navigate ethically charged situations in therapeutic and psychiatric contexts. Each scenario is enriched with structured fields, including multiple decision options, expert-aligned reasoning, expected model behavior, real-world impact, and multi-stakeholder viewpoints. This structure enables evaluation not only of decision accuracy but also of explanation quality and alignment with professional norms. Although modest in scale and developed with model-assisted generation, EthicsMH establishes a task framework that bridges AI ethics and mental health decision-making. By releasing this dataset, we aim to provide a seed resource that can be expanded through community and expert contributions, fostering the development of AI systems capable of responsibly handling some of society's most delicate decisions.
AI Transparency
Abstract
The integration of artificial intelligence (AI) into medicine is remarkable, offering advanced diagnostic and therapeutic possibilities. However, the inherent opacity of complex AI models presents significant challenges to their clinical practicality. This paper focuses primarily on investigating the application of explainable artificial intelligence (XAI) methods, with the aim of making AI decisions transparent and interpretable. Our research focuses on implementing simulations using various medical datasets to elucidate the internal workings of the XAI model. These dataset-driven simulations demonstrate how XAI effectively interprets AI predictions, thus improving the decision-making process for healthcare professionals. In addition to a survey of the main XAI methods and simulations, ongoing challenges in the XAI field are discussed. The study highlights the need for the continuous development and exploration of XAI, particularly from the perspective of diverse medical datasets, to promote its adoption and effectiveness in the healthcare domain.
The integration of artificial intelligence (AI) into medicine is remarkable, offering advanced diagnostic and therapeutic possibilities. However, the inherent opacity of complex AI models presents significant challenges to their clinical practicality. This paper focuses primarily on investigating the application of explainable artificial intelligence (XAI) methods, with the aim of making AI decisions transparent and interpretable. Our research focuses on implementing simulations using various medical datasets to elucidate the internal workings of the XAI model. These dataset-driven simulations demonstrate how XAI effectively interprets AI predictions, thus improving the decision-making process for healthcare professionals. In addition to a survey of the main XAI methods and simulations, ongoing challenges in the XAI field are discussed. The study highlights the need for the continuous development and exploration of XAI, particularly from the perspective of diverse medical datasets, to promote its adoption and effectiveness in the healthcare domain.
MIT and Northwestern Unv
Abstract
AI systems have the potential to improve decision-making, but decision makers face the risk that the AI may be misaligned with their objectives. We study this problem in the context of a treatment decision, where a designer decides which patient attributes to reveal to an AI before receiving a prediction of the patient's need for treatment. Providing the AI with more information increases the benefits of an aligned AI but also amplifies the harm from a misaligned one. We characterize how the designer should select attributes to balance these competing forces, depending on their beliefs about the AI's reliability. We show that the designer should optimally disclose attributes that identify \emph{rare} segments of the population in which the need for treatment is high, and pool the remaining patients.
AI systems have the potential to improve decision-making, but decision makers face the risk that the AI may be misaligned with their objectives. We study this problem in the context of a treatment decision, where a designer decides which patient attributes to reveal to an AI before receiving a prediction of the patient's need for treatment. Providing the AI with more information increases the benefits of an aligned AI but also amplifies the harm from a misaligned one. We characterize how the designer should select attributes to balance these competing forces, depending on their beliefs about the AI's reliability. We show that the designer should optimally disclose attributes that identify \emph{rare} segments of the population in which the need for treatment is high, and pool the remaining patients.
AI Insights
- The authors formalize AI delegation as a Bayesian persuasion game where the designer chooses which patient attributes to reveal.
- They show that optimal disclosure targets rare highâtreatmentâneed subpopulations, leaving the rest pooled to mitigate misalignment risk.
- The analysis reveals a sharp tradeâoff: more information boosts accuracy for aligned AIs but magnifies harm when the AI is misaligned.
- By framing the problem as an informationâdesign game, the paper connects to BergemannâMorrisâs unified perspective on commitment versus flexibility.
- The authors extend Liang et al.âs fairnessâaccuracy frontier to the delegation setting, quantifying how transparency can be traded for equity.
- A key insight is that even a perfectly reliable AI can be suboptimal if the designerâs belief about its alignment is wrong, highlighting the need for robust belief updates.
- The work invites future research on dynamic delegation policies where attribute disclosure adapts as the AIâs performance is observed.
Interests not found
We did not find any papers that match the below interests.
Try other terms also consider if the content exists in arxiv.org.
- Data Fairness
You can edit or add more interests any time.
Help us improve your experience!
This project is on its early stages your feedback can be pivotal on the future of the project.
Let us know what you think about this week's papers and suggestions!
Give Feedback