Page 22 - U4SSC Guiding principles for artificial intelligence in cities

• Ensuring AI systems get trained, tested, and evaluated in lifelike environments which accurately
represent actual operating conditions in real life (including realistic datasets and algorithms);
• Trading off optimality and robustness, where needed, to cautiously make decisions; and

• Develop governance or advisory committees to evaluate these trade-offs and assign
accountability for their deployment.

3.2.8 Assessed for impact and sustainability

AI systems are designed to handle certain tasks and achieve a certain level of performance as
discussed in the previous principle. However, it is important to take a broad perspective and assess
the impact of AI systems in cities, i.e. to identify the future consequences of using these systems.

Impact assessment can be used as a forward-looking tool and considers the positive and adverse
impacts during such an assessment. Impact assessment helps in balancing current and future
sustainability with respect to AI systems. It is an invaluable tool for cities and their stakeholders by
providing a well-informed and more holistic perspective of their AI systems.

Consequently, this principle allows cities to assess the impact of AI systems and to evaluate and ensure
their sustainability. Recommendation ITU-T L.1480 “Enabling the Net Zero transition: Assessing how
the use of information and communication technology solutions impact greenhouse gas emissions
of other sectors” give guidance on how to assess the positive effect of AI implementation.

Implementation Considerations: Cities can adopt various mechanisms to assess the impact of
their AI systems and to evaluate their sustainability. These mechanisms include:

• Conducting well-defined impact assessments for AI systems (e.g., Recommendation ITU-T Y.4905
Smart sustainable city impact assessment and Recommendation ITU-T L.1410 Methodology for
environmental life cycle assessments of information and communication technology goods,
networks and services);

• Taking a broad social, economic, and environmental perspective during the impact assessment
(e.g., social issues would include communal and individual aspects such as health and well-
being, fears and aspirations, and cultural heritage; economic issues would include GDP, business
output, employment and wages; environmental issues would include energy, GHG emissions,
etc.);
• Defining the geographical scope (e.g., community, city, nation) and the time horizon clearly for
the AI systems impact assessment (would help in quantitative and qualitative aspects);

• Determining the details of research and analyses for AI systems’ impact assessment (e.g.,
stakeholders involved, interviews, surveys, economic and environmental models);

• Taking a holistic approach for AI systems’ long-term sustainability from different perspectives
and evaluating the costs, benefits and the risks; and

12