A whirlwind month discussing youth, AI, and mental health!
Originally posted on Linkedin:
June 2026 was a whirlwind month discussing youth wellbeing and responsible AI across ACM UMAP, #DPH2026, and ACM FAccT. Across all these events, one challenge has stayed with me:
Adaptive AI is often framed as improving when systems know more, respond faster, and personalize more precisely. But for (youth) wellbeing, adaptation may only become meaningful when young people can shape how a system understands, supports, and responds to them. Three tensions I observed:
1. ๐๐ฌ๐ค๐ข๐ง๐ ๐๐๐๐จ๐ซ๐ ๐ข๐ง๐๐๐ซ๐ซ๐ข๐ง๐
At ACM UMAP, I was grateful to present our joint work on youth-created personas for reflexive co-design of LLM personalization requirements. I enjoyedSubigya Nepalโs talk on momentarily salient adaptive systems, which also connected closely with my poster discussion about whether systems should infer a userโs context, or ask for it.
Thoughtful questions helped me understand that personalization is never only about relevance. It is also about ๐ณ๐ฆ๐ฑ๐ณ๐ฆ๐ด๐ฆ๐ฏ๐ต๐ข๐ต๐ช๐ฐ๐ฏ: what picture the system builds of a young person, and how much space they have to revise it.
2. ๐๐ก๐๐ง ๐๐จ-๐๐๐ฌ๐ข๐ ๐ง ๐ฌ๐ญ๐๐ซ๐ญ๐ฌ
At #DPH2026, I was honoured to join Digital Transformations for Health Lab's panel on youth digital public health with Tobias Silberzahn, Rossana Alessandrello, and Jordi Piera Jimรฉnez.
A question that kept coming up was: ๐ธ๐ฉ๐ฆ๐ฏ ๐ฅ๐ฐ๐ฆ๐ด ๐ค๐ฐ-๐ฅ๐ฆ๐ด๐ช๐จ๐ฏ ๐ข๐ค๐ต๐ถ๐ข๐ญ๐ญ๐บ ๐ฃ๐ฆ๐จ๐ช๐ฏ? Too often, young people are invited in after the system logic has already been decided: what data matters and what responses are available. What does it look like if we include youth as technical stakeholders, not just end-users?
3. ๐๐ก๐๐ญ ๐๐จ๐ฎ๐ง๐ญ๐ฌ ๐๐ฌ ๐ ๐ฌ๐๐๐ ๐ฉ๐๐ญ๐ก๐ฐ๐๐ฒ?
At #FAccT2026, I presented our participatory approach to operationalizing governance for youth-facing LLMs.
Escalation generated especially thoughtful discussion. Parents and social workers often emphasized boundaries, authority limits, and referral to human support. Young people proposed intermediary strategies, including retrieval-based ways to surface peer support stories before formal help. For youth-facing AI, safety may not always mean removing steps between a young person and formal help. Perhaps ๐ต๐ฉ๐ฆ ๐ช๐ฏ๐ต๐ฆ๐ณ๐ฎ๐ฆ๐ฅ๐ช๐ข๐ต๐ฆ ๐ด๐ต๐ฆ๐ฑ๐ด ๐ฎ๐ข๐บ ๐ฃ๐ฆ ๐ธ๐ฉ๐ข๐ต ๐ฎ๐ข๐ฌ๐ฆ ๐ฉ๐ฆ๐ญ๐ฑ ๐ฑ๐ฐ๐ด๐ด๐ช๐ฃ๐ญ๐ฆ.
๐๐ก๐๐ญ ๐ซ๐๐ฆ๐๐ข๐ง๐ฌ ๐ฎ๐ง๐ซ๐๐ฌ๐จ๐ฅ๐ฏ๐๐
I'm very grateful to my co-authors for trusting me to represent our joint work, and to the conference communities for such thoughtful conversations!
Iโm super curious to hear othersโ perspectives: ๐๐ฐ๐ธ ๐ค๐ข๐ฏ ๐ฑ๐ข๐ณ๐ต๐ช๐ค๐ช๐ฑ๐ข๐ต๐ช๐ฐ๐ฏ ๐ฃ๐ฆ๐ค๐ฐ๐ฎ๐ฆ ๐ด๐ฑ๐ฆ๐ค๐ช๐ง๐ช๐ค ๐ฆ๐ฏ๐ฐ๐ถ๐จ๐ฉ ๐ต๐ฐ ๐ฎ๐ข๐ต๐ต๐ฆ๐ณ, not only as a broad principle, but in how responsibility, trust, autonomy, and care are distributed between young people, adults, institutions, and AI systems?
