Researchers behind a new civic AI project called EJAI asked a useful question: Can artificial intelligence help people examine a complicated public issue without deciding the issue for them?

Their test case involved disputes over data centers. EJAI combined maps, source-grounded evidence, statements about missing information and two AI personas representing different stakeholder perspectives. In a small benchmark, the evidence retrieved by the system supported 98.4% of its factual claims. The researchers also found an important weakness: one persona sometimes omitted evidence that conflicted with the priorities it had been assigned.

That combination points toward an interesting role for civic AI.

Consider offshore drilling in Florida. An environmental organization may concentrate on ecological damage and spill risk. An energy company may emphasize jobs, investment and domestic production. Government may examine tax revenue, regulation and energy supply. A hotel owner on the Gulf Coast may care most about what even a low-probability spill could do to tourism.

Each perspective can contain accurate information. Each also selects which information deserves attention.

Most people experience those arguments separately. We read the environmental group's report, the company's economic analysis or the government's staff presentation. Social media pushes the separation further by surrounding us with people and sources that tend to share our priorities.

AI gives us the ability to assemble those perspectives in one place.

St. Petersburg has already lived through decisions where that capability would have been useful. The proposed Rays stadium and Historic Gas Plant redevelopment combined stadium financing, tourism revenues, redevelopment infrastructure, affordable housing commitments, economic development and the history of a displaced Black community.

Pinellas County approved $312.5 million in tourist development tax funding toward the proposed stadium in 2024. City redevelopment documents also contemplated $212.5 million for the stadium project and $130 million for Gas Plant infrastructure. The development proposal included housing, jobs and community-benefit commitments.

People examining that decision could start from very different places. One resident might concentrate on the public cost. Another might study projected economic activity. A business owner might focus on downtown traffic and customers. Someone connected to the Gas Plant community might give much greater weight to restorative development.

A useful civic AI system could stage those perspectives together, attach evidence to each and show where information is incomplete.

The user could then direct the exploration.

Someone who favors public investment could ask the system to explain the strongest arguments against it. Someone skeptical of a stadium subsidy could ask what supporters see that they may be discounting. A resident preparing for a City Council meeting could ask, "Help me explain my position to someone whose priority is economic development."

That last use moves beyond neutral briefing into persuasion, but it still benefits from understanding the other person's evidence and priorities.

The EJAI research also shows why the design requires care. Creating multiple AI personas does not automatically produce balance. If each persona selectively presents evidence supporting its assigned position, we have simply automated the information silos we already have.

A civic system needs a common evidence layer underneath the perspectives, visible sources and mechanisms that surface significant evidence a particular viewpoint leaves out.

This is one of the capabilities we envision for AICOE's Civic Stack and Cityverse in St. Petersburg. Local civic information already exists across government records, journalism, organizations and public discussion. AI can help organize that information around the decisions residents are actually trying to understand.

A Practical Model for Multi-Perspective Civic AI

Build the shared evidence base.
Gather authoritative records, credible research and stakeholder claims while preserving their sources.

Separate evidence from priorities.
Identify which disagreements concern facts and which result from people assigning different weight to the same facts.

Stage the perspectives together.
Let users see how different stakeholders interpret the issue and what each perspective emphasizes.

Let the user drive deeper.
Allow residents to challenge their own position, examine another perspective or learn how to communicate across different priorities.

The value of civic AI may come partly from giving people a wider field of vision. Public decisions rarely arrive with one fact that settles everything. They involve evidence, tradeoffs and people who assign different importance to each.

AI can help put more of that decision on the table at the same time.

‍