The Three Levels of Public Interest AI

The transition from the Biden-Harris administration to the second Trump administration will bring big changes to presidential leadership and priorities on artificial intelligence (AI). The Trump agenda includes rescinding President Biden’s sweeping executive order on AI that placed equity and civil rights on the same level as innovation and competitiveness. The Biden administration’s Blueprint for an AI Bill of Rights that spelled out key principles for AI’s impact on society will likely also get buried.

But bringing AI to heel for the common good does not need to stop on Inauguration Day. Concern about AI’s risks is bipartisan, even if there are differences over priorities and solutions. The Biden administration kept the first Trump administration’s 2019 AI executive order alive, hopefully setting a precedent for some degree of bipartisan continuity. Moreover, state legislatures and governors’ offices are fizzing with AI initiatives and turning the laboratories of democracy into laboratories for AI policy and practice innovation. What are the priorities for public interest AI?

These are turbulent times, and public interest AI must compete for attention in a crowded political space and rapidly changing commercial environment. But getting to grips with it can be a convoluted affair requiring an understanding of a vast terrain of technologies, challenges, and insider jargon. It is too much for all but the most expert or interested to get their arms around. To cut through the fog, the field should prioritize painting a vivid picture of what it wants for society from safe and responsible AI. What are we solving for? As Dr. Alondra Nelson, one of the key framers of the AI Bill of Rights, wrote in Foreign Affairs: "[When] tackling AI governance, it is crucial for leaders to consider not only what specific threats they fear from AI but what type of society they want to build.”

The AI field can look to the military’s three levels of warfare — tactical, operational, and strategic — for inspiration. The tactical level of public interest AI covers how organizations plan and execute targeted efforts advancing objectives. For example, a health insurer might use generative AI-powered co-pilots to boost the productivity and job satisfaction of member service agents. Tactical use cases feed into the operational level that connects organization-wide resources to strategic objectives, such as the insurer investing in AI to smooth the transition to value-based care that centers health outcomes, provider performance, and the patient experience.

A suite of tools and services are available to support the operational and tactical levels, such as existing consumer protection and civil rights laws applied to AI, non-enforceable but influential soft law tools such as the OECD AI Principles, red teaming, AI nutrition labels, and methods such as AI alignment. But even at its best, operational and tactical efforts are scattershot without an overarching strategic charge bringing public interest priorities into focus. That’s where the strategic level of public interest AI comes in. While the operational and tactical levels are concerned with mid-to-downstream matters, the strategic level seeks to shape the start of the stream by connecting bold and measurable public interest goals to AI research and development, investment and procurement, policy and governance, and other efforts shaping a field. It can be applied to vertical markets, such as improving health care or government services, and horizontal markets, such as older adults or racial equity, that are touched by a variety of industries and sectors using AI. Finally, strategic level work is relevant across the AI horizon, from mature fields, such as narrow AI, to new frontiers, such as large language models, to pre-dawn AI futures, such as the integration of AI and biotechnology.

Making the strategic level real can take the form of public interest AI term sheets that codify the terms and conditions for AI to best serve specific communities or concerns. Components include:

• A preamble articulating the key goals and other measurable objectives for the addressable public interest market that is agnostic to AI’s involvement;
• Defining critical problems and use cases relevant to AI’s capabilities; and
• Terms for key performance, ethics, safety, governance, and other requirements of AI models, systems, and products aligned with advancing strategic objectives.

Building out credible term sheets requires having the right mix of expertise from across socio, technical, and commercial fields and the right mix of people and interests at the drafting table.

How might this look? Consider government public benefits where AI and automation have a track record of causing harm. A Term Sheet for Public Interest AI and Government Benefits could be grounded in a customer service moonshot, ensuring everyone needing help can be screened for benefits such as SNAP (food assistance) and Medicaid (health care) in less than twenty minutes, receive an eligibility determination within twenty-four hours, and enjoy dignity-enhancing customer service along the way. Today’s commercial AI solutions can help government process paperwork, crunch data, and automate dreary tasks currently performed by humans more efficiently but left to be a traditional commercial transaction without public interest guardrails, AI could achieve the twenty minutes, twenty-four hours goal but cause harm by making service delivery cruder and less equitable and making already difficult and poorly paid benefits administration jobs more miserable.

Or consider AI serving the nearly 60 million strong older Americans market. A Term Sheet for Public Interest AI and Older Adults could be grounded in AARP’s Vision for a National Plan on Aging’s four goals:

1. Healthy living and access to affordable, high-quality health care,
2. Maximizing the dignity, independence, and protection of older adults,
3. Expanding opportunities for financial wellbeing, and
4. Creating age-friendly, livable communities supporting aging in place.

The vision includes data-driven decision-making and accountability as a cross cutting principle and calls out the role of technology in improving outcomes as people age. A term sheet would drill down on defining AI’s role in, for example, transforming the health care system and financial services, both big commercial opportunities in their own right, with an eye towards the unique needs and best interests of an aging population.

The three levels of public interest AI cannot replace sound laws and regulations or responsible business practices but give those worried about AI a practical way to assert agency over its evolution and make it easier to use the levers of politics, the marketplace, and communities to drive the change and progress we need for AI to benefit all.