AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles (e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities, and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.

Our results suggest that the prevalence of child and adult hearing impairment is substantially higher in middle- and low-income countries than in high-income countries, demonstrating the global need for attention to hearing impairment.

Abstract. In some reinforcement learning problems an agent may be provided with a set of input policies, perhaps learned from prior experience or provided by advisors. We present a reinforcement learning with policy advice (RLPA) algorithm which leverages this input set and learns to use the best policy in the set for the reinforcement learning task at hand. We prove that RLPA has a sub-linear regret of O( √ T ) relative to the best input policy, and that both this regret and its computational complexity are independent of the size of the state and action space. Our empirical simulations support our theoretical analysis. This suggests RLPA may offer significant advantages in large domains where some prior good policies are provided.

While mobile phones have found broad application in bringing health, financial, and other services to the developing world, usability remains a major hurdle for novice and low-literacy populations. In this article, we take two steps to evaluate and improve the usability of mobile interfaces for such users. First, we offer an ethnographic study of the usability barriers facing 90 low-literacy subjects in India, Kenya, the Philippines, and South Africa. Then, via two studies involving over 70 subjects in India, we quantitatively compare the usability of different points in the mobile design space. In addition to text interfaces such as electronic forms, SMS, and USSD, we consider three text-free interfaces: a spoken dialog system, a graphical interface, and a live operator. Our results confirm that textual interfaces are unusable by first-time low-literacy users, and error prone for literate but novice users. In the context of healthcare, we find that a live operator is up to ten times more accurate than text-based interfaces, and can also be cost effective in countries such as India. In the context of mobile banking, we find that task completion is highest with a graphical interface, but those who understand the spoken dialog system can use it more quickly due to their comfort and familiarity with speech. We synthesize our findings into a set of design recommendations.

Deciding how to act in partially observable environments remains an active area of research. Identifying good sequences of decisions is particularly challenging when good control performance requires planning multiple steps into the future in domains with many states. Towards addressing this challenge, we present an online, forward-search algorithm called the Posterior Belief Distribution (PBD). PBD leverages a novel method for calculating the posterior distribution over beliefs that result after a sequence of actions is taken, given the set of observation sequences that could be received during this process. This method allows us to efficiently evaluate the expected reward of a sequence of primitive actions, which we refer to as macro-actions. We present a formal analysis of our approach, and examine its performance on two very large simulation experiments: scientific exploration and a target monitoring domain. We also demonstrate our algorithm being used to control a real robotic helicopter in a target monitoring experiment, which suggests that our approach has practical potential for planning in real-world, large partially observable domains where a multi-step lookahead is required to achieve good performance.

Intelligent machines using machine learning algorithms are ubiquitous, ranging from simple data analysis and pattern recognition tools to complex systems that achieve superhuman performance on various tasks. Ensuring that they do not exhibit undesirable behavior—that they do not, for example, cause harm to humans—is therefore a pressing problem. We propose a general and flexible framework for designing machine learning algorithms. This framework simplifies the problem of specifying and regulating undesirable behavior. To show the viability of this framework, we used it to create machine learning algorithms that precluded the dangerous behavior caused by standard machine learning algorithms in our experiments. Our framework for designing machine learning algorithms simplifies the safe and responsible application of machine learning.

No abstract

Online education is rapidly expanding in response to rising demand for higher and continuing education, but many online students struggle to achieve their educational goals. Several behavioral science interventions have shown promise in raising student persistence and completion rates in a handful of courses, but evidence of their effectiveness across diverse educational contexts is limited. In this study, we test a set of established interventions over 2.5 y, with one-quarter million students, from nearly every country, across 247 online courses offered by Harvard, the Massachusetts Institute of Technology, and Stanford. We hypothesized that the interventions would produce medium-to-large effects as in prior studies, but this is not supported by our results. Instead, using an iterative scientific process of cyclically preregistering new hypotheses in between waves of data collection, we identified individual, contextual, and temporal conditions under which the interventions benefit students. Self-regulation interventions raised student engagement in the first few weeks but not final completion rates. Value-relevance interventions raised completion rates in developing countries to close the global achievement gap, but only in courses with a global gap. We found minimal evidence that state-of-the-art machine learning methods can forecast the occurrence of a global gap or learn effective individualized intervention policies. Scaling behavioral science interventions across various online learning contexts can reduce their average effectiveness by an order-of-magnitude. However, iterative scientific investigations can uncover what works where for whom.