Today's artificial intelligence is technology-driven, data-intensive and resource-aggressive. Today's applications of AI technology in industry include centralized, edge and cloud-based back-end AI with very resource-intensive algorithmic strategies for machine learning (computing power, memory, energy), massive environmental impact (greenhouse gas emissions) for model training and inference, and complicated collection, generation and handling of huge training datasets. Recent advances in conversational generative AI (OpenAI GPT-4, Bard, LaMDA, Bedrock, etc.) have shown that the application of AI can have a significant impact on the environment.

The Streaming AI project aims to advance the development of AI for industrial applications. In contrast to conventional (i) pre-trained, (ii) holistic and (iii) resource-intensive AI, it seeks to harmonize AI technology and methodology with the reality of digital transformation in industry by introducing (i) streaming machine learning methods, i.e. training models on the fly, thus avoiding the need for mass training data, and connecting with (ii) on-device machine learning methods for AI federations distributed across different manufacturing plants, machines, processes and devices.