“Accountable AI” analyzes AI and automated decision-making (ADM) systems used to assist government decision-making and considers legal accountability issues when civil and administrative justice bodies use these systems in decision-making. The latest report forms part of a series of LCO-issued papers examining AI use in the Canadian justice system.
Based on the report, the LCO said that AI systems offer significant potential benefits to governments and the public, such that “AI might finally help crack the code of mass adjudication, improve accuracy, reduce inconsistency, and cut down on rampant backlogs that plague government agencies.”
The LCO added that many believe that government AI systems have the potential “to reduce discrimination, enhance democratic and legal accountability, and significantly improve the administrative state.”
The LCO noted that governments worldwide use AI across a broad range of areas, services, and functions. Some notable uses of AI in government to date include child risk assessment tools, immigration detention tools, biometric surveillance systems, predictive policing systems, bail and sentencing algorithms, and tax compliance algorithms.
However, the LCO said that despite AI’s potential, the government’s use of AI has remained “controversial,” There are many examples of government AI systems that have proven to be “biased, illegal, secretive, or ineffective.”
“Experience with government AI systems across North America, Europe, Australia, and New Zealand demonstrates the serious risk of bias, data discrimination, lack of due process, and ‘black box’ decision-making,” the LCO wrote.
The LCO also found that there are AI systems than can diminish judicial and administrative discretion, establish automation bias, and disrupt the legitimacy of government decision-making.
“The burden of these harms and risks is not shared equally, as they fall disproportionately on racialized or otherwise vulnerable communities,” the LCO wrote.
To prevent these harms and risks, the LCO said many governments, including Ontario, have adopted “Trustworthy AI” frameworks. These frameworks assure the public and stakeholders that government AI development and use will be “transparent, legal and beneficial.”
“[We] believe that widespread adoption of AI systems has the potential to transform the legal and policy landscape of government decision-making,” the LCO wrote. “As a result, government AI accountability strategies must be designed to respond to the unique features of AI decision-making.”
The LCO concluded that “accountable AI” depends on “a mix of law reform tools and strategies,” such as front-end regulation, substantive law reform, enhanced due process protections, and innovative initiatives to improve access to justice.
The LCO also made 19 recommendations to address “bias” in the government AI systems, particularly in Ontario.
To ensure that government AI is appropriately regulated, the LCO suggested that Ontario incorporate its “Trustworthy AI” framework in legislation, including provisions that promote access to justice, address bias and discrimination, and mitigate harms.
To guarantee that government AI systems comply with human rights and administrative law requirements, the LCO advised that the provincial government, the Ontario Human Rights Commission, technologists, human rights experts, and community members should work altogether to develop a provincial human rights strategy for Ontario AI systems.
The LCO added that the provincial government should develop and adopt an AI directive to guide provincial decision-making and serve as a template for other public organizations under the provincial jurisdiction.
The LCO recommended that Ontario’s laws of evidence be monitored to gauge whether the current law evaluating expert evidence is sufficient and effective when applied to AI litigation. Moreover, new AI-specific civil procedure rules should be considered.