Accountable, Efficient Reinforcement Learning for Human Impact

The question of how an agent can learn to make decisions in a noisy, uncertain world is at the heart of artificial intelligence. I am interested in how such artificial intelligent agents can amplify human intelligence. Thinking about this requires us to advance the foundations and frameworks of reinforcement learning-- new algorithms that enable agents to quickly learn to make good decisions, and do so in a way that enables strong accountability through performance guarantees and bounds. My work is inspired and grounded in our applications to education and healthcare.

Ethics and Governance of AI

AI is a transformative technological development. What does AI enable others to do? What responsibilities does this imply for innovators, policymakers, and citizens? How can AI be aligned with human interests — as a technical matter, through governance frameworks, or by moral codes and commitments?

Safe and reliable integration of autonomous systems in our society

Taking a human-centered approach, I focus on developing algorithms for robots that interact, collaborate, and positively influence humans. My work combines model-based and data-driven techniques to learn computational models of human behavior, predict humans’ intents, and further study how humans adapt to one another or to AI agents. I work on developing policies for autonomous agents and robots that influence humans with the goal of advancing safe and interactive human-robot systems such as autonomous vehicles or domestic robots interacting with people.

Ensuring ethical use of Artificial Intelligence in healthcare

The use of machine intelligence in health care has the potential to revolutionize how we decide whether to act, when, and for whom. Given the promise, how will we balance fairness, equity, and patient privacy with the need to compile the massive datasets that AI needs? How will we decide on the degree of autonomy and decision-making authority that we grant to algorithmic agents? At HAI we solve these challenges via collaborations between medical researchers, computer scientists, social scientists, and humanists.

AI can be sexist and racist - it’s time to make it fair

Can we identify sources of human bias before building them into algorithms? Does gendering social robots reinforce social inequalities? We seek to understand technology in social context. Schiebinger directs Gendered Innovations, an international, collaborative project that harnesses the creative power of sex and gender analysis for innovation and discovery. Considering gender may add a valuable dimension to research. It may take research in new directions.
James Zou and Londa Schiebinger

How can computers learn, understand, and communicate in human languages?

In recent years, Chris has been full steam ahead on research using neural networks for language understanding, part of research breakthroughs that have seen enormous progress in speech recognition and generation, machine translation, question answering, and text generation. However, you’ll also find Chris deep into questions of human language structure and meaning. The primary questions of how languages can be learned and used, as humans do, remain largely unsolved, and Chris hopes for a new convergence between machine learning, cognitive science, knowledge-based reasoning, and language.

Smarter medicine?

How can HAI improve our interpretation of medical data? We focus on early detection of disease through the application of deep learning to data from wearables, cardiac imaging, circulating cells, and hospitalized patients. We strive to provide a framework for intelligent early disease detection and better patient outcomes and treatments.