This moonshot aims to supplement and support human thinking. This field is known as Intelligence Augmentation (IA). This will reduce the cognitive load of time-consuming tasks such as semantic queries that clinicians, engineers or drivers experience in complex settings. This will provide more time to think more strategically, focus on more human tasks, and provide a way to verify their approach.
IA does not aim to replace human cognitive systems, but to use AI to augment human intelligence. To enable this, we need to optimize human-machine collaboration and interaction. We will be focusing on the unique and complementary set of strengths and weaknesses that both the human and the machine bring.
Working with cognitive machines will inspire, enrich and challenge our own cognitive and emotional abilities. It is also expected to bring a radical transformation of creative endeavors in science, engineering, design and the arts, as well as a deeper appreciation of the nature of intelligence itself.
Cognitive Models as Surrogate Models for Explainable AI
Biological intelligence is explained with the help of cognitive models: mathematical or computational models that reproduce capacities such as visual categorization, language learning, and decision making. Can cognitive models also be used to explain the behavior of “black box” artificial intelligence?
In this project, we evaluate the usefulness of different cognitive modeling frameworks for understanding, predicting, and intervening on the behavior of “black box” systems such as deep neural networks. We consider different stakeholders, from expert developers to end-users and external regulators, and determine the extent to which models that center on mental representations, complex dynamics, and/or statistical inference can be used to satisfy these stakeholders' explanatory needs. In this way, we aim to identify normative guidelines and best-practice methods for Explainable AI, and facilitate comparisons between human and artificial intelligence.Read more
The Thinking Assistant for Software Systems
In a system development project, there are various types of developers (architect, programmer, tester, …) that can ask various types of questions to the intelligent assistant. A security-engineer might ask: Which security mechanisms are used in this system? For a maintenance task, a software developer might ask: across which parts of the system together implement the feature F?
What is a Thinking Assistant for Software Systems? A thinking assistant will need to mine and combine information from various sources about the system – these can range from requirements, architecture-descriptions, source codes and test scripts – into a knowledge-base.
Reasoning and dialogue systems on top of this knowledge based should produce the answer to questions asked by developers.
Assisting medical decision with Explainable AI
Modern laboratory experiments in bio-medicine generate large amounts of structured, heterogeneous data that can be used to build models and assist critical decisions in clinical environments. Due to the delicate nature of these decisions, in real-world scenarios, models must be interpretable or explainable, i.e., domain experts must be able to inspect and understand the rationale underlying decisions.Read more
Personalized healthcare processes using AI for improved treatment
Health care has developed significantly in the past decades. Advances in medicine and technology allowed to treat more and more diseases and in a more effective way, with visible improvements in terms of both duration and quality of life for individuals. However, at the same time the complexity and the costs of the health care systems have grown exponentially. Nowadays, multiple solutions often exist to cure a particular disease, and determining the best option for the patient at hand is a challenging issue, where many factors have to be taken into account. In recent year, significant efforts have been made to develop standardized care pathways, representing best practices for a number of treatment processes, with the goal of standardizing and improving the quality of care. However, several studies have shown that the proposed care pathways are often either not used or used only as guidelines, while the actual treatment processes can deviate quite significantly from them.Read more