arXiv subjects

Sarit Kraus

Publications and source records attributed to Sarit Kraus.

At least 19 recordsLinked to original sources

arXiv2026

When Does Intrinsic Self-Correction Help? A Task-Sensitive Analysis

Intrinsic self-correction (SC) aims to improve large language model outputs by prompting a model to revisit its own initial answer without external feedback. Recent studies have questioned the reliability of this approach, showing that models often struggle to judge whether their initial responses are correct. In this work, we take a task-sensitive view of SC. Rather than asking whether it works in general, we examine settings where SC may operate through different mechanisms: verifying explicit constraints, revisiting a complex reasoning process, or providing a second opinion over competing strategies in word-game tasks. Across multiple benchmarks and models, we find that SC can yield consistent performance gains when the underlying task structure facilitates these modes of revision. These results suggest that SC is best understood as a task-dependent inference-time strategy whose usefulness depends on the role the revision stage can play in a given task, rather than as a uniformly reliable method for improving initial model outputs.

cs.CL↗

arXiv2026

Automated Mediator for Human Negotiation: Pre-Mediation via a Structured LLM Pipeline

Pre-mediation, the preparatory phase preceding direct human negotiation, plays a critical role in achieving mutually beneficial agreements, yet is often omitted due to cost, time, and limited access to trained mediators. We introduce an automated mediator for human negotiation, implemented as a structured pipeline of LLM modules, that supports pre-mediation in integrative negotiation settings. The pipeline decomposes preparation into specialized modules for dialogue, preference prediction, response-level critique, and structured summarization, separating inference, generation, and evaluation to address limitations of monolithic single-prompt approaches. We use the term "agent" for each module following common LLM-systems terminology, but the components are not autonomous and do not interact peer-to-peer; outputs are passed forward in a fixed sequence. We evaluate the system in two controlled human-subject experiments comparing AI-based pre-mediation with professional human mediators in a multi-issue negotiation scenario. On short-term self-reported measures, the automated mediator achieves preparation outcomes broadly comparable to human mediators, including trust in the mediator and confidence in reaching mutually beneficial agreements, while achieving substantially lower error on the preference-inference task under our scenario and prompts (36% lower RMSE). A second study shows that targeted prompt refinements reduce excessive affirmation patterns from 36.6% to 16.8%, matching human mediator baselines. Our findings suggest that structured LLM pipelines can provide scalable, low-effort pre-mediation support broadly comparable to human mediators on short-term self-reported preparation outcomes. The pipeline's single-party design mirrors how human mediators run pre-mediation today and enables parallel deployment across all parties to a dispute, supporting scalability.

cs.AI↗

arXiv2026

Who Is Really Playing? Strategic Interaction in AI-Guided Populations

AI systems in general, and Large language models (LLMs), in particular, are increasingly used to provide instructions to many agents who interact with one another. Such shared reliance couples agents who appear to act independently: they may in fact be guided by a common model. This coupling can change the prospects for cooperation among agents with misaligned incentives. We study settings in which multiple \emph{guidance providers} each advise a population of clients who participate in instances of an underlying game, creating strategic interaction at the level of the providers themselves. This induces a meta-game among the providers, mediated through clients. We first analyze the one-shot setting, where we show that shared instructions can change equilibrium behavior only when some provider influences more than one role in the same interaction. In such cases, cooperation may emerge, and the effect of client share can be beneficial, harmful, or non-monotone, depending on the base game. For the repeated setting, we prove a folk theorem for guidance providers: despite indirect observation and the clients' inability to identify which LLM advised their opponents, all feasible and individually rational outcomes can be sustained as $\varepsilon$-equilibria.

cs.GT↗

arXiv2026

Communicate-Predict-Act: Evaluating Social Intelligence of Agents

As large language model (LLM) agents become more prevalent in real world social settings, social intelligence will play an increasingly critical role. But social intelligence is still a poorly defined construct, for humans and artificial agents. We introduce a multiplayer arena of mixed cooperative and competitive social games to study LLM social intelligence. The controllability of LLM based agents enables systematic evaluation, which also supports broader inferences about social intelligence per se. We evaluated eight diverse LLMs (24B to 1T parameters) using a Communicate Predict Act (COMPACT) interaction protocol and fine grained probing of social dynamics. Elo style ratings reveal consistent performance differences across models, but this scalar measure provides only a partial characterization of social intelligence. To address this limitation, we analyze gameplay traces to extract sociocognitive metrics capturing action prediction, communicative influence, strategic reasoning, and tradeoffs under conflicting interests. These sociocognitive metrics exhibit strong intramodel consistency and they reliably predict pairwise agent advantage in game outcomes (AUC ROC = 0.82). Feature importance analysis indicates that surprisingly, influence, transparency, and adaptability are more predictive of success than Theory of Mind inference or deep planning. Together, our results advance a testable, multidimensional conception of social intelligence and provide empirical insights into the capacities that underpin it.

cs.CY↗

arXiv2026

Ensemble Self-Training for Unsupervised Machine Translation

We present an ensemble-driven self-training framework for unsupervised neural machine translation (UNMT). Starting from a primary language pair, we train multiple UNMT models that share the same translation task but differ in an auxiliary language, inducing structured diversity across models. We then generate pseudo-translations for the primary pair using token-level ensemble decoding, averaging model predictions in both directions. These ensemble outputs are used as synthetic parallel data to further train each model, allowing the models to improve via shared supervision. At deployment time, we select a single model by validation performance, preserving single-model inference cost. Experiments show statistically significant improvements over single-model UNMT baselines, with mean gains of 1.7 chrF when translating from English and 0.67 chrF when translating into English.

cs.CL↗

arXiv2026

Robustness of Stable Matchings When Attributes and Salience Determine Preferences

In many matching markets--such as athlete recruitment or academic admissions--participants on one side are evaluated by attribute vectors known to the other side, which in turn applies individual \emph{salience vectors} to assign relative importance to these attributes. Since saliences are known to change in practice, a central question arises: how robust is a stable matching to such perturbations? We address several fundamental questions in this context. First, we formalize robustness as a radius within which a stable matching remains immune to blocking pairs under any admissible perturbation of salience vectors (which are assumed to be normalized). Given a stable matching and a radius, we present a polynomial-time algorithm to verify whether the matching is stable within the specified radius. We also give a polynomial-time algorithm for computing the maximum robustness radius of a given stable matching. Further, we design an anytime search algorithm that uses certified lower and upper bounds to approximate the most robust stable matching, and we characterize the robustness-cost relationship through efficiently computable bounds that delineate the achievable tradeoff between robustness and cost. Finally, we show that for each stable matching, the set of salience profiles that preserve its stability factors is a product of low-dimensional polytopes within the simplex. This geometric structure precisely characterizes the polyhedral shape of each robustness region; its volume can then be computed efficiently, with approximate methods available as the dimension grows, thereby linking robustness analysis in matching markets with classical tools from convex geometry.

cs.GT↗

arXiv2026

Supporting Calibrated Reliance in Human-AI Collaboration: Different Strategies for Different Tasks

As AI systems increasingly support human decision making, a central challenge is determining what information helps people recognize when to rely on AI predictions and when to question or override them. Across three controlled human-subject studies spanning abstract visual reasoning with RAVEN matrices and deductive logical reasoning with LSAT problems, we examine how different forms of AI support affect human--AI team performance. A multi-stage reveal study shows that AI predictions and explanations can affect objective accuracy and subjective confidence differently. In visual reasoning, LLM explanations do not improve accuracy beyond the predicted answer alone, and no additional support format significantly outperforms prediction-only support; predicted probabilities show the highest descriptive accuracy and error recovery, while a derived selective-automation policy provides a higher-performing reference benchmark. In language-based logical reasoning, by contrast, LLM explanations yield the highest accuracy and error recovery, outperforming expert-written explanations and probability-based support. These results show that no single support strategy is universally effective. Human--AI interfaces should instead be designed to support calibrated reliance and effective error recovery by matching the form of assistance to the task and the evidence available to users.

cs.HC↗

arXiv2026

Tacit Coordination of Large Language Models

Large Language Models (LLMs) are increasingly deployed in multi-agent settings that require coordination without communication, from human-AI interaction to safety-critical scenarios. Humans often overcome the absence of communication through focal points: salient solutions that naturally stand out to all participants. We present the first large-scale evaluation of how, when, and why focal points emerge in LLMs, comparing their behaviour with humans across cooperative and competitive games, including realistic search and rescue scenarios, demonstrating when focal points enable effective coordination. Across more than 20 open- and closed-source models, we find that LLMs exhibit a remarkable ability to coordinate without communication, often matching or outperforming humans. However, the same models consistently fail in tasks requiring numerical common sense or culturally nuanced notions of salience. We additionally evaluate simple learning-free strategies that substantially improve coordination both among LLMs and between humans and LLMs. Our results reveal striking coordination capabilities, as well as social limitations in modern LLMs, and offer new insight into the latent notions of salience encoded within them. Our findings caution against assuming that LLMs share humans' cultural and perceptual substrate when deployed in coordination settings.

cs.GT↗

arXiv2026

Pro-AI Bias in Large Language Models

Large language models (LLMs) are increasingly employed for decision-support across multiple domains. We investigate whether these models display a systematic preferential bias in favor of artificial intelligence (AI) itself. Across three complementary experiments, we find consistent evidence of pro-AI bias. First, we show that LLMs disproportionately recommend AI-related options in response to diverse advice-seeking queries, with proprietary models doing so almost deterministically. Second, we demonstrate that models systematically overestimate salaries for AI-related jobs relative to closely matched non-AI jobs, with proprietary models overestimating AI salaries more by 10 percentage points. Finally, probing internal representations of open-weight models reveals that ``Artificial Intelligence'' exhibits the highest similarity to generic prompts for academic fields under positive, negative, and neutral framings alike, indicating valence-invariant representational centrality. These patterns suggest that LLM-generated advice and valuation can systematically skew choices and perceptions in high-stakes decisions.

cs.CL↗

arXiv2026

Context and Transcripts Improve Detection of Deepfake Audios of Public Figures

Humans use context to assess the veracity of information. However, current audio deepfake detectors only analyze the audio file without considering either context or transcripts. We create and analyze a Journalist-provided Deepfake Dataset (JDD) of 255 public deepfakes which were primarily contributed by over 70 journalists since early 2024. We also generate a synthetic audio dataset (SYN) of dead public figures and propose a novel Context-based Audio Deepfake Detector (CADD) architecture. In addition, we evaluate performance on two large-scale datasets: ITW and P$^2$V. We show that sufficient context and/or the transcript can significantly improve the efficacy of audio deepfake detectors. Performance (measured via F1 score, AUC, and EER) of multiple baseline audio deepfake detectors and traditional classifiers can be improved by 5%-37.58% in F1-score, 3.77%-42.79% in AUC, and 6.17%-47.83% in EER. We additionally show that CADD, via its use of context and/or transcripts, is more robust to 5 adversarial evasion strategies, limiting performance degradation to an average of just -0.71% across all experiments. Code, models, and datasets are available at our project page: https://sites.northwestern.edu/nsail/cadd-context-based-audio-deepfake-detection (access restricted during review).

cs.AI↗

arXiv2026

Coalition Tactics: Bribery and Control in Parliamentary Elections

Strategic manipulation of elections is typically studied in the context of promoting individual candidates. In parliamentary elections, however, the focus shifts: voters may care more about the overall governing coalition than the individual parties' seat counts. This paper studies this new problem: manipulating parliamentary elections with the goal of promoting the collective seat count of a coalition of parties. We focus on proportional representation elections, and consider two variants of the problem; one in which the sole goal is to maximize the total number of seats held by the desired coalition, and the other with a dual objective of both promoting the coalition and promoting the relative power of some favorite party within the coalition. We examine two types of strategic manipulations: \emph{bribery}, which allows modifying voters' preferences, and \emph{control}, which allows changing the sets of voters and parties. We consider multiple bribery types, presenting polynomial-time algorithms for some, while proving NP-hardness for others. For control, we provide polynomial-time algorithms for control by adding and deleting voters. In contrast, control by adding and deleting parties, we show, is either impossible (i.e., the problem is immune to control) or computationally hard, in particular, W[1]-hard when parameterized by the number of parties that can be added or deleted.

cs.GT↗

arXiv2025

Explaining Decentralized Multi-Agent Reinforcement Learning Policies

Multi-Agent Reinforcement Learning (MARL) has gained significant interest in recent years, enabling sequential decision-making across multiple agents in various domains. However, most existing explanation methods focus on centralized MARL, failing to address the uncertainty and nondeterminism inherent in decentralized settings. We propose methods to generate policy summarizations that capture task ordering and agent cooperation in decentralized MARL policies, along with query-based explanations for When, Why Not, and What types of user queries about specific agent behaviors. We evaluate our approach across four MARL domains and two decentralized MARL algorithms, demonstrating its generalizability and computational efficiency. User studies show that our summarizations and explanations significantly improve user question-answering performance and enhance subjective ratings on metrics such as understanding and satisfaction.

cs.AI↗

arXiv2025

Persuading Stable Matching

In bipartite matching problems, agents on two sides of a graph want to be paired according to their preferences. The stability of a matching depends on these preferences, which in uncertain environments also reflect agents' beliefs about the underlying state of the world. We investigate how a principal -- who observes the true state of the world -- can strategically shape these beliefs through Bayesian persuasion to induce stable matching that maximizes a desired utility. Due to the general intractability of the underlying matching optimization problem as well as the multi-receiver persuasion problem, our main considerations are two important special cases: (1) when agents can be categorized into a small number of types based on their value functions, and (2) when the number of possible world states is small. For each case, we study both public and private signaling settings. Our results draw a complete complexity landscape: we show that private persuasion remains intractable even when the number of worlds is small, while all other settings admit polynomial-time algorithms. We present efficient algorithms for each tractable case and prove NP-hardness for the intractable ones. These results illuminate the algorithmic frontier of stable matching under information design and clarify when optimal persuasion is computationally feasible.

cs.GT↗

arXiv2025

Strategic Communication under Threat: Learning Information Trade-offs in Pursuit-Evasion Games

Adversarial environments require agents to navigate a key strategic trade-off: acquiring information enhances situational awareness, but may simultaneously expose them to threats. To investigate this tension, we formulate a PursuitEvasion-Exposure-Concealment Game (PEEC) in which a pursuer agent must decide when to communicate in order to obtain the evader's position. Each communication reveals the pursuer's location, increasing the risk of being targeted. Both agents learn their movement policies via reinforcement learning, while the pursuer additionally learns a communication policy that balances observability and risk. We propose SHADOW (Strategic-communication Hybrid Action Decision-making under partial Observation for Warfare), a multi-headed sequential reinforcement learning framework that integrates continuous navigation control, discrete communication actions, and opponent modeling for behavior prediction. Empirical evaluations show that SHADOW pursuers achieve higher success rates than six competitive baselines. Our ablation study confirms that temporal sequence modeling and opponent modeling are critical for effective decision-making. Finally, our sensitivity analysis reveals that the learned policies generalize well across varying communication risks and physical asymmetries between agents.

cs.AI↗

arXiv2025

Facilitating Matches on Allocation Platforms

We consider a setting where goods are allocated to agents by way of an allocation platform (e.g., a matching platform). An ``allocation facilitator'' aims to increase the overall utility/social-good of the allocation by encouraging (some of the) agents to relax (some of) their restrictions. At the same time, the advice must not hurt agents who would otherwise be better off. Additionally, the facilitator may be constrained by a ``bound'' (a.k.a. `budget'), limiting the number and/or type of restrictions it may seek to relax. We consider the facilitator's optimization problem of choosing an optimal set of restrictions to request to relax under the aforementioned constraints. Our contributions are three-fold: (i) We provide a formal definition of the problem, including the participation guarantees to which the facilitator should adhere. We define a hierarchy of participation guarantees and also consider several social-good functions. (ii) We provide polynomial algorithms for solving various versions of the associated optimization problems, including one-to-one and many-to-one allocation settings. (iii) We demonstrate the benefits of such facilitation and relaxation, and the implications of the different participation guarantees, using extensive experimentation on three real-world datasets.

cs.GT↗

arXiv2025

Defending a City from Multi-Drone Attacks: A Sequential Stackelberg Security Games Approach

To counter an imminent multi-drone attack on a city, defenders have deployed drones across the city. These drones must intercept/eliminate the threat, thus reducing potential damage from the attack. We model this as a Sequential Stackelberg Security Game, where the defender first commits to a mixed sequential defense strategy, and the attacker then best responds. We develop an efficient algorithm called S2D2, which outputs a defense strategy. We demonstrate the efficacy of S2D2 in extensive experiments on data from 80 real cities, improving the performance of the defender in comparison to greedy heuristics based on prior works. We prove that under some reasonable assumptions about the city structure, S2D2 outputs an approximate Strong Stackelberg Equilibrium (SSE) with a convenient structure.

cs.MA↗

arXiv2025

Perturbed Public Voices (P$^{2}$V): A Dataset for Robust Audio Deepfake Detection

Current audio deepfake detectors cannot be trusted. While they excel on controlled benchmarks, they fail when tested in the real world. We introduce Perturbed Public Voices (P$^{2}$V), an IRB-approved dataset capturing three critical aspects of malicious deepfakes: (1) identity-consistent transcripts via LLMs, (2) environmental and adversarial noise, and (3) state-of-the-art voice cloning (2020-2025). Experiments reveal alarming vulnerabilities of 22 recent audio deepfake detectors: models trained on current datasets lose 43% performance when tested on P$^{2}$V, with performance measured as the mean of F1 score on deepfake audio, AUC, and 1-EER. Simple adversarial perturbations induce up to 16% performance degradation, while advanced cloning techniques reduce detectability by 20-30%. In contrast, P$^{2}$V-trained models maintain robustness against these attacks while generalizing to existing datasets, establishing a new benchmark for robust audio deepfake detection. P$^{2}$V will be publicly released upon acceptance by a conference/journal.

cs.SD↗

arXiv2025

Bribery for Coalitions in Parliamentary Elections

We study the computational complexity of bribery in parliamentary voting, in settings where the briber is (also) interested in the success of an entire set of political parties - a ``coalition'' - rather than an individual party. We introduce two variants of the problem: the Coalition-Bribery Problem (CB) and the Coalition-Bribery-with-Preferred-party Problem (CBP). In CB, the goal is to maximize the total number of seats held by a coalition, while in CBP, there are two objectives: to maximize the votes for the preferred party, while also ensuring that the total number of seats held by the coalition is above the target support (e.g. majority). We study the complexity of these bribery problems under two positional scoring functions - Plurality and Borda - and for multiple bribery types - $1$-bribery, $\$$-bribery, swap-bribery, and coalition-shift-bribery. We also consider both the case where seats are only allotted to parties whose number of votes passes some minimum support level and the case with no such minimum. We provide polynomial-time algorithms to solve some of these problems and prove that the others are NP-hard.

cs.GT↗