Leveraging the Mutual Strengths of Human and Artificial Intelligence
Humans and today’s AI exhibit strongly complementary skills and “forms of intelligence”. Our goal is to explore and develop new human-centric paradigms for AI that intertwine humans and machines in a more collaborative and mutually beneficial process. Therefore, we adopt a transdisciplinary research approach that combines data mining, machine learning, interactive visualisation, visual analytics, and human-computer interaction.
Research Focus
In our research, we focus on the deeper integration of humans and machines through interaction and collaboration. In this context, the learning process of humans and AI is conceived as a joint process involving both humans and machines in which they enter into a bidirectional dialogue to jointly solve a problem. This allows synergies to be exploited on both sides. The learning process of the AI can be guided by the expertise of humans and thus better aligned with the actual user purpose or need.
Human intuition and perception can be leveraged to quickly discover complex patterns, which is often difficult for AI. Beyond this, the collaborative process supports humans in gaining deeper insights into the AI, its workings and learning progress, which can lead to increased explainability, transparency, and trust.
Our collaborative and human-centred approach to AI has numerous fields of application. Wherever AI-based assistance systems are to be integrated, it can help to adapt the systems to the actual requirements of the users to iteratively improve AI methods and to make them more transparent. Especially in areas where large amounts of unlabelled data are available, collaborative learning can help to bootstrap and accelerate the training of AI systems. Concrete application areas are shown below.
Research foci in the field of human-centred AI include
We develop novel concepts for collaborative machine learning building upon methodology from active learning, reinforcement learning, and machine teaching as well as concepts from human-computer interaction (HCI) for a deeper engagement of the human in the learning process.
In this way, the learning process of humans and machines is combined and considered a temporally dynamic and mutually influencing process. We develop novel interaction metaphors for collaborative machine learning as well as learning strategies based on user observation and insights from learning theory.
VIAL combines machine learning and visualisation with the goal of jointly optimising three processes: the labelling of data, the incremental training of machine learning models, and the knowledge generation for the user.
In KAVA, we focus on the design and development of interactive and visually guided methods for the integration, management and formalisation of expert knowledge for novel AI-based data exploration solutions for experts and lay users.
We develop methodology for the explanation of AI-based decisions and design AI-based systems that learn to explain themselves in a way that is comprehensible for human users. We further aim at creating transparency by integrating interactive visual interfaces into the learning and decision-making process and developing visual representations of complex data and models
See also: Trustworthy AI
You want to know more? Feel free to ask!
Media Computing Research Group
Institute of Creative\Media/Technologies
Department of Media and Digital Technologies