Instance-based interpretation methods have been widely studied for supervised learning methods as they help explain how black box neural networks predict.

Computational drug repurposing for rare diseases is especially challenging when no prior associations exist between drugs and target diseases. Therefore, knowledge graph completion and message-passing GNNs have little reliable signal to learn and propagate, resulting in poor performance. We present RareAgent, a self-evolving multi-agent system that reframes this task from passive pattern recognition to active evidence-seeking reasoning. RareAgent organizes task-specific adversarial debates in which agents dynamically construct evidence graphs from diverse perspectives to support, refute, or entail hypotheses. The reasoning strategies are analyzed post hoc in a self-evolutionary loop, producing textual feedback that refines agent policies, while successful reasoning paths are distilled into transferable heuristics to accelerate future investigations. Comprehensive evaluations reveal that RareAgent improves the indication AUPRC by 18.1% over reasoning baselines and provides a transparent reasoning chain consistent with clinical evidence.

The method is general .

The AI alignment problem comprises both technical and normative dimensions. While technical solutions focus on implementing normative constraints in AI systems, the normative problem concerns determining what these constraints should be. This paper examines justifications for democratic approaches to the normative problem -- where affected stakeholders determine AI alignment -- as opposed to epistocratic approaches that defer to normative experts. We analyze both instrumental justifications (democratic approaches produce better outcomes) and non-instrumental justifications (democratic approaches prevent illegitimate authority or coercion). We argue that normative and metanormative uncertainty create a justificatory gap that democratic approaches aim to fill through political rather than theoretical justification. However, we identify significant challenges for democratic approaches, particularly regarding the prevention of illegitimate coercion through AI alignment. Our analysis suggests that neither purely epistocratic nor purely democratic approaches may be sufficient on their own, pointing toward hybrid frameworks that combine expert judgment with participatory input alongside institutional safeguards against AI monopolization.

Recent advances in Large Language Models (LLMs) have enabled multi-agent systems that simulate real-world interactions with near-human reasoning. While previous studies have extensively examined biases related to protected attributes such as race, the emergence and propagation of biases on socially contentious issues in multi-agent LLM interactions remain underexplored. This study explores how LLM agents shape public opinion through debates on five contentious topics. By simulating over 2,500 debates, we analyze how initially neutral agents, assigned a centrist disposition, adopt specific stances over time. Statistical analyses reveal significant group conformity mirroring human behavior; LLM agents tend to align with numerically dominant groups or more intelligent agents, exerting a greater influence. These findings underscore the crucial role of agent intelligence in shaping discourse and highlight the risks of bias amplification in online interactions. Our results emphasize the need for policy measures that promote diversity and transparency in LLM-generated discussions to mitigate the risks of bias propagation within anonymous online environments.

Training data attribution (TDA) methods aim to attribute model outputs back to specific training examples, and the application of these methods to large language model (LLM) outputs could significantly advance model transparency and data curation. However, it has been challenging to date to apply these methods to the full scale of LLM pretraining. In this paper, we refine existing gradient-based methods to work effectively at scale, allowing us to retrieve influential examples for an 8B-parameter language model from a pretraining corpus of over 160B tokens with no need for subsampling or pre-filtering. Our method combines several techniques, including optimizer state correction, a task-specific Hessian approximation, and normalized encodings, which we find to be critical for performance at scale. In quantitative evaluations on a fact tracing task, our method performs best at identifying examples that influence model predictions, but classical, model-agnostic retrieval methods such as BM25 still perform better at finding passages which explicitly contain relevant facts. These results demonstrate a misalignment between factual *attribution* and causal *influence*. With increasing model size and training tokens, we find that influence more closely aligns with factual attribution. Finally, we examine different types of examples identified as influential by our method, finding that while many directly entail a particular fact, others support the same output by reinforcing priors on relation types, common entities, and names. We release our prompt set and model outputs, along with a web-based visualization tool to explore influential examples for factual predictions, commonsense reasoning, arithmetic, and open-ended generation for an 8B-parameter LLM.

Tech companies like to make two grand pronouncements about the future of artificial intelligence. First, the technology is going to usher in a revolution akin to the advent of fire, nuclear weapons, and the internet. And second, it is going to cost almost unfathomable sums of money. Silicon Valley has already triggered tens or even hundreds of billions of dollars of spending on AI, and companies only want to spend more. Their reasoning is straightforward: These companies have decided that the best way to make generative AI better is to build bigger AI models.

In recent years Artificial Intelligence has emerged as a fundamental tool in medical applications. Despite this rapid development, deep neural networks remain black boxes that are difficult to explain, and this represents a major limitation for their use in clinical practice. We focus on the segmentation of medical images task, where most explainability methods proposed so far provide a visual explanation in terms of an input saliency map. The aim of this work is to extend, implement and test instead an influence-based explainability algorithm, TracIn, proposed originally for classification tasks, in a challenging clinical problem, i.e., multiclass segmentation of tumor brains in multimodal Magnetic Resonance Imaging. We verify the faithfulness of the proposed algorithm linking the similarities of the latent representation of the network to the TracIn output. We further test the capacity of the algorithm to provide local and global explanations, and we suggest that it can be adopted as a tool to select the most relevant features used in the decision process. The method is generalizable for all semantic segmentation tasks where classes are mutually exclusive, which is the standard framework in these cases.

For years now, Hakwan Lau has suffered from an inner torment. Lau is a neuroscientist who studies the sense of awareness that all of us experience during our every waking moment. How this awareness arises from ordinary matter is an ancient mystery. Several scientific theories purport to explain it, and Lau feels that one of them, called integrated information theory (IIT), has received a disproportionate amount of media attention. He's annoyed that its proponents tout it as the dominant theory in the press.