3AI – The Third Wave of Artificial Intelligence aims to explore the Third Wave of Artificial Intelligence: AI systems should acquire human-like communication and reasoning capabilities and be able to recognize, classify and adapt to new situations autonomously.
Our research approach goes far beyond the level of performance of AI and machine learning methods achieved in recent decades: AI systems should no longer act merely as tools that execute rules programmed by humans or derive solutions to problems from datasets curated by humans, but should be able to act as ‘colleagues’. The goal is not to replace human intelligence, but to extend it reliably and for the benefit of society in an increasingly complex world.
The AI systems we research should not only be able to learn, but they should also be able to grasp – novel – facts and be able to link them to forms of abstract thought. They will draw logical conclusions and make contextual decisions and learn from them again. In addition to algorithmic fundamentals, new methods of system design, new methods of software engineering and data management for AI in particular will play a key role for this. In the long term, the paradigm of systems AI should form the foundation for the development of the “third wave of AI” – artificial intelligences that learn, reason, build knowledge, and (inter)act in partnership with humans in a context-aware manner. To explore the foundations of Third Wave AI, we are working closely together in a research team of Computer Science, Artificial Intelligence, Cognitive Science, and Life Sciences.
To meet the grand challenge of Third Wave AI, we need to rethink AI from the ground up and create new foundations that seamlessly integrate machine learning, optimization, and reasoning-from spatial and temporal, to physical and domain-specific, to cognitive models-because one component alone is not enough to develop complex AI systems with human-like capabilities. As a bracket and leitmotif for 3AI, we envision a programming paradigm for system AI that makes the process of developing complex learning-based AI systems efficient, safe, and easier to reproduce.
3AI is coordinated by Prof. Mira Mezini, Prof. Kristian Kersting, Prof. Jan Peters and Prof. Stefan Roth (all TU Darmstadt).
In the Foundations of Systems AI Design mission, we bring together our previous contributions to probabilistic programming and learning, self-updating computation, AI-supported programming, and deep databases to create the software foundations of systems AI.
Foundations of a common Systems AI software framework
Linking continuous & combinatorial AI to improve generalization & robustness of AI
Ensuring good performance, robustness and trustworthiness, esp. when built & used by non-AI experts
The goal of the Foundations of AI Methods mission is to show step-by-step how to solve challenging AI/ML tasks in NLP, robotics, and computer vision using the Systems AI paradigm.
Multimodal learning and learnable labeling & data transformations
(Interactive) learning during tasks & grappling new situations, w/o forgetting lessons learnt
Achieving consistency by informing systems about the governing mechanisms & constraints of a domain
The Life Sciences Tasks mission takes up the systems AI paradigm developed in the other two missions as a challenge in one of the most important application areas of AI.
Systems AI for medicine at a holistic level, i.e., in the light of complex medical knowledge
Better clinical decisions for solid cancer (prostate, bladder) based on virtual slides & metadata
Integration of structured clinical pathological findings, histological images, molecular data & known interactions between gene alterations and drugs