A Mechanism-Based Approach to Mitigating Harms from Persuasive Generative AI

Recent generative AI systems have demonstrated more advanced persuasive capabilities and are increasingly permeating areas of life where they can influence decision-making. Generative AI presents a new risk profile of persuasion due the opportunity for reciprocal exchange and prolonged interactions. This has led to growing concerns about harms from AI persuasion and how they can be mitigated, highlighting the need for a systematic study of AI persuasion. The current definitions of AI persuasion are unclear and related harms are insufficiently studied. Existing harm mitigation approaches prioritise harms from the outcome of persuasion over harms from the process of persuasion. In this paper, we lay the groundwork for the systematic study of AI persuasion. We first put forward definitions of persuasive generative AI. We distinguish between rationally persuasive generative AI, which relies on providing relevant facts, sound reasoning, or other forms of trustworthy evidence, and manipulative generative AI, which relies on taking advantage of cognitive biases and heuristics or misrepresenting information. We also put forward a map of harms from AI persuasion, including definitions and examples of economic, physical, environmental, psychological, sociocultural, political, privacy, and autonomy harm. We then introduce a map of mechanisms that contribute to harmful persuasion. Lastly, we provide an overview of approaches that can be used to mitigate against process harms of persuasion, including prompt engineering for manipulation classification and red teaming. Future work will operationalise these mitigations and study the interaction between different types of mechanisms of persuasion.

Fine-tuning language models to find agreement among humans with diverse preferences

Recent work in large language modeling (LLMs) has used fine-tuning to align outputs with the preferences of a prototypical user. This work assumes that human preferences are static and homogeneous across individuals, so that aligning to a a single "generic" user will confer more general alignment. Here, we embrace the heterogeneity of human preferences to consider a different challenge: how might a machine help people with diverse views find agreement? We fine-tune a 70 billion parameter LLM to generate statements that maximize the expected approval for a group of people with potentially diverse opinions. Human participants provide written opinions on thousands of questions touching on moral and political issues (e.g., "should we raise taxes on the rich?"), and rate the LLM's generated candidate consensus statements for agreement and quality. A reward model is then trained to predict individual preferences, enabling it to quantify and rank consensus statements in terms of their appeal to the overall group, defined according to different aggregation (social welfare) functions. The model produces consensus statements that are preferred by human users over those from prompted LLMs (>70%) and significantly outperforms a tight fine-tuned baseline that lacks the final ranking step. Further, our best model's consensus statements are preferred over the best human-generated opinions (>65%). We find that when we silently constructed consensus statements from only a subset of group members, those who were excluded were more likely to dissent, revealing the sensitivity of the consensus to individual contributions. These results highlight the potential to use LLMs to help groups of humans align their values with one another.

Assessing Group-level Gender Bias in Professional Evaluations: The Case of Medical Student End-of-Shift Feedback

Although approximately 50% of medical school graduates today are women, female physicians tend to be underrepresented in senior positions, make less money than their male counterparts and receive fewer promotions. There is a growing body of literature demonstrating gender bias in various forms of evaluation in medicine, but this work was mainly conducted by looking for specific words using fixed dictionaries such as LIWC and focused on recommendation letters. We use a dataset of written and quantitative assessments of medical student performance on individual shifts of work, collected across multiple institutions, to investigate the extent to which gender bias exists in a day-to-day context for medical students. We investigate differences in the narrative comments given to male and female students by both male or female faculty assessors, using a fine-tuned BERT model. This allows us to examine whether groups are written about in systematically different ways, without relying on hand-crafted wordlists or topic models. We compare these results to results from the traditional LIWC method and find that, although we find no evidence of group-level gender bias in this dataset, terms related to family and children are used more in feedback given to women.

Can language models learn from explanations in context?

Large language models can perform new tasks by adapting to a few in-context examples. For humans, rapid learning from examples can benefit from explanations that connect examples to task principles. We therefore investigate whether explanations of few-shot examples can allow language models to adapt more effectively. We annotate a set of 40 challenging tasks from BIG-Bench with explanations of answers to a small subset of questions, as well as a variety of matched control explanations. We evaluate the effects of various zero-shot and few-shot prompts that include different types of explanations, instructions, and controls on the performance of a range of large language models. We analyze these results using statistical multilevel modeling techniques that account for the nested dependencies among conditions, tasks, prompts, and models. We find that explanations of examples can improve performance. Adding untuned explanations to a few-shot prompt offers a modest improvement in performance; about 1/3 the effect size of adding few-shot examples, but twice the effect size of task instructions. We then show that explanations tuned for performance on a small validation set offer substantially larger benefits; building a prompt by selecting examples and explanations together substantially improves performance over selecting examples alone. Hand-tuning explanations can substantially improve performance on challenging tasks. Furthermore, even untuned explanations outperform carefully matched controls, suggesting that the benefits are due to the link between an example and its explanation, rather than lower-level features of the language used. However, only large models can benefit from explanations. In summary, explanations can support the in-context learning abilities of large language models on

Growing knowledge culturally across generations to solve novel, complex tasks

Knowledge built culturally across generations allows humans to learn far more than an individual could glean from their own experience in a lifetime. Cultural knowledge in turn rests on language: language is the richest record of what previous generations believed, valued, and practiced. The power and mechanisms of language as a means of cultural learning, however, are not well understood. We take a first step towards reverse-engineering cultural learning through language. We developed a suite of complex high-stakes tasks in the form of minimalist-style video games, which we deployed in an iterated learning paradigm. Game participants were limited to only two attempts (two lives) to beat each game and were allowed to write a message to a future participant who read the message before playing. Knowledge accumulated gradually across generations, allowing later generations to advance further in the games and perform more efficient actions. Multigenerational learning followed a strikingly similar trajectory to individuals learning alone with an unlimited number of lives. These results suggest that language provides a sufficient medium to express and accumulate the knowledge people acquire in these diverse tasks: the dynamics of the environment, valuable goals, dangerous risks, and strategies for success. The video game paradigm we pioneer here is thus a rich test bed for theories of cultural transmission and learning from language.

Improving Coherence and Consistency in Neural Sequence Models with Dual-System, Neuro-Symbolic Reasoning

Human reasoning can often be understood as an interplay between two systems: the intuitive and associative ("System 1") and the deliberative and logical ("System 2"). Neural sequence models -- which have been increasingly successful at performing complex, structured tasks -- exhibit the advantages and failure modes of System 1: they are fast and learn patterns from data, but are often inconsistent and incoherent. In this work, we seek a lightweight, training-free means of improving existing System 1-like sequence models by adding System 2-inspired logical reasoning. We explore several variations on this theme in which candidate generations from a neural sequence model are examined for logical consistency by a symbolic reasoning module, which can either accept or reject the generations. Our approach uses neural inference to mediate between the neural System 1 and the logical System 2. Results in robust story generation and grounded instruction-following show that this approach can increase the coherence and accuracy of neurally-based generations.

Communicating Natural Programs to Humans and Machines

The Abstraction and Reasoning Corpus (ARC) is a set of tasks that tests an agent's ability to flexibly solve novel problems. While most ARC tasks are easy for humans, they are challenging for state-of-the-art AI. How do we build intelligent systems that can generalize to novel situations and understand human instructions in domains such as ARC? We posit that the answer may be found by studying how humans communicate to each other in solving these tasks. We present LARC, the Language-annotated ARC: a collection of natural language descriptions by a group of human participants, unfamiliar both with ARC and with each other, who instruct each other on how to solve ARC tasks. LARC contains successful instructions for 88\% of the ARC tasks. We analyze the collected instructions as `natural programs', finding that most natural program concepts have analogies in typical computer programs. However, unlike how one precisely programs a computer, we find that humans both anticipate and exploit ambiguities to communicate effectively. We demonstrate that a state-of-the-art program synthesis technique, which leverages the additional language annotations, outperforms its language-free counterpart.

A practical introduction to the Rational Speech Act modeling framework

Recent advances in computational cognitive science (i.e., simulation-based probabilistic programs) have paved the way for significant progress in formal, implementable models of pragmatics. Rather than describing a pragmatic reasoning process in prose, these models formalize and implement one, deriving both qualitative and quantitative predictions of human behavior -- predictions that consistently prove correct, demonstrating the viability and value of the framework. The current paper provides a practical introduction to and critical assessment of the Bayesian Rational Speech Act modeling framework, unpacking theoretical foundations, exploring technological innovations, and drawing connections to issues beyond current applications.

The Language of Generalization

Language provides simple ways of communicating generalizable knowledge to each other (e.g., "Birds fly", "John hikes", "Fire makes smoke"). Though found in every language and emerging early in development, the language of generalization is philosphically puzzling and has resisted precise formalization. Here, we propose the first formal account of the language of generalization that makes quantitative predictions about human understanding. We test our model in three diverse domains: generalizations about categories (generic language), events (habitual language), and causes (causal language). The model explains the gradience in human endorsement through the interplay between a simple truth-conditional semantic theory and diverse beliefs about properties. This work opens the door to understanding precisely how abstract knowledge is learned from language.

Practical optimal experiment design with probabilistic programs

Scientists often run experiments to distinguish competing theories. This requires patience, rigor, and ingenuity - there is often a large space of possible experiments one could run. But we need not comb this space by hand - if we represent our theories as formal models and explicitly declare the space of experiments, we can automate the search for good experiments, looking for those with high expected information gain. Here, we present a general and principled approach to experiment design based on probabilistic programming languages (PPLs). PPLs offer a clean separation between declaring problems and solving them, which means that the scientist can automate experiment design by simply declaring her model and experiment spaces in the PPL without having to worry about the details of calculating information gain. We demonstrate our system in two case studies drawn from cognitive psychology, where we use it to design optimal experiments in the domains of sequence prediction and categorization. We find strong empirical validation that our automatically designed experiments were indeed optimal. We conclude by discussing a number of interesting questions for future research.