Human-Centred AI Systems
We believe that AI-enabled systems must serve the collective needs of humanity. To that end, TU Delft supports the evolution of the science and practice of AI towards a "human-centred" approach. Human-centred AI emphasizes that the improvement on AI-technology should not only focus on issues of accuracy and speed, but also on the relationship between this powerful technology and their context of design, implementation, and deployment.
The unique perspective of TU Delft on the topic of human-centred AI is our strength in system’s human-centred design and engineering. Several organisations are strong in formulating requirements that AI systems should fulfil, in terms of its adherence to norms and values, safety and reliability, usability, and social embedding. But there are few organizations that take steps towards realizing systems that comply to such norms and values. Similarly, there are plenty of organizations that excel at the AI technology level, greatly optimizing individual algorithms and components. Eventually, these algorithms and components need to be integrated into broader socio-technical systems where small gains or technofixes in one component may not lead to proper outcomes in implementation. Achieving an overall desirable result is a stronghold of the Delft engineer, which we can deliver by working together across faculties and disciplines. The research community around Human-Centered AI Systems, also based on the TU Delft AI Labs that deal with human-centered AI is led by Alessandro Bozon, professor Human Centered AI. This community is concerned with research and teaching about understanding, design, and engineering of AI behaviour.
Key Properties of Human-Centredness AI Systems
TU Delft aligns its human-centred AI research and education with its engineering strengths. For that reason, we prefer to speak about human centred AI-systems, emphasizing the need to design and engineer systems in which human-centredness is embedded, as detailed in the following key properties:
- The AI-enabled system complements and augments human intelligence and abilities, rather than serve as a mere substitution;
- The design, development, control, and operation of AI-enabled systems are well-situated around actual human characteristics, values, intentions, and behaviours, rather than as we want or assume them to be;
- The AI-enabled system relies on a computational infrastructure that allows for relevant values to be reflected in the AI-enabled systems by design;
- The AI-enabled system integrates safely and appropriately in the broader socio-technical context;
- The success of the AI-enabled system is measured by the positive impact they have on society and how they safeguard against real harms to citizens, and not by improved narrow technical metrics.
Human-centric AI is inherently transdisciplinary as it brings together science and practice in human-machine interaction, computer science, design, systems engineering, psychology, work and organization, law and policy, ethics and philosophy, political economy, and an entire range of application-specific disciplines. In that sense, the truly ‘human-centred’ approach to AI is a fundamentally new field.
TU Delft AI Labs
This TU Delft AI Labs are closely involved in research and education in the field of responsible design of human-centric AI and data-driven systems:
In the mood for mudAs a biologist and soil scientist Julia Gebert often finds herself getting her hands dirty. With mud, that is. In the past two years she has many times boarded a small boat to cruise among the huge container ships in the port of Hamburg while collecting mud samples.
Increased road safety with high-resolution automotive radarIf it is up to Cicero Vaucher, cars will be transformed into robots with the capability of sensing the environment, thinking, and acting autonomously.
Contactless interaction between man and robotIn traffic, in the supermarket or in the factory: in the near future, robots will no longer be standalone machines, but systems that operate and make decisions within the same environment as people. This is placing different demands on the design and development of such robots.
So much more than 3D visualisationElmar Eisemann and Ruben Wiersma tell about how mathematics and computer graphics will enhance our perception of art.
Counting grains of sand to understand coastal dunesThe Dutch coastline is continually changing. Coastal engineer Sierd de Vries is using ventilators, wind tunnels, drones, jet skis and laser scanners to gain an insight into the development of coastal areas and dune formation. He can often be found counting grains of sand.
Nynke Dekker's insatiable curiosityHer research takes place at the interface of physics, chemistry and biology. Professor of Molecular Biophysics Nynke Dekker is internationally renowned for her pioneering research into the interactions between individual proteins and DNA and RNA molecules, as well as the advanced techniques she developed to make these interactions visible. Colleagues call her ambitious, thorough, content-driven and insatiably curious.
There’s still room on the water!Freight transport by water is cheaper and uses less energy than by road. What’s more, there’s more room on the water than on our congested roads. We could fully exploit these benefits by first solving a few notorious bottlenecks, such as container ships that aren’t loaded to capacity, suboptimal vessel navigation and congested locks.
A supermodel for all the glaciers on EarthThe Dutch coastline is continually changing. Coastal engineer Sierd de Vries is using ventilators, wind tunnels, drones, jet skis and laser scanners to gain an insight into the development of coastal areas and dune formation. He can often be found counting grains of sand.
Read more about this theme: