The machine does not refuse alone

Since 2019, the Government of Canada has governed its own algorithms through a Treasury Board directive on automated decision-making. It covers decisions that affect people's rights or interests, such as reviewing a benefits application or entry into the country, and applies to systems designed or purchased after April 1, 2020. Its goal: transparent, accountable and fair decisions.

Before launching a system, a department must complete an Algorithmic Impact Assessment: 65 questions on risks and 41 on ways to reduce them. The result places the system on one of four impact levels and must be published on the Open Government Portal. Revised in 2023, the directive strengthens bias testing and the obligation to explain decisions.

A concrete example: since 2018, Immigration, Refugees and Citizenship Canada has used advanced analytics to sort temporary resident visa applications, first those from China and India. The system identifies simple files, whose eligibility can be assessed 87% faster. But officers still make the final decision, and the department says it reviews its tools to detect bias.

The lesson is simple: a machine can help, but it is a human being who must carry the weight of a refusal, and answer for it.