Generative visual AI makes civic design conversations clearer and faster but doesn’t usurp planners: it acts as an iterative co‑design tool that lowers barriers to expression, yet its influence on actual projects materialises only where institutions can convert participatory inputs into decisions.

Main Finding

Generative visual AI (UrbanistAI) used inside official urban participation processes improved communication, iteration, and co-design during workshops—acting as a low‑threshold expressive and deliberative tool that augments (but does not replace) professional expertise. Its influence on built outcomes depends on institutional feedback loops that can act on participatory inputs. The study frames this as a "designed distribution of agency": machines admitted co‑agency in imaginative phases while human actors retained accountable judgement in decision phases.

Key Points

• Distinction introduced:
  • Participation with AI — using AI tools as part of participatory activities (visualisation, iteration).
  • Participation in AI — public involvement in governance of algorithmic systems (not the main focus here).
• Immediate, iterable visualisations:
  • Clarified abstract ideas and made proposals tangible.
  • Strengthened iterative co‑design and joint evaluation during workshops.
  • Lowered barriers to expression for lay participants.
• Professional expertise and procedures:
  • AI did not replace planners’ expertise or formal procedures.
  • The tool functioned as a complement (a design aid), not a substitute.
• Impact on actual built outcomes:
  • Dependent on whether participatory outputs entered an institutional feedback loop with authority and capacity to act.
  • Without such links, richer participation risked having limited real-world effect.
• Theoretical contribution:
  • "Designed distribution of agency" specifies how a tool structures human–machine roles: machine co‑agency in imagination; human accountability in final design and implementation.

Data & Methods

• Empirical base: five early adoption cases (2023–2024) where UrbanistAI was integrated into official participation processes.
• Geographical contexts: Finland, Kosovo, United Arab Emirates.
• Evidence: eight semi‑structured interviews with planners and consultants; qualitative content analysis of the cases and workshop artifacts.
• Methodological scope & limits:
  • Qualitative, small‑N, early‑adopter focus—rich descriptive insights but limited generalisability and no quantitative outcome measurement.
  • Findings reflect practitioner experience and observed process effects rather than measured impacts on final built environment outcomes.

Implications for AI Economics

• Productivity and task reallocation
  • Complementarity: Visual generative AI complements skilled planners rather than substituting them, shifting time and effort from routine visualisation work toward higher‑level judgement and implementation tasks.
  • Potential time/cost savings in producing visual alternatives and running more iterative workshops; quantification of these savings is an important next step for economic assessment.
• Labour market effects
  • Demand shift toward roles combining domain expertise with AI‑mediated facilitation and evaluation skills (e.g., planners who can curate prompts, interpret outputs, and integrate feedback).
  • Lower‑threshold tools may broaden participation but do not eliminate need for professional accountability—affecting skill‑premium dynamics within planning professions.
• Public‑sector procurement and governance incentives
  • Realisation of value from participatory AI depends on institutional capacity: where feedback loops exist, benefits can translate into better outcomes; where they do not, participation may generate political or reputational but not material returns.
  • Incentives to adopt these tools will therefore track not only cost and efficiency gains but also institutional design (who can act on inputs, budgetary rules, legal responsibilities).
• Market design and externalities
  • Positive externalities: better informed public inputs could improve allocative efficiency of public investments if institutions act on them.
  • Risks of wasted effort or legitimacy loss where participatory outputs have no channel to influence decisions—this represents a governance externality worth internalising through procurement rules or mandated feedback mechanisms.
• Accountability, liability, and regulation
  • The "designed distribution of agency" implies normative choices about where accountability sits; policymakers and procurement frameworks should explicitly allocate responsibilities (e.g., who vets AI outputs, who signs off on designs).
  • Regulatory standards and transparency requirements (audit trails of prompts/outputs, human sign‑off) will affect adoption costs and firms’ liability exposure.
• Research and evaluation priorities for AI economics
  • Quantify: time/cost savings per project, changes in participation rates, and effects on decision quality and implementation outcomes.
  • Comparative studies: cases with vs. without institutional feedback loops to estimate causal effect on built outcomes.
  • Welfare analysis: weigh gains from improved deliberation against costs of adoption, governance failures, and possible reputational harms.
  • Market impact: study how such tools affect prices for planning services, competition among consultancies, and entry of tech providers into civic procurement.

Suggestions for policymakers and researchers: measure and incentivise institutional pathways that convert richer participatory inputs into actionable decisions; build procurement standards that define accountability for AI‑generated content; and conduct quantitative cost–benefit studies to assess economic impacts of these tools in public planning.