Essential Role of Statistics in the Evolution of Trustworthy AI

I recently read an article in MIT Sloan Management Review titled “AI and Statistics: Perfect Together.” This title reflects my already firm belief that statistics has an essential role to play in AI development. AI can affect society in a positive way, helping address societal challenges ranging from health and nutrition to equality and inclusion. In our daily work, it can be an effective assistant for tasks such as coding, summarizing, and reviewing literature. It can also help automate (daily) activities subject to human error, such as driving.

In the past year, there have been many exciting developments, especially in generative AI. However, the news has not been uniformly positive. Along the way, there have been negative headlines—for example, the rollout of Microsoft’s chatbot with the headline “Dataset Trained Microsoft Chatbot to Spew Racist Tweets.” In response, the European Union introduced new regulatory frameworks for AI. They were also needed in the United States, reflected in the White House executive order on the safe, secure, and trustworthy development and use of AI.

AI is nearly always associated with computer science and engineering. However, AI has required substantial statistical input along its entire developmental path, though it depends on massive computational resources. Statisticians have made groundbreaking contributions to both existing and emerging areas (ranging from public polling to automated driving systems). But not everyone recognizes the value added of our discipline or the central role the American Statistical Association plays in sustaining and enhancing our influence.

In fact, statistics and statisticians are essential to the deployment of safe, secure, and trustworthy AI.

ASA Initiatives Emphasizing Our Role in the Evolution of Safe AI

To quote the ASA Committee on Data Science and Artificial Intelligence in the ASA Statement on the Role of Statistics in Data Science and Artificial Intelligence:

With that elegant statement setting the tone, I would like to highlight a few initiatives in our community designed to showcase the role of our discipline, brainstorm around problem solving, and build bridges between our field and computer science engineering.

Joint Statistical Meetings
My JSM invited speaker was Jason Matheny, president and chief executive officer of RAND. Before becoming RAND’s president, Matheny led White House policy on technology and national security at the National Security Council and Office of Science and Technology Policy. In his JSM speech, “Working at the Intersection of Statistics and AI Policy,” Matheny discussed open problems with AI and statistics, focusing on AI’s technical issues, AI policy, and how they manifest in real-world consequences. “Statisticians are poised to play critical roles in the design, implementation, and evaluation of AI policy,” he said, noting societal impacts (e.g., workforce, economic benefits, security). Matheny also explored how statisticians can support the development of fairness metrics, security measures, and model evaluation.

IDEA Forum
The ASA president and members of the board of directors will host the annual Influencing Discovery Exploration & Action (IDEA) Forum as a virtual event on November 22 from 3:00 p.m. to 4:30 p.m. ET. In collaboration with a working group led by Lingzhou Xue of Penn State and Mark Glickman of Harvard University, we have decided to focus on trustworthy AI and related principles. This follows President Joe Biden’s lead with the White House executive order. The event will feature presentations from leaders in the field about trustworthy AI and related principles such as fairness and transparency, followed by a moderated Q&A.

Our first panelist is Jeannette Wing, a leading expert in trustworthy AI. Wing is executive vice president for research and professor of computer science at Columbia University. She thinks the promise of AI is immense, but noting the systems can be brittle and unfair, she says, “For society to reap the benefits of AI systems, society needs to be able to trust them.” Wing envisions a future in which we deliver on the promise of the benefits of AI and address scenarios that have life-critical consequences for people and society.

Our second panelist is Arvind Narayanan, a leading expert in fairness and machine learning. Narayanan is an associate professor of computer science at Princeton University, where he studies the societal impact of AI. His new book, AI Snake Oil, cuts through AI hype and provides an essential understanding of how AI works and why it often doesn’t. His book is based on years of research on limits to prediction, the dangers of predictive decision-making, the reproducibility crisis in machine learning–based science, the limits of large language models, and the risks of social media algorithms.

We are in the process of confirming additional speakers. The IDEA Forum is particularly significant for statistical practitioners. As the integration of AI into various aspects of data analysis and decision-making continues, understanding the principles of trustworthy AI and related principles becomes essential.

CSAB Update

At the August board meeting, I welcomed an update about the ASA role in CSAB from past president Kathy Ensor of Rice University and Dave Hunter of Penn State. My takeaways were that we are showcasing our essential role in data science and strengthening ties to computer science and engineering, which will help facilitate future collaborations.

In April 2021, the ASA became a full member of CSAB, formally known as the Computer Science Accreditation Board, joining the world’s two largest professional and technical societies for computing: the Association for Computing Machinery and Institute of Electrical and Electronics Engineers Computer Society. As the lead ABET, formally known as the Accreditation Board for Engineering and Technology, member society for computing, CSAB is responsible for developing accreditation criteria for several programs, including data science.

In their respective roles, Ensor and Hunter have ensured the ASA’s collective voice is being heard in existing and new initiatives within CSAB. A few of those merit mention. First, CSAB successfully made the case within ABET that CSAB should lead the development of data science criteria. Also, CSAB’s board recognizes statisticians need to be an integral part of formulating these criteria, given the widely acknowledged strong role statistics plays in data science.

Ensor’s term on the CSAB Board has ended. Hunter will assume the role of board representative, with Kelly McConville of Bucknell University assuming the role of alternate. This fall, CSAB will formally request that ABET commissions CSAB to be the leading society for AI and machine learning programs. We are grateful for Ensor’s leadership and service, as well as to Hunter and McConville for continuing to represent the ASA in this important arena.

I want to thank each of you for your continued support and involvement in our community’s initiatives. It’s your enthusiasm and commitment that make all these projects possible and meaningful. We have some exciting times ahead, and I can’t wait to see how all our initiatives evolve. Remember, your feedback and ideas are always welcome, so don’t hesitate to reach out. Let’s keep the momentum going. Let’s keep making a difference!