The purpose of DASH is to allow humans to efficiently tell an AI-based system what to do without having to spell out all details, and without losing meaningful human control over that system.

Problem Context

Delegation has been recognized as a crucial notion within the context of human-machine teaming. It allows a delegator to set an agenda either broadly or specifically but leaves some authority to the subordinate to decide exactly how to achieve the commands supplied by the delegator. Research has made significant progress on a wide range of issues that are essential for enabling humans to delegate tasks to machines, such as developing appropriate levels of trust, allowing the human to have situation awareness of a diverse range of aspects of the work domain, and characterizing different levels of automation between human delegator and the machine.

Recent developments in artificial intelligence and robotics have significantly increased the application possibilities of delegation, yet raised new challenges as well in the domains of meaningful human control and distributed delegation. The idea behind meaningful human control is that, in a human-machine teaming, the human must always remain in control of ethically sensitive decisions. This challenge is especially pressing, because AI is becoming ever more capable. It is being applied to an increasingly wide range of problems, including ethically sensitive ones such as medical diagnosis, balancing driving speed and safety, and warfighting. Designing delegation that safeguards meaningful human control is therefore one of present time’s profound challenges.

The challenge of distributed delegation refers to the fact that most current delegation systems assume a one-to-one relation between delegator and subordinate. However, an AI is hardly used by one user, nor does it take place at one particular moment. For example, an organization might set behavioral constraints to a system during deployment, while an operator commands the AI system to pursue a certain goal during operation. Both of these instructions determine the AI system’s behavior. Furthermore, the machine may also act as a delegator, e.g. for delegating sub-tasks to other machines or even back to humans (as is the case when humans and machines work together as equal teammates). To be future proof, these mechanisms should be extended to enable various forms of delegation that are distributed in time and source of the command. The most advanced form of delegation in a Human-agent teams also allows machines to delegate tasks to humans, leading to complex delegation chains.


The challenges of distributed delegation and meaningful human control are closely related. The DASH project aims at confronting these challenges by iteratively designing, prototyping and evaluating a uniform delegation system (called DASH) for controlling AI-based systems such as robots and information agents.


The video below demonstrates DASH’s intended use. DASH acts as an intermediary between the human and the AI system in a human-agent team. DASH can be used to assign both high-level goals to the AI systems or to control the systems at a lower-level. The latter is possible by taking a look at how DASH breaks down the higher-level goals into tasks for the AI systems to take on and by adjusting these tasks as desired. Finally, it is possible for humans to use DASH to place constraints on all AI systems (such as not falling in enemy hands).


  • Jurriaan van Diggelen, Sr Research Scientist, TNO, e-mail:
  • Jasper van der Waa, Research Scientist, TNO, e-mail: