Detroit has formally adopted a policy governing the use of Artificial Intelligence (AI) by its employees, consultants, volunteers, and third-party vendors. The policy, which went into effect over the summer, aims to enhance city services through AI while safeguarding against potential risks to residents.
The policy broadly defines AI as technologies capable of making predictions, recommendations, or decisions based on objectives provided through programming. These systems process machine- and human-based inputs using algorithms and large datasets to generate outputs that mimic human reasoning. However, the policy acknowledges that AI's reliance on extensive human input can lead to the replication of stereotypes, bias, and even the generation of false information, known as "hallucinations."
Currently, city employees are restricted to using Microsoft CoPilot Chat as their only approved text-based generative AI model, according to an internal memo from last month. Other large language models (LLMs) such as ChatGPT, Claude, Gemini, Alexa, or Siri have not received approval for city use. The policy also explicitly prohibits the use of AI to create images or media of real events, or in the evaluation of people or programs, including employee reviews. Generally, any AI system handling sensitive or personally identifying information must undergo human review.
Art Thompson, chief information officer with the city’s Department of Innovation and Technology (DoIT), underscored the city's commitment to data security, stating, "Detroit’s data is Detroit’s data. We own it. We do not farm it out, share it, give it out. It is a very sacred thing." City personnel are required to be transparent about their use of AI and to fact-check all AI-generated outputs before their application.
The policy is built upon nine core city values: Human-Centered Design, Security & Safety, Privacy, Transparency, Equity, Accountability, Effectiveness, Workforce Empowerment, and Compliance with Applicable Laws, including the City’s Community Input Over Government Surveillance (CIOGS) ordinance. These values are designed to ensure AI systems are developed and deployed with public impact in mind, maintaining confidentiality, managing potential bias, and ensuring human oversight.
The policy framework is designed to be flexible, recognizing the rapid evolution of AI technology. City employees can submit requests for new AI technologies or upgrades to existing systems for review by various entities within DoIT, including Thompson himself. Detroit has prior experience with citywide AI, including an AI pilot program named Emily, which previously handled resident calls and messages for the Department of Neighborhoods and has since concluded, performing well according to Thompson and DON Director Bryan Peckinpaugh.
Enforcement of the policy falls under the city’s employee handbook, with general violations potentially leading to discipline, up to and including discharge. Illegal activities are subject to legal action.
The drafting process for this comprehensive policy began in April 2023, involving extensive collaboration across multiple city departments. Input came from the Law Department, DoIT, Human Resources, the City Council’s Legislative Policy Division, Procurement, and both the police and fire departments. Detroit also gained insights from its participation in the GovAI Coalition, an international group dedicated to promoting responsible AI in the public sector.
The policy directly addresses the inherent flaws of AI, including sections dedicated to "AI Risks" such as data breaches and "adverse outcomes" stemming from bias and misinformation. The primary mitigations for these issues are a requirement for human involvement in refining AI outputs and a cautious approach to data handling.
Thompson expressed his biggest concern regarding the varying levels of AI literacy among residents and employees, particularly among older individuals, children, and those with limited access to technological education. He noted the technology's rapid advancement and the challenges this poses for individuals in protecting themselves from threats like deepfakes and advanced cybersecurity issues. While Detroit's AI literacy ecosystem is considered good, Thompson believes there is room for improvement, highlighting programs and events hosted by organizations such as New Lab, Black Tech Saturdays, Rocket, and Connect313.
Residents with concerns or feedback regarding the AI policy are encouraged to participate in City Council meetings, including public comment periods, to ensure their voices are heard by Thompson and his team.