Border agency expands use of tool to identify ‘higher-risk travellers’

Cars pass a monument marking the border between the United States and Canada on Wednesday, Sept. 1, 2021, at Highgate Springs, Vt. (AP Photo/Wilson Ring)

The federal government plans to roll out an artificial intelligence tool at all its land borders to help determine which travellers might need a secondary examination before entering the country, CTV News has confirmed.

According to the Canada Border Services Agency (CBSA), the traveller compliance indicator (TCI) was developed in-house to “process travellers more efficiently.”

The tool compiles predictive data that is “already available in multiple systems” in real time and flags travellers who are at a higher risk of violating border requirements. It’s meant to help border agents decide whether certain travellers should be referred to a secondary inspection.

“The goal of the Traveller Compliance Indicator is to give officers a real-time summary of available data to help them identify and process compliant travellers faster and focus on unknown and higher-risk travellers,” wrote CBSA spokesperson Luke Reimer in an email to CTV News.

The TCI is part of a broader modernization plan by the agency, which has been ongoing since last August.

“The Traveller Compliance Indicator is just that, an indicator, and does not replace officer judgement or automatically determine outcomes,” Reimer wrote.

“The actual decision on whether to refer a traveller for a secondary examination rests with the border services officer, whose specialized training, expertise, and knowledge allow them to always be on the lookout for potential threats,” Reimer added.

The expansion of CBSA’s use of the TCI was first reported by the Toronto Star.

The TCI has been in place as a pilot project at six land ports of entry since 2023, with full roll-out planned by late 2027. Reimer did not lay out a timeline for the technology to be implemented at air or marine ports of entry.

“We expect that the Traveller Compliance Indicator will help to reduce referrals for secondary examination that do not result in the discovery of traveller non-compliance,” Reimer wrote.

Statistics provided by Reimer on the number of travellers who received secondary examinations has been largely steady or on the decline in the last year.

According to University of Toronto professor Ebrahim Bagheri, however, any use of AI that impacts human behaviour and decisions is likely to result in inherent biases.

Bagheri — who specializes in the responsible development of AI — pointed to instances of artificial intelligence designed to assess risk resulting in minority groups being negatively profiled.

“The risk here is that similar types of biases get picked up,” Bagheri said of the TCI tool, in an interview with CTV News.

He said AI, because of the way it’s developed and trained, has the potential to “pick up on biases in historical data, exacerbate them, and put them into action.”

Another potential cause for concern, Bagheri said, is automation bias — the phenomenon that leads most people to favour suggestions by automated systems, instead of relying on their own knowledge.

Bagheri said that in the case of the TCI, while the CBSA’s intention appears to be leaving any final decision about which travellers need secondary examination to human border agents, automation bias means those agents are likely to defer decision-making to the algorithm.

The U of T professor said anyone using this type of system would need extensive training to understand the reliability of it, but also that inherent biases and automation bias are “bound to happen.”

He said artificial intelligence technology could be deployed to help border agents in other ways, instead of specifically being designed to build risk scores for travellers.

Reimer said the CBSA is “actively” working to minimize the impact of potential biases of the TCI, adding the agency “has already taken several important steps” to that end.

Those include plans to monitor the tool’s performance across different equity groups and “continuously” work to mitigate biases.

“The main issue with this is that typically the people or groups who build a system, they’ve already done their best to avoid any biases,” Bagheri said. “The only way you can make a system better is if you allow independent scrutiny of the system.”

“It becomes really hard to see how the CBSA can effectively ensure that their system is gradually looking at and monitoring biases and making it better,” he added.

The TCI came in slightly under budget, Reimer said, costing the government about $15.3 million, who added that once the technology is fully implemented, the operational costs for ongoing development and maintenance are expected to be about $700,000 annually.