System 2 AI
Artificial intelligence has made rapid progress, yet most systems still learn in fixed phases: trained once, deployed once, and only rarely improving through experience. At System 2 AI, we are building a different kind of intelligence — AI that learns continuously while operating in the real world.
System 2 AI develops AI that can learn on the job from its own experience. Inspired by human cognition, we focus on building systems capable of sustained learning, reflective reasoning, and adaptive decision-making in dynamic environments.
This requires AI to develop a coherent internal understanding of the world — capturing relationships, temporal structure, and causal dependencies rather than isolated facts. Systems that have this can reason more effectively, plan across longer horizons, and update their knowledge in a consistent way as new experiences arrive.
We believe the next generation of AI will be defined not by scale alone, but by the ability to learn continuously, reason coherently, and integrate knowledge through experience — and we are building toward that by combining foundational research with practical systems that operate, learn, and improve in the real world.
Continual learning
Accumulating knowledge without catastrophic forgetting — enabling AI to grow from experience rather than starting fresh.
Improved reasoning and planning
Supporting deliberate, multi-step problem solving that goes beyond pattern matching.
World models and knowledge internalization
Forming coherent representations that capture relations, time, and causality — allowing systems to simulate, predict, and plan.
Active learning
Selecting informative experiences through interaction and exploration — seeking out what matters rather than passively consuming data.
Our work combines large-scale computation with interactive training. We leverage online and reinforcement learning loops that tightly integrate data collection and model updates, enabling systems to improve through experience. This work is supported by access to high-performance infrastructure including the LUMI supercomputer, allowing experimentation at scale.
We are a small, highly collaborative team where roles overlap and everyone contributes across research, engineering, infrastructure, and company building.
Former ML scientist at Apple; co-founder of Zenrobotics and Curious AI
Former professor at Aalto University; ML scientist at Amazon and Curious AI
Former ML scientist at Apple; co-founder of Curious AI
We actively collaborate with multiple research groups at Aalto University.