An Effective AI Strategy for Canadian Youth Must Address Age-Related Risks

The federal government recently announced two policies that will shape the future of Canadian youth’s interactions with artificial intelligence: the AI for All strategy and the Safe Social Media Act.

The government says the AI for All strategy will give “access to AI training and education for all Canadians” to address Canada’s “AI adoption gap,” arguing that “adoption will drive AI’s benefits for Canadians.”

The Safe Social Media Act proposes establishing a regulator to protect youth from egregious harms posed by AI chatbots, such as “reinforcing harmful behaviours and providing unsafe responses.” It aims to maximize the safe adoption of AI by Canadian youth.

However, research on generative AI (genAI) use among youth pinpoints design features that pose age-specific risks, which must be addressed before we adopt genAI in our schools and homes.

The AI for All Strategy includes free AI literacy training via online courses, K-12 educator training, making AI chatbots available to all post-secondary students, job opportunities via funded internships and safety-focused governance of these systems.

The Safe Social Media Act introduces a framework to improve online safety. It proposes regulating AI chatbots by requiring companies to protect privacy and not promote harmful content (cyberbullying, violent, sexual or hateful content) or harmful behaviours (impersonation/deepfakes, manipulative engagement techniques or encouraging self-harm).

Risky design features of genAI

I recently contributed as a scientific adviser to two research reports that identify risky design features of genAI.

The first is a mapping of genAI’s impact on child development that was presented by a coalition of research groups as part of the G7 summit in France.

In the report, collaborators and I argue that AI must:

• Be regulated through a developmental lens, ensuring it is age-appropriate;

• Support rather than replace young people’s skills;

• Be evaluated through youth-centred standards, evidence and safeguards.

Importantly, youth must not be treated as a single entity or user category, as is done in the current regulations.

AI will affect children, adolescents and young adults differently, including cognitive, emotional, social and educational elements of their lives.

Given this, our research team identified the opportunities and associated risks of genAI for children and youth from birth to age 18, with policy recommendations.

From birth to age 3: GenAI should strengthen caregiver responsiveness and not replace it, helping caregivers notice and respond to infants and young children. Risks arise when genAI is designed to be a source of responsiveness or stimulation for infants. Policy must recognize that genAI systems designed to mediate infant-caregiver interactions must be treated with more caution than ones designed as caregiver-support tools.

Ages 3-6: GenAI must remain a tool within adult-guided interaction and not a stand-alone source of explanation, entertainment or social interaction. The features that drive generative AI engagement — fluent, personalized and emotionally resonant language — risk making these interactions feel authoritative and relationally authentic to children.

Read more:
AI products for kids promising friendship and learning? 3 things to consider

Ages 6-12: GenAI must support learning and social development and not offload learning and outsource friendship-building. The design features of personalized learning, automatic feedback and social mediation risk reducing cognitive effort, weakening self-regulation and displacing human interaction. Policy must ensure the design features of AI strengthen and don’t weaken the skills that underpin children’s educational and social foundations.

Ages 12-18: GenAI must support the building of autonomy, identity and judgment so adolescents become self-directed individuals. Policy must ensure design choices and default features of AI do not weaken independent judgment, intensify reliance and set improper expectations for healthy social relationships.

Children & Screens research institute

The second report is a review of research on AI in education published by Children & Screens, an independent research institute that conducts and shares interdisciplinary studies on the impact of digital media on children.

The report covers how AI is used in education, the risks posed for each use and how educators and families can evaluate AI for learning. It concludes that genAI must be carefully integrated into learning by evaluating the opportunities and risks posed by each use, as well as risks arising from genAI being a tech company product.

GenAI as a tool should help students complete specific learning tasks efficiently, such as organizing notes, formatting materials and making learning products (for example, editing a video report to remove transitions between sections) so students can focus on understanding and thinking about the learning content. Risks arise when the tool is designed to do the thinking for the student (such as editing, paraphrasing or brainstorming). Policies must ensure GenAI tools deployed in schools build skills and do not replace them.

GenAI as a tutor should provide feedback, guided practice and explanations to help students move through a process of productive struggle. However, serious risks arise due to genAI hallucinations and the burden they place on students to correctly prompt the systems for guidance. Students are ill-positioned to judge if an AI tutor is correct or hallucinating. Students also do not know how to structure a lesson; they’re not teachers. Policies must ensure AI tutors are designed to address hallucinations and do not force students to become their own teachers.

GenAI as a resource should expand students’ access to explanations, examples and translations of learning content and practice materials while ensuring this content is both accurate and reflective of the students’ cultural context. Here, the risks arise in how the genAI is trained, on what information (internet versus trusted sources) and how it is designed to present information to students (linking to verifiable sources versus not).

Policies must ensure that genAI systems be accurate and legally and ethically trained on Canadian content (for example: partnering with Indigenous communities to include Indigenous history and language) and present verifiable information. They must also ensure AI literacy training includes awareness of AI hallucinations and bias.

GenAI as a tech company product should give students access to effective and safe genAI that is independently tested and proven to aid learning. Risks from tech companies include the “deploy now, test later” approach that puts unproven genAI into students’ hands; business practices that turn private student data (for example, related to age, backgrounds, grades, assignments, preferences, what they think, how they learn) into corporate assets to be monetized; and design features that maximize engagement (using friendly, human-like language to foster para-social relationships with genAI) instead of learning.

Developmentally responsible ‘AI for All’

Policies should go beyond getting youth AI-ready. They must regulate genAI systems and interactions to make them youth-ready.

This will require following evidence, developing critical AI literacy initiatives and drafting regulations that ensure AI is designed to maximize academic and social skills, not tech company profits.