Webinar Addresses Ethical Principles for Statistical Practitioners

Jana Eggers is the CEO and a board member at NARA Logics, a company focusing on neuroscience-based artificial intelligence. She is a mathematician and computer scientist who brings extensive leadership experience and cutting-edge technology expertise to NARA Logics. She previously worked with companies such as Intuit, Lycos, American Airlines, and Los Alamos National Laboratory, as well as various startups. As Eggers advances the frontiers of AI, her dedication to ethical implementation remains a central aspect of her work.

Andrew Gelman is a professor of statistics and political science at Columbia University. He has a talent for explaining complex concepts to both a technical audience and the public, as demonstrated by his numerous articles in technical journals and an equal number in The New York Times and other popular press outlets. Gelman explores the intersection of statistics and ethics, emphasizing the importance of integrity in communicating statistical analysis.

Jana Eggers, CEO of Nara Logics, and Andrew Gelman, Columbia University professor of statistics and political science, participated in a webinar related to the ASA’s Statement on Ethical AI Principles for Statistical Practitioners on February 6. The event was sponsored by the ASA Committee on Professional Ethics and Committee on Data Science and Artificial Intelligence.

Last fall, CoPE and CoDSAI collaborated to develop the principles, recognizing that AI and statistics are critical pillars in modern practice. The essence is that the ASA advocates for trustworthy AI through robust ethical guidelines. The statement encapsulates its message through the following three guiding principles:

1.Accountability. This principle emphasizes the importance of auditability and governance while maintaining professional competence and avoiding blind reliance on AI. It calls for a clear operational framework to monitor the use and deployment of AI systems.

2.Transparency. This principle highlights the need for clear communication about AI tasks, benefits, and risks. This includes providing thorough documentation and addressing biases and limitations inherent in AI systems.

3.Fairness. This principle advocates balancing individual and collective interests to promote equity and mitigate biases to prevent exploitation.

Mark Glickman from CoDSAI highlighted the relevance of ethical AI principles for statistical practitioners, providing three examples of how statisticians can engage in this dialogue. The first example was related to accountability in algorithmic hiring practices. The second emphasized transparency in the context of robo-advisers in finance. And the third example addressed fairness, particularly in criminal justice. Through these examples, Glickman underscored the importance of statisticians as advocates for ethical AI practice in interdisciplinary teams.

By educating collaborators and addressing potential ethical pitfalls, statisticians can ensure AI tools are developed and deployed responsibly, fostering a more equitable and transparent landscape in statistical practice and AI development.

After the introduction, Stephanie Shipp from CoPE guided the conversation and Q&A with the audience. Below is a summary of Eggers and Gelman’s conversation.

Why is ethical AI important to you, especially in the statistical and scientific world?

Eggers emphasized the importance of ethical considerations at the intersection of AI and traditional statistical methods, drawing from her experience in computational chemistry at Los Alamos. She pointed out the challenges faced when transitioning from statistical techniques to AI, particularly in explaining models to material scientists who relied on her calculations for fabrication decisions. This made her passionate about ethical AI, focusing on verifying data, algorithms, and objectives, especially when working with sensitive applications such as military technology. She noted that understanding data and algorithms’ intricacies is crucial to ensuring ethical practices and that all three elements—the algorithm, data, and objective—should be scrutinized for ethical implications in AI development.

Gelman noted that the importance of ethical AI, particularly in statistics and science, cannot be understated, especially when considering the potential political and military implications. While discussions often revolve around algorithmic fairness and privacy, the more pressing concerns involve the scenarios of AI misapplication, such as autonomous weapons and AI-engineered pandemics. In his view, these issues pose a greater threat than academic concerns, and he emphasized the need to frame the discussion around AI ethics in the context of its potential misuse.

Ethics in AI is fundamentally rooted in human behavior and the organizational structures that encourage cooperative action. Statisticians tend to operate in a more collaborative environment, which places them in a unique position to contribute to discussions about ethical AI.

In answering Eggers’s question about the ethical role storytelling plays in statistical practice, Gelman said storytelling is crucial in understanding complex concepts, often influencing how we interpret information and methods, especially in fields such as statistics and science. According to Gelman, rather than simply presenting ideas, compelling storytelling involves exploring the assumptions behind them and the logical consequences that follow, creating a narrative in which the outcome is not immediately obvious. This approach fosters reflection on surprising findings and challenges in predictive models, such as AI, where acknowledging failures can enhance credibility and encourage a more constructive dialogue among proponents and critics. By embracing examples of where methods fall short, both sides can find common ground, advancing understanding and collaboration in pursuing knowledge.

Share examples related to the ASA’s Statement on Ethical AI Principles or examples that continue to help us learn about AI ethics in
statistics and science.

Gelman emphasized the delicate balance statistics occupies between data abundance and scarcity. He noted that while vast amounts of data allow one to draw conclusions directly, this relationship shifts in many fields, particularly those with limited data such as political science. He argued that core statistics thrive in scenarios in which data is sufficient to inform decisions but not overwhelming to render analysis unnecessary. This balance is essential for ensuring statistical methods add value, rather than simply reaffirming existing biases or misinterpretations.

Eggers built upon Gelman’s insights, emphasizing the importance of accountability and critical thinking in statistical analysis. She encourages practitioners to question the objectives behind data collection and the technology used for analysis. By stressing the importance of not relying unthinkingly on advanced technologies, she advocates considering whether simpler methods, such as decision trees, might be more effective and maintainable.

What is the most important aspect of AI ethics, and how can we make a difference in promoting AI ethics?

Eggers highlighted the significance of explainable AI in fostering trust and understanding in technology. She explained that as AI systems become more integrated into decision-making processes, users must comprehend how these systems arrive at their conclusions. This transparency boosts confidence and allows for better accountability in AI applications.

Additionally, Eggers emphasized the importance of maintaining open communication within teams, as it fosters collaboration and trust among members. She pointed out that team members feeling comfortable sharing their ideas and concerns leads to more innovative solutions and a stronger sense of belonging.

How do we balance AI’s environmental impact with the potential benefits of deploying it in certain scenarios?

Eggers expressed concern about the widespread adoption of AI without a conscious awareness of its environmental impact. She emphasized that their AI is designed to be computationally efficient and runs on low-cost hardware. Despite this efficiency, she noted many users don’t prioritize environmental considerations when choosing software, often driven by a short-term mindset that equates technological abundance with necessity. She believes this mindset is influenced by societal messaging promoting resource-intensive solutions, even when alternatives exist.

As educators in a data analytics program, how do you thoughtfully integrate AI tools into the curriculum?

Gelman suggested a grading strategy that de-emphasizes homework and emphasizes class participation. He focuses on creating an engaging classroom environment by encouraging students to put away their phones and ensuring exams are conducted in class. He acknowledged that different instructors might approach the use of AI tools in several ways, admitting that he does not have a definitive answer to the ethical considerations surrounding their use.

Eggers suggested emphasizing a broader understanding of tools beyond large language models. She expressed concern about the prevalent focus on LLMs, noting that while they facilitate communication, decision-making occurs differently within the brain, often not reliant on language. To address this, she encourages students to explore a variety of tools and approaches. In her organization, they foster a culture of learning through initiatives like their AI chat sessions and a paper club, where participants engage with diverse topics, ranging from technical subjects to contemporary research, thus promoting critical thinking and active participation.

Watch the full webinar and download the slides.