Humans and Artificial Intelligence Sharing Control and Decisions – HAISCoDe

We propose a research program of 10 PhD theses coordinated by Normandie Université (NU), which federates the most important institutions of higher education and research in Normandy. These PhD grants will be co-funded by Région Normandie, in line with its strategy for research and innovation.

The program aims at developing novel artificial intelligence (AI) approaches in contexts where humans and AI together control systems and make decisions. An example of such contexts is autonomous vehicles, where most piloting is performed automatically, but control must be taken by the human in specific situations (e.g., reduced visibility). Another example is medical imaging-based diagnosis, where AI is used to assist physicians in the tedious task of image analysis by focusing their attention to the most meaningful images or regions of images. Yet another example is the case of a complex organization, like a harbor, in which human decisions about the management of traffic must be supported by models and prediction tools, which themselves embed expert knowledge.

Our objective is to develop ambitious and original AI techniques for such contexts and understand their fundamental underpinnings, along two main directions as suggested by the above examples:
- sharing of control, where AI and humans together control a system in an interactive loop (for instance, control passing back and forth between an automated and a human pilot), and
- sharing of decisions, where the interaction occurs essentially in two phases (for instance, AI-based segmentation of medical images, then physicians making decisions based on them).

These scientific themes are very active research areas at NU, where many teams have gained solid national and international reputation. One of the added value of this project will be to foster interaction between researchers and PhD students in both themes, around the numerous aspects common to both directions: taking into account human knowledge, modeling the human’s expectations, ensuring that decisions are explainable, reliable, robust, and ethical, to cite a few.

This general objective will be instantiated on two main fields of application, for which NU has important forces: transportation and health. Normandy indeed has an important expertise on these two fields, specifically on autonomous vehicles, logistics, and medical imaging. It also enjoys important infrastructures, especially the harbor in Le Havre (2 nd most important port in France, and 11 th in Europe), the ARCHADE platform for hadrontherapy and the CYCERON platform for medical imaging in Caen, and the Rouen Normandy Autonomous Lab for autonomous vehicles.

These applications will drive research in various facets of AI for which NU has internationally renowned research teams and individual researchers. One can cite in particular: automated decision and planning in interaction with humans; machine learning, especially deep learning; discrete and continuous optimization; knowledge extraction and data analysis; multi-agent systems. Moreover, at the crossroads between all these techniques lie important common scientific problems, notably large-scale optimization and modeling of data, knowledge, users, and systems.

We propose to fund 10 PhD theses on these topics, with a strategy for fostering interaction between them through scientific events at different levels held on a regular basis. Our proposal does not include a definitive list of PhD subjects, since research in the targeted scientific fields and innovation
in the targeted application areas evolve very fast. We will rather open a call for PhD subjects once this program is funded, have the proposals and candidates reviewed and ranked by external experts, and decide which to fund depending on these evaluations and on a balance between the scientific directions outlined above, a process successfully used for several years at NU.