We ensure that improvements in AI model performance are aligned with improvements in human outcomes.
AI systems are rapidly reshaping how humans learn, think, and connect—yet emerging evidence of AI-induced social, emotional, and cognitive stunting in users signals that AI systems are not serving human developmental needs. How might such systems reinforce rather than rust long-term human flourishing?
We first synthesize interdisciplinary research across neuroscience, developmental psychology, and the learning sciences to identify and validate metrics for positive developmental outcomes—attachment security, growth orientation, resilience, sense of belonging. We then engage frontier labs and education systems in integrating these metrics directly into model evaluation pipelines, creating a shared language between AI alignment research and social institutions.
The goal: that no major AI system—especially those impacting children or learners—is deployed without being evaluated through a developmental lens, the way we now expect environmental impact assessments for infrastructure.
Summer 2026. Presenting the human-AI developmental sciences framework in talks around Europe (Max Planck Institute for Human Development in Berlin, Foresight Institute in Berlin and the UK). The framework synthesizes seminal traditions and contemporary literature across the learning and developmental sciences and AI—Dewey, Piaget, Turkle, Turing, Rumelhart—to identify short- and long-term avenues of productive research. By midsummer, workshopping the framework-to-translation pipeline with a network of practitioners (Erin Mote, Sunanna Chand), researchers (Carol Dweck, Eric Hanushek), and engineers (Joel Lehman, Nick Haber). The final draft identifies key developments, risks, and opportunities—from reliance on AI for social interaction, to the role of AI in shaping identity and motivation, to cognitive debt—with a manuscript submitted to Nature Perspectives by the end of summer.
Fall–Winter 2026. Focusing the research agenda on the developmental topic of transformative aspiration, while convening those who aim to do their life's work in human-AI developmental sciences: weekly "demo dinners" where developmental practitioners critique prototypes; biweekly salons that workshop emerging ideas; monthly hackathons on open problems. Each event is followed by a blogpost to build thought leadership in the space.
Spring 2027. With a core team of researchers and developers recruited, building open-source tools and materials for public release—engaging a broader audience in the design, iteration, and implementation of our tools, metrics, and methods. An invite-only summit reviews shorter-term experiments and sets a long-term R&D agenda with global experts, funders, and community members.
Summer 2027 and beyond. Launching a series of empirical human-AI developmental science studies—especially longitudinal and cross-cultural work—translating findings into concrete design principles and policy recommendations, and training facilitators to run reading groups, bootcamps, and convenings that recruit more people into the field.
Last month, I asked a group of education professionals at The Tech Interactive's National AI Literacy Day if they'd heard of Maslow's Hierarchy of Needs. Every hand in the room went up. Yet when I introduced a newer developmental framework—one that substantially challenged Maslow's—no one recognized it.
The next five years are the window for massively updating our human development systems to align with research-backed and field-sourced insights on human capabilities, as elevated by AI. This is not where technology and society are currently headed. AI and social media present sycophantic and misinformed ideas on what it means to live a good life; practitioners and institutions remain fixated on developmental ideas from the 1940s; and young people today are experiencing fewer adult-like experiences before graduating high school. When our environments, media, and education systems are not aligned with developmental science, we handicap our ability to expand human capability and economic productivity—and we harm children's abilities to grow into the best possible versions of themselves.
My goal is to transform legacy education systems into lifelong learning ecosystems that support population flourishing—designing and deploying human-AI systems that don't just optimize our first-order desires (revealed preferences) but discover and promote our second-order desires (aspirational preferences). I designed Future You to give everyone the opportunity to experience a digital environment that normalized and incentivized growth; my research on "downward spiral" human-AI interactions has prepared me to study "upward spiral" interactions, achievable through aspirational alignment of AI systems.
This is why, more than just launching my own research, I would like to launch the field of human-AI developmental sciences. The cognitive revolution sowed the seeds for today's intelligent technologies. I believe the 21st century will be the developmental revolution—seeded by a better understanding of human capabilities and how they can be cultivated, through systems that scaffold how people mature and change over time.
Peggy Yin is a Stanford psychology PhD student and Knight-Hennessy Scholar working on building AI systems for cognitive and behavioral change. She previously led projects at Teach for America, Ars Electronica, and the Stanford Institute.
I'd like to play a part in ensuring that AI model performance is aligned with human outcomes.