The ever-increasing speed of computing devices, connectivity of the Internet, the Internet of things, and the sophistication of algorithms comes with the promise of immense prosperity. However, these advances have revealed their dark side; algorithms can increase economic inequity, can be discriminatory, reinforce human prejudices, polarize opinions, accelerate the spread of misinformation, and are generally not as reliable as they are widely thought to be.
Our goal is to reconsider the foundational principles that underlie the design of modern AI systems, take inspiration from nature, incorporate knowledge about processes that generate the data that underlies their training, and study the impact of such algorithms on humans and society at large. Our aim is that this effort leads to the development of tools – both algorithmic and regulatory – that synthesize knowledge from various relevant disciplines. Ultimately, these developments will enable well-informed recommendations for approaches to not only design AI systems but also for policy-making and governance that are adapted to this new ecosystem.
Examples of questions and directions include:
- Understand the foundations of intelligence: Understand principles of intelligent phenomena in nature, draw inspiration from natural systems to search for novel AI paradigms, develop new methods to reduce dependence on compute and data of modern AI systems.
- Understand how humans behave in the face of algorithms: Measure how human behavior changes when interacting with and through algorithms, and how these changes affect individual and group outcomes.
- Quantify/understand how algorithms impact humans: Develop quantitative measures and models of how training data affects algorithmic outputs, and how these algorithms, in turn, impact the options available to, opinions, and decision-making abilities of humans.
- Design computational methods that empower society: Develop a theory and practice of regulating algorithms and develop tools that empower citizens and policymakers alike over key issues such as privacy, learning, and decision-making.
- Devise and derive recommendations: Suggest approaches to policy/lawmaking and governance that are adapted to this new world of ubiquitous algorithmic decision-making.
Assistant Professor of Statistics and Data Science