Technologists warn of AI bias, misreads and governance gaps for ALPRs and drones

At the Saint Paul hearing, multiple speakers with technology and data backgrounds cautioned that AI-powered surveillance tools are not neutral—algorithms rely on probabilistic associations and can be biased, they said, and those limitations can produce false identifications and disproportionate impacts on over-policed communities.

Chris Daniel, a technology professional, recommended the council review model transparency frameworks such as ACLU's CCOPS and criticized private vendors for running ahead of regulators. Daniel said the city should "take the time to articulate our intent and understand our limitations under the law." Daniel and others stressed that corporate settlements are often absorbed into the business model and do not meaningfully deter misuse.

Daniel Hartman and other speakers made technical arguments that machine-learning models are nondeterministic and subject to bias based on training data and deployment patterns. Speakers said that concentrating cameras and analytic tools in certain neighborhoods increases the risk of skewed policing and false positives; one witness described an instance he summarized as "1 in 10" Flock plate misreads. Multiple witnesses urged legislative reforms to data law to produce accountability.