The Promise and Perils of Artificial Intelligence in Advancing Participatory Science and Health Equity in Public Health

Model-based reinforcement learning (MBRL) is a promising route to sampleefficient policy optimization. However, a known vulnerability of reconstructionbased MBRL consists of scenarios in which detailed aspects of the world are highly predictable, but irrelevant to learning a good policy. Such scenarios can lead the model to exhaust its capacity on meaningless content, at the cost of neglecting important environment dynamics. While existing approaches attempt to solve this problem, we highlight its continuing impact on leading MBRL methods —including DreamerV3 and DreamerPro — with a novel environment where background distractions are intricate, predictable, and useless for planning future actions. To address this challenge we develop a method for focusing the capacity of the world model through synergy of a pretrained segmentation model, a task-aware reconstruction loss, and adversarial learning. Our method outperforms a variety of other approaches designed to reduce the impact of distractors, and is an advance towards robust model-based reinforcement learning.

Vafa et al. (2024) introduced a transformer-based econometric model, CAREER, that predicts a worker’s next job as a function of career history (an “occupation model”). CAREER was initially estimated (“pre-trained”) using a large, unrepresentative resume dataset, which served as a “foundation model,” and parameter estimation was continued (“fine-tuned”) using data from a representative survey. CAREER had better predictive performance than benchmarks. This paper considers an alternative where the resume-based foundation model is replaced by a large language model (LLM). We convert tabular data from the survey into text files that resemble resumes and fine-tune the LLMs using these text files with the objective to predict the next token (word). The resulting fine-tuned LLM is used as an input to an occupation model. Its predictive performance surpasses all prior models. We demonstrate the value of fine-tuning and further show that by adding more career data from a different population, fine-tuning smaller LLMs surpasses the performance of fine-tuning larger models.

Language Model Programs, i.e. sophisticated pipelines of modular language model (LM) calls, are increasingly advancing NLP tasks, but they require crafting prompts that are jointly effective for all modules. We study prompt optimization for LM programs, i.e. how to update these prompts to maximize a downstream metric without access to module-level labels or gradients. To make this tractable, we factorize our problem into optimizing the free-form instructions and few-shot demonstrations of every module and introduce several strategies to craft task-grounded instructions and navigate credit assignment across modules. Our strategies include (i) program- and data-aware techniques for proposing effective instructions, (ii) a stochastic mini-batch evaluation function for learning a surrogate model of our objective, and (iii) a meta-optimization procedure in which we refine how LMs construct proposals over time. Using these insights we develop MIPRO, a novel algorithm for optimizing LM programs. MIPRO outperforms baseline optimizers on five of seven diverse multi-stage LM programs using a best-in-class open-source model (Llama-3-8B), by as high as 13% accuracy. We have released our new optimizers and benchmark in DSPy at [http://dspy.ai](http://dspy.ai).

Measuring the impact of online misinformation is challenging. Traditional measures, such as user views or shares on social media, are incomplete because not everyone who is exposed to misinformation is equally likely to believe it. To address this issue, we developed a method that combines survey data with observational Twitter data to probabilistically estimate the number of users both exposed to and likely to believe a specific news story. As a proof of concept, we applied this method to 139 viral news articles and find that although false news reaches an audience with diverse political views, users who are both exposed and receptive to believing false news tend to have more extreme ideologies. These receptive users are also more likely to encounter misinformation earlier than those who are unlikely to believe it. This mismatch between overall user exposure and receptive user exposure underscores the limitation of relying solely on exposure or interaction data to measure the impact of misinformation, as well as the challenge of implementing effective interventions. To demonstrate how our approach can address this challenge, we then conducted data-driven simulations of common interventions used by social media platforms. We find that these interventions are only modestly effective at reducing exposure among users likely to believe misinformation, and their effectiveness quickly diminishes unless implemented soon after misinformation’s initial spread. Our paper provides a more precise estimate of misinformation’s impact by focusing on the exposure of users likely to believe it, offering insights for effective mitigation strategies on social media.