Human-centered AI Systems
Human-centered AI systems are designed to interact with humans and deliver explainable and comprehensible results.
At the Lamarr Institute, we’re developing human-centered approaches that bridge the gap between Machine Learning (ML) methods and human cognition. On one hand, human-centered AI systems are designed to align with human goals, concepts, values, and cognitive processes. On the other hand, these systems aim to optimize their functionality by harnessing the power of human perception and intelligence. Visual Analytics serves as a pivotal tool in integrating human and machine intelligence, leveraging human-computer interaction to enable the transfer of expert knowledge into ML models. This makes them capable not only of accurate predictions and problem-solving but also of generating human-understandable explanations.
AI-Human Communication and Collaboration
The main objective of human-centered AI systems is to combine and make the most of human and computer intelligence. When developing ML models that complement and augment human abilities, we need to focus on tasks and situations where human experts have difficulties and avoid replacing humans where they perform well.
To achieve these goals, we take into account two sides of human-centered AI systems. Firstly, we have ML for humans which means computers adapting to human thinking and perception, providing human-friendly and in particular visual information.
Secondly, there is the concept of humans for ML which means computers taking advantages of the power of human perception and intelligence, thus making optimal use of human intellectual capabilities for developing ML models.
The following diagram shows the relationship between human-centered AI systems and the Lamarr Institute’s research paradigm of Triangular AI. A key aspect is humans transferring expert knowledge while taking into account the context in which the ML model is going to be used.
Contact persons
Visual Analytics: Synergies between Human and Machine Intelligence
In the context of visual analytics, interactive visual interfaces are developed which play a key role for enabling synergies between human and machine intelligence. Such interfaces provide a means of human-computer communication which facilitates understanding as well as more abstractive perception and thinking by humans.
Visual analytics provides essential support for Machine Learning models to learn from human experts, particularly regarding domain concepts, logic, causal relationships, and explanatory methods. Additionally, visual interfaces can enhance the usability and adoption of these models by presenting and explaining their functionality and results in a user-friendly manner.
Ultimately, visual interfaces are meant to support collaborative human-AI problem solving and decision making, envisioning a partnership with complementary contributions, discourse, and mixed initiative. Future human-centered AI systems should be compliant with general human cognitive abilities and limitations and ready to adapt to specific user categories, for instance experts, decision makers or patients.