A Blueprint for Equitable AI

The Aspen Institute Science & Society Program and DeepMind today released A Blueprint for Equitable AI, a new report that considers how to build and distribute artificial intelligence for equitable outcomes. The report acknowledges the transformative potential of AI in our society while underscoring the importance of involving members of marginalized communities in defining what constitutes truly equitable AI, with their visions tangibly communicated to developers. AI literacy and education was also discussed as a necessary prerequisite, with a clear need to develop strategies that span the public and private sectors.

“At DeepMind, we know that for the long-term impact of AI to be equitable, we need to pioneer responsibly, collaborate with civil society and seek feedback,” said Lila Ibrahim, Chief Operating Officer, DeepMind. “It was a privilege for us to lead this open dialogue – one of what I am sure will be many as AI technology scales within our societies.”

“Our hope is that this blueprint will serve as a starting point for technologists and the communities they serve,” said Aaron Mertz, Director of the Aspen Institute Science & Society Program. “To realize the full potential of AI, it is critical to engage companies, civil society, and policymakers to ensure that equity considerations are centered as these technologies continue to evolve.

A Blueprint for Equitable AI represents the discussions of two diverse groups of multidisciplinary, international experts convened by DeepMind and the Aspen Institute in late 2022. The groups included participants from technology companies including DeepMind, Meta, Amazon, and Salesforce; academic institutions such as Georgetown University, Cornell University, and the University of Oxford; and policy and advocacy organizations including the Brookings Institution, American Civil Liberties Union (ACLU), Aapti Institute, and the American Association of People with Disabilities. The full list of participants is available on page four of the report.

In the context of the conditions required for equitable AI, participants considered questions of definitions, access, scale, incentives, education, participation, and law. The groups agreed that open discussion of equitable outcomes will play a vital role in building the muscle and future institutions required for civil discourse on the role technology plays in our lives. Key points included:

• Involving members of marginalized communities in the process of defining equitable AI is key, so that their visions are clearly and tangibly communicated to developers, while ensuring that this work takes place without undue burdens and tokenization.
• AI literacy is a necessary prerequisite to equitable AI – for example, through K-12 curricula, massive open online courses (MOOCs), and town halls.
• As many current concerns regarding how AI will be developed and deployed stem from historic, entrenched patterns of marginalization, individual actors are unlikely to be able to remedy them alone. This underlines the importance of all groups – industry, governments and civil society – working together to set standards and co-create solutions.
• Significant work remains to create a universally accepted definition of equitable AI, and debate continues as to whether such a streamlined definition would be desirable.
• Further discussion and collaboration will be needed to translate thinking into concrete positions and practical recommendations. Asking the right questions, and inviting the right mix of people and groups to answer them, can be an impactful step towards greater equity.

The full report can be found here: www.aspeninstitute.org/publications/blueprint-for-equitable-ai/

‍