Growing AI like a Child, at Scale, Safely

Humans never "learn" intelligence. Humans develop intelligence. Biological lives on this planet take heavy advantage of intelligent primitives embedded in their genes. Cats never "learn" to backflip. Birds never "learn" to fly. In the same way, humans never "learn" to cognize. Humans are born with a set of core cognition, that sets the foundation for our perception and action in the physical world.

Our core cognition unravels through a specific developmental trajectory as we grow into adulthood. Here, we seek to do the same for our machines, leveraging heavy cognitive literature in developmental psychology, e.g., Piagetian theory of cognitive development, to design our growing up curriculum. In addition, we also want to learn from the current success of machine intelligence, specifically the scaling law which has repeatedly led to the development of highly competent AI systems.

However, competence is not the only thing that matters. We also seek a deep sense of human sensibility. More than just being tremendously powerful, it should engage with us in ways that reflect an understanding of the richness yet fragility of human existence—offering comfort in sorrow, sharing in moments of happiness, and standing up with courage when harm is near. It must also understand, care, and connect with us in meaningful and safe ways.

Instead of putting growing up and scaling up, competence and safety into opposite camps, we argue the next step towards human-like artificial general safe intelligence is to grow AI like a child, at scale, safely. We come together as GrowAI, an open-source community uniting researchers from computer science, cognitive science, psychology, linguistics, philosophy, and beyond. Our ongoing research focuses on the following areas:

• Cognitive Competence: Investigating and evaluating the cognitive behaviors and limitations of pre-trained models beyond leaderboard chasing.
• Core Knowledge: Identifying and building fundamental knowledge and core cognitive scenarios for benchmarking and evaluating human-like intelligence.
• Developmental AI: Understanding the developmental trajectories and training dynamics of scalable systems toward human capabilities.
• Cognitive AI for AI Safety: Essential AI alignment research from cognitive science perspectives to ensure the reduction of human existential risks.