Expected Outcome:
Project results are expected to contribute to some of the following expected outcomes:
- Development of General Purpose AI (GPAI) models and architectures demonstrating enhanced capabilities, such as formal reasoning, mathematical problem-solving, confidence level estimation, long-term planning, and seamless adaptation to dynamic and non-stationary environments.
- Innovative learning approaches combining self-supervised learning with hybrid learning, active learning, reinforcement learning, transfer learning, relational learning or continual learning and evolutionary learning.
- Theoretical insights to advance the understanding of synergies between self-supervised and complementary learning paradigms in GPAI model development.
Scope:
Current large-scale AI models have demonstrated remarkable capabilities that have transformed numerous fields. They excel at tasks like natural language processing, image generation, and playing complex games. However, despite these successes, current models often struggle in several key areas. They lack the adaptability to seamlessly adjust to changing conditions in real-world environments. Additionally, their reasoning abilities remain limited, when facing complex tasks that require logical deduction, mathematical problem-solving, or multi-step planning. Moreover, current GPAI models frequently fail to recognise their own limitations, leading them to generate erroneous outputs when presented with queries outside their domains of knowledge. These limitations underscore the need for advancements in General Purpose AI (GPAI) that go beyond pattern recognition and towards robust, adaptive systems capable of a wider range of intelligent behaviours, for example, taking inspirations from biological and collective systems.
To push the boundaries of current AI technology, this topic seeks the development of groundbreaking GPAI models that combine self-supervised learning with complementary learning strategies. These strategies include hybrid learning, which integrates symbolic reasoning and knowledge representation; active learning, which allows models to actively seek information to improve their performance; reinforcement learning, which enables models to learn through interaction with their environment; relational learning, which focuses on learning from relational data structures; continual learning, which allows models to continuously adapt and acquire new knowledge without forgetting previous tasks; and evolutionary learning, which draws inspiration from biological evolution to optimize model architectures and parameters; and physics-based learning, which considers physical properties in the models’ architectures. By leveraging these complementary approaches, the aim is to create GPAI models that exhibit enhanced capabilities, overcome existing limitations, and pave the way for a new generation of intelligent systems capable of tackling complex, real-world challenges.
This topic prioritizes proposals that explore innovative approaches to developing GPAI models, focusing on at least one of the following key research areas:
- Hybrid Learning Architectures for Advanced Reasoning: Development of architectures integrating self-supervised learning with symbolic reasoning, knowledge representation, and neuro-symbolic methods to foster robust reasoning, complex planning, and problem-solving abilities within GPAI.
- Continual and Evolutionary Learning for Dynamic Environments: Research on paradigms enabling GPAI models to seamlessly adapt, learn from changing conditions, and retain knowledge essential for operation in dynamic, real-world environments.
- Reinforcement Learning Integration: Research on the fusion of self-supervised learning and reinforcement learning to overcome challenges like non-stationary data, algorithm sensitivity, and computational cost.
- Explainable AI and Trustworthy Decision-Making: Integration of robust XAI methodologies, exploring causal inference and counterfactual reasoning techniques to enhance transparency, accountability, and responsible use of GPAI models in alignment with European values and principles.
- Other Novel Paradigms: Research on the combination of self-supervised learning with other learning paradigms, such as active learning, relational learning, and embodied learning, to equip GPAI models with new advanced capabilities.
Proposed projects should aim for a balanced approach between theoretical advancements and practical applications, with a strong emphasis on the development of GPAI models that align with European values and principles, including the AI Act.
The potential impact of this research extends beyond scientific advancements, as it has the potential to transform key European industries and sectors, including advanced robotics, personalized healthcare, mobility, manufacturing, sustainable energy solutions, and the scientific sector, and to contribute to the EU climate neutrality objective through energy efficiency. Successful projects will contribute to the development of GPAI models that enhance productivity, improve decision-making, and foster innovation across a wide range of domains.
This topic strongly encourages the formation of interdisciplinary teams combining the necessary technical expertise. Such a collaborative approach will ensure that assessments accurately capture real-world capabilities and risks, and that the developed tools are responsive to the concerns of all relevant stakeholders.
Proposals must adhere to Horizon Europe's requirements regarding Open Science. Open access to research outputs should be provided unless there is a legitimate reason or constraint.
All proposals are expected to incorporate mechanisms for assessing and demonstrating progress, including qualitative and quantitative KPIs, benchmarking, and progress monitoring. This should include participation in international evaluation contests and the presentation of illustrative application use-cases that demonstrate concrete potential added value. Communicable results should be shared with the European R&D community through the AI-on-demand platform, and if necessary, other relevant digital resource platforms to bolster the European AI, Data, and Robotics ecosystem by disseminating results and best practices.
This topic implements the co-programmed European Partnership on AI, data and robotics (ADRA), and all proposals are expected to allocate tasks for cohesion activities with ADRA and the CSA HORIZON-CL4-2025-03-HUMAN-18: GenAI4EU central Hub.
Proposals should also build on or seek collaboration with existing projects and develop synergies with other relevant European, national or regional initiatives, funding programmes and platforms. Regarding European programmes, proposals are expected to develop synergies and complementarities with relevant projects funded under Horizon Europe but also under the Digital Europe Programme (DEP).