Question Period Note: AI AND AUTOMATION FOR IMMIGRATION DECISIONS

Concerns about IRCC’s use of automation and artificial intelligence in decision-making on client applications

• IRCC uses advanced analytics (AA) and other automated systems to help officers identify routine applications for streamlined processing, as well as to perform other functions, such as the sorting of applications based on common characteristics.
• These systems do not automatically learn or adjust on their own, and are not used to refuse any applications or deny entry to Canada.
• IRCC officers remain central to immigration processing and continue to exercise their delegated authority in decision-making.
If pressed
• The use of advanced analytics (AA) enables IRCC to automate some processing steps for routine applications. By leveraging technology, IRCC is able to direct officer resources toward more complex or sensitive applications, and increase the efficiency of our processing.
• IRCC follows TBS directives and conducts Algorithmic Impact Assessments (AIA) for all automated systems that play a role in administrative decision-making, whether these systems use AI or not, to promote transparency. The AIA is GC’s oversight process intended to determine risk and reduce potential negative impacts of automated systems.

IRCC use of automated systems
• IRCC is using a number of innovative approaches to manage high application volumes, improve service delivery and enhance the client experience.
• In support of our clients, we are using digital tools to create processing efficiencies where appropriate. IRCC’s approach to automated systems, including those developed and leveraging advanced analytics and AI is, by and large, facilitative in nature. Models are put in place to streamline processing and approve straightforward cases, never to automatically refuse applicants. All models go through a rigorous review process before implementation, to ensure they are equitable, explainable, privacy protecting and technically sound.
• IRCC’s use of automated systems can be divided into the following broad categories:
o Automating positive eligibility determinations
o Distributing applications between officers based on available capacity or the characteristics of the application (e.g. one that requires a decision-maker with local knowledge)
o Identifying applications that may require additional verification
o Creating ‘annotations’ that summarize basic information on each client to reduce officer searches in the Global Case Management System
o Triaging client emails to enable faster replies, and responding to client enquiries by providing publicly available information
o Assessing biometrics

• None of IRCC’s automated systems, including those that have been developed with AA and AI , can refuse an application, nor can they recommend a refusal to an officer. All final decisions to refuse applications are made by officers after thorough review. Officers are provided with training on IRCC’s automated decision support tools, in order to ensure they understand that a lack of an automated approval does not constitute a recommendation to refuse an application.
• IRCC does not allow the use of generative AI tools, such as ChatGPT, in support of decision-making on client applications. The Department is exploring potential benefits to using generative AI in certain other capacities, such as synthesizing information in support of research and policy development. IRCC is approaching generative AI usage with an emphasis on caution, in line with guidance for public servants provided by Treasury Board Secretariat.
Key projects
• In 2018, IRCC began using advanced analytics) to help officers triage online Temporary Resident Visa (TRV) applications from China and India. In January 2022, the China and India models were updated, and a third model was introduced for TRV applications from all other countries. All three of these advanced analytics models function in the same way: they sort incoming files to streamline officer review, and they automatically approve the eligibility portion of certain straightforward applications.
• In spring 2021, IRCC launched another advanced analytics project for in-Canada Family Class spousal and common-law applications. Similar to the TRV models, this project aimed to automate some positive eligibility determinations and enable the strategic allocation of resources so that all applications can benefit from this initiative. All applications that do not receive an automated eligibility approval by the model are sent for an individualized assessment by officers in accordance with standard practice.. In May 2024, IRCC expanded its use of advanced analytics to all spousal and partner applications under the family class. Two new tools were launched to help process overseas family class spousal and partner applications to reunite families sooner by automating some positive eligibility determinations.
• In 2023, the Integrity Trends Analysis Tool (ITAT), was launched as part of broader departmental efforts to become a more data-driven organization – one that makes evidence-based decisions with efficiency, consistency and a commitment to program integrity. The goal was to leverage proven technology to optimize globalized risk management in the ever-growing inventory of temporary resident applications.

• ITAT uses AI (ML/AA) to uncover risk patterns, which the Department then uses to identify incoming applications that may require further verification. ITAT’s outputs are not shared directly with officers making decisions on individual applications, and ITAT does not automatically put “red flags” on certain clients’ files. Instead, ITAT looks for cases where our officers have confirmed that clients have previously attempted to misuse our programs – for example, by verifying supporting documents and determining that they are fraudulent.

• A two-step, “human-in-the-loop” process helps to guard against the risk that officers might refuse an application simply because it was flagged by an automated system. ITAT only reviews data in client applications. It does not automate or replace human decision-making, and it is not used for immigration and criminal investigations.

• More recently, in October 2022, IRCC began using advanced analytics to help identify low-complexity in-Canada temporary resident applications (Visitor Records) for positive eligibility determination. The model also optimizes processing by triaging the remainder of cases into homogeneous groups to optimize file management and work sharing.
• In late 2023, IRCC launched an automated decision support tool for International Experience Canada work permit applications. In addition to the posting of an Algorithmic Impact Assessment, the department ramped up transparency efforts by also posting a peer review and a GBA Plus assessment for the model. IRCC will continue to work to post additional assessments along with Algorithmic Impact Assessments for our automated decision support tools to provide a larger window into how these tools are developed and utilized.
• Media coverage has incorrectly suggested that IRCC’s ‘Chinook’ tool employs AI or AA to automate decisions. Chinook is a processing aid that extracts basic information from client applications and displays it in a clear format that is more user-friendly for officers. Chinook is not powered by AI or AA, and does not make or recommend decisions on applications.
Ensuring responsible use of data-driven technologies
• The department has developed detailed guidance, including a Policy Playbook on Automated Support for Decision-making, to help consider how these technologies can be used responsibly, effectively and efficiently. IRCC has also established an internal governance framework to ensure that new decision support tools go through a rigorous review and approval process. A broad-based committee of senior executives acts as the key oversight body in this governance framework.
• IRCC is always working to be a leader in the responsible use of data-driven technologies, and has developed its approach to align with the Treasury Board Directive on Automated Decision-Making, as well as other key legal and privacy requirements. To date, IRCC has published more Algorithmic Impact Assessments on Canada’s Open Government Portal than any other federal department or agency.
• Protecting individuals’ personal information continues to be a priority for IRCC as the Department tests new and innovative approaches. The use of personal information for analytics-based processing is in accordance with the Immigration and Refugee Protection Act and the Privacy Act. Its use is consistent with the purpose for which it was initially collected.

None