Improving oversight, asking the right questions, and cultivating diversity in AI systems
AI Ethics: Global Perspectives is a free, online course jointly launched by The Governance Lab (The GovLab) at the NYU Tandon School of Engineering, The Global AI Ethics Consortium (GAIEC), Center for Responsible AI @ NYU (R/AI), and the TUM Institute for Ethics in Artificial Intelligence (IEAI), which aims to convey the breadth and depth of the ongoing interdisciplinary conversation around AI ethics. The course seeks to bring together diverse perspectives from the field of ethical AI, to raise awareness and help institutions work towards more responsible use.
On February 24, 2021, The GovLab, GAIEC, R/AI, IEAI, and NYU Tandon School of Engineering hosted the first panel discussion for AI Ethics: Global Perspectives, to accompany the release of four new modules in the month of February and enable deeper dialogue on the issues surfaced in each. Moderated by Stefaan Verhulst, Co-Founder and Chief Research and Development Officer of The GovLab and Julia Stoyanovich, Director of the Center for Responsible AI at NYU, the panel featured distinguished AI Ethics faculty members:
- Idoia Salazar, President and Co-Founder of OdiselA (Module: Alexa vs Alice: Cultural Perspectives on the Impact of AI);
- Danya Glabau, Director of Science and Technology studies at the NYU Tandon School of Engineering (Module: AI for whom?); and
- Jerry John Kponyo, Associate Professor of Telecommunication Engineering, Kwame Nkrumah University of Science and Technology (KNUST) (Module: Applications of Artificial Intelligence (AI) in Transport and Safety).
In this 60-minute panel, moderators and panelists discussed the importance of diversity in AI systems, ways to improve the design and development of AI in the public sector, and the role that Institutional Review Boards (IRB) can play in the oversight.
Learning from multi-disciplinary approaches
Panelists spoke in-depth about their experience with AI technologies in their respective fields of work, from which several cross-cutting ethical concerns emerged.
Given his module’s focus on application of AI for transportation systems, which can be deployed to reduce travel time and improve safety procedures, Jerry Kponyo discussed challenges in gathering traffic data to appropriately train AI systems for transportation management; currently, many systems are trained on historical datasets that may be outdated or biased. Jerry raised other emerging ethical concerns in the field of AITS (AI in Transportation Systems), such as electronic policing, in which information gathered about civilians using transportation systems is exploited for nefarious purposes, raising serious privacy concerns.
Danya Glabau, meanwhile, framed her thoughts in relation to the deployment of AI in health care systems. She discussed popular applications of AI in health care, such as in automated risk detection and improving efficiencies for appointment scheduling. Danya emphasized the importance of consistently framing and asking the right questions to improve the design and evaluation process of AI systems:
“Is the motivation to adopt AI only to save money and time? Or are we thinking about whole patients? Are we thinking about patients in the ways that they interface with their communities? In the infrastructural system in which they work? [..] How can you design the process to be more inclusive of a wider range of stakeholders?”
Opportunities to improve AI Systems through public consultation and education
Panelists also discussed ways to improve the design, development, and implementation of AI. Danya noted the importance of public consultation, and seeking input from stakeholders with varying degrees of expertise, such as community organizers or ethicists, who can contextualize the application of systems.
Idoia Salazar is the Co-Founder of the Observatorio Del Impacto Social y Ético de la Inteligencia Artificial (OdiselA), an organization which aims to create a global awareness towards the responsible use of AI. Idoia discussed lessons from her work at the observatory, where she has seen the need to especially cultivate this awareness among high-level stakeholders in the private sector and government, such as corporate managers. She also highlighted the importance of overturning public misconceptions about the design and use of AI by investing in more public education.
Suggestions for improving oversight of AI systems
During the Q&A session of the discussion, Julia Stoyanovich asked panelists about ways in which we might improve oversight for AI systems.
Danya spoke to the role that IRBs might play in asking the right questions about the design and development process in the early research stages. Beyond external oversight, we can also adopt research policies and standards from adjacent field:
“Applying some of the questions that are asked in clinical research, or social science research, to algorithmic research would go a long way.”
Jerry reinforced this idea, discussing the need not only for oversight and standards, but also for legislation; once these standards and procedures are outlined in the law, the path to implementation becomes clearer.
“Without legislation”, he said, “enforcement becomes difficult.”
Idoia also highlighted the importance of diversifying oversight systems, to not think about regulating AI as a whole but to focus on specific use cases such as facial recognition technologies or autonomous vehicles.
Looking ahead: cultivating diversity across AI systems
The panel concluded with a question from Julia on how diversity plays a role in AI solutions — not limited to diversity in data sets, but also in teams, goals, and models.
In his concluding thoughts, Jerry summed up many points discussed throughout the panel, stressing the importance of diversity in making sure AI is used for its intended purpose and is relevant to the stakeholders it is being designed for.
“What diversity brings,” Jerry stated, “is to open the floor to allow you to be able to look at different perspectives as far as coming to a decision is concerned.”
Danya closed the panel by emphasizing the importance of diversity and the need to think beyond it.
“Diversity does not always equate a focus on justice [and equity]. A diverse individual could still be interested in the bottom line of making money over community well-being. […] But in thinking in terms of applications and appropriate applications of AI systems, [diversity thinking] can get us farther than where we sometimes seem to be with new AI systems today.”