What is meaningful human control?
When a hiring algorithm mischaracterizes you and deprives you of a job, you want someone to give you the dignified consideration each job seeker deserves. When prices for an Uber ride skyrocket due to people attempting to flee from a terrorist attack, you want someone to make it possible to leave without spending $100. In both cases, we want meaningful control: a way to affect the outcome so that humans (e.g. users, designers, developers) can take responsibility and act upon it. This can be hard when artificial intelligence is embedded into a system, where it takes over decision making in myriad applications, such as the driving of a car or the pricing of services. It is therefore a challenge to design systems such that meaningful human control is maintained, and humans are still responsible for the end result. At AiTech we take up that challenge and aim to design human-AI systems such that we can keep the benefits of automation, but maintain meaningful human control at the same time.
Why meaningful human control?
Today’s engineers create systems that are ever more equipped with artificial intelligent technologies. Autonomous behavior of cars, robots, and decision support algorithms is becoming a reality. Our vision is that scientists should not only research the technology that makes intelligent autonomy possible, but also act upon the responsibility to ensure that design, engineering, and use of such systems embrace human values and meaningful human control.
Our ‘how to’ approach
Meaningful human control is particularly important in cases of failures or conflicts with the normative foundations of society, social conventions, and human acceptability. We believe these challenges demand a interdisciplinary effort, bringing together researchers across a wide range of fields. Our aim is to provide answers to ‘how to’ build autonomous intelligent systems that collaborate with humans towards societal and economic prosperity and the sustainable development of our planet.
AiTech’s objectives
- Understand the implications of meaningful human control for the science, design, and engineering of autonomous intelligent systems
- Build, test, break, and learn from systems under meaningful human control in practice
- Develop educational programs on the use of meaningful human control in autonomous intelligent systems