R3: Sustainability and Efficiency in Human-centered Environments

Current AI models for ITS can contain hundreds of billion model parameters and tokens for training. These exhaustive resources limit their availability for fast model adaptation in human-machine interaction and continued adaptation during their life-cycle. Furthermore, there is a need for data- and resource-efficient AI technologies in the light of AI’s ecological footprint. While research begins to address technological demands such as learning from few data or resource-efficient model distillation, we also need to consider efficiency regarding human-computer-interaction and cognitive constraints of a human partner. SAIL leverages and connects cognitive components, human-computer interaction expertise, and novel engineering technologies to create innovative approaches for challenges pertaining to data economy and knowledge integration, energy efficiency, and cognitive efficiency. Central research questions are as follows:

  • How can the amount of necessary training data for large AI models for ITS be reduced without harming performance? Can prior knowledge (e.g., physical models) be integrated to generate reliable AI models at low cost. How is prior knowledge best modeled to improve run-time, amount of training data, reliability, or usability?
  • How can the energy efficiency of AI algorithms be improved? Is domain-specific computing a viable path towards energy-efficient AI? Do approximative solutions provide sufficient guarantees for human-centered AI?
  • What are viable models of human empowerment in this socio-technical context? Which types of human interventions are natural and efficient, thus limiting the ‘cognitive load’ of required interventions? How do human-centered objectives relate to common learning objectives and knowledge representation models of AI models for ITS?