Andrea Wynn

Andrea Wynn, Harsh Satija, Gillian Hadfield

While multi-agent debate has been proposed as a promising strategy for improving AI reasoning ability, we find that debate can sometimes be harmful rather than helpful. Prior work has primarily focused on debates within homogeneous groups of agents, whereas we explore how diversity in model capabilities influences the dynamics and outcomes of multi-agent interactions. Through a series of experiments, we demonstrate that debate can lead to a decrease in accuracy over time - even in settings where stronger (i.e., more capable) models outnumber their weaker counterparts. Our analysis reveals that models frequently shift from correct to incorrect answers in response to peer reasoning, favoring agreement over challenging flawed reasoning. We perform additional experiments investigating various potential contributing factors to these harmful shifts - including sycophancy, social conformity, and model and task type. These results highlight important failure modes in the exchange of reasons during multi-agent debate, suggesting that naive applications of debate may cause performance degradation when agents are neither incentivised nor adequately equipped to resist persuasive but incorrect reasoning.

Andrea Wynn, Ilia Sucholutsky, Thomas L. Griffiths

How can we build AI systems that can learn any set of individual human values both quickly and safely, avoiding causing harm or violating societal standards for acceptable behavior during the learning process? We explore the effects of representational alignment between humans and AI agents on learning human values. Making AI systems learn human-like representations of the world has many known benefits, including improving generalization, robustness to domain shifts, and few-shot learning performance. We demonstrate that this kind of representational alignment can also support safely learning and exploring human values in the context of personalization. We begin with a theoretical prediction, show that it applies to learning human morality judgments, then show that our results generalize to ten different aspects of human values -- including ethics, honesty, and fairness -- training AI agents on each set of values in a multi-armed bandit setting, where rewards reflect human value judgments over the chosen action. Using a set of textual action descriptions, we collect value judgments from humans, as well as similarity judgments from both humans and multiple language models, and demonstrate that representational alignment enables both safe exploration and improved generalization when learning human values.

Andrea Wynn, Metod Jazbec, Charith Peris et al. · amazon-science

Large language models (LLMs) demonstrate a remarkable ability to learn new tasks from a few in-context examples. However, this flexibility introduces safety concerns: LLMs can be influenced by incorrect or malicious demonstrations -- for example, if an adversary tampers with or injects harmful examples without a human supervisor noticing. This motivates principled designs in which the system itself includes built-in mechanisms to guard against such attacks. We propose a novel approach to limit the degree to which harmful demonstrations can degrade model performance. First, we define a baseline ``safe'' behavior for the model -- the model's performance given no in-context demonstrations (zero-shot). Next, we apply distribution-free risk control (DFRC) to control the extent to which in-context samples can decay performance below zero-shot. We achieve this by leveraging dynamic early exit prediction, ignoring later attention heads that attend the most to the unsafe inputs. Finally, we propose modifications to DFRC that allow it to both control risk for harmful inputs \textit{and} leverage performance and efficiency gains on helpful inputs. We present both theoretical and empirical results showing that our approach can effectively control risk for harmful in-context demonstrations while simultaneously achieving substantial computational efficiency gains with helpful demonstrations.