Algorithmic Foundations of Trustworthy AI

Algorithms play a significant role in our lives, from news recommendations in the morning, to medical diagnostics when we are sick— and they are drastically transforming our society. However, the use of social algorithms has revealed that today’s well-established algorithm theories are no longer adequate for our modern society. For many years, the focus of the algorithmic theory was on computational efficiency and correctness. But how could it have happened that technology intended to help people instead produces algorithms with various biases?

When algorithms directly impact people’s lives, this raises questions of fundamental social values like fairness, ethics, privacy, justice, safety, accountability, or even morality. Our mission is to build a computational framework to understand, develop and design Algorithms for Trustworthy AI — socially responsible/responsive algorithms that incorporate human values into AI systems. 

This innovative approach includes the use of multivariable algorithms coupled with social theory to establish the foundation for trustworthy AI. The lack of theoretical understanding is a severe obstacle to the trustability of modern AI algorithms.

The two strongly interconnected research questions we propose to address are:

• How to develop algorithms with guaranteed performance (such as scalability, running time, approximation, or space complexity) for the needs of constraining algorithmic optimization? 
• How to design models with the right balance between algorithmic bias and algorithmic effectiveness? 

In this project, we will exploit cross-domain synergies and provide a collaborative platform for today’s leading scientists and the promising ones of tomorrow to pursue ambitious scientific objectives around new and innovative algorithms for social good.

Our approach is: Algorithmic Principles + Formal Social Models = Trustworthy AI.  

AI generated using Microsoft Designer