The aim is to develop new, breakthrough AI approaches that go beyond current generative AI and LLMs, with a focus on three cognitive capabilities:

• Deep Reasoning – causal, logical and probabilistic reasoning, handling uncertainty and complex knowledge.
• Deep Abstraction – learning higher-level concepts, analogies and transferable representations from limited data.
• Deep Planning – long-term, hierarchical and adaptive planning in complex and uncertain environments.

The emphasis is on trustworthy, explainable, human-centric AI, rather than simply developing a larger or better language model.

What kind of innovation is expected?

The call encourages radically new approaches, potentially including:

• neuro-symbolic AI,
• new cognitive architectures,
• formal methods,
• brain-/biology-inspired approaches,
• novel learning and reasoning mechanisms,
• combinations of these approaches.

Projects should demonstrate their technology in a working cognitive AI system at around TRL 4.

Important additional expectations

Projects should address:

• trustworthiness, explainability, safety and security;
• new ways of evaluating reasoning and AI reliability;
• efficient use of computational resources;
• FAIR data and models;
• relevant real-world applications such as scientific discovery, healthcare, robotics, industry, mobility or cybersecurity.

An important feature is that projects will be selected as a portfolio. The Commission wants complementary projects covering different approaches and applications, rather than several projects developing essentially the same technology.