Law enforcement is faced with huge amounts of data from online platforms, digital marketplaces, or communication services. Finding evidence in such large collections of text, images, and other data and bringing it to court is a time-consuming process. Artificial Intelligence tools are a promising way to make this more efficient and effective. But currently no clear legal regulations for AI tools are in place. In AI4Intelligence we let AI tool development, the use of the tools by investigators, and legal regulations go hand-in-hand so investigations lead to trustworthy evidence that is admissible in court.

Public organizations, such as the police, increasingly use algorithms for work processes. However, this useage has resulted in concerns regarding undesirable outcomes such as a bias or lack of fairness, in contact with citizens. This project investigates how; (1) value-sensitivity and (2) transparency can strengthen citizen trust in the use of these algorithms.

Many discrete optimization challenges in Computer Science, such as the travelling salesman problem or train scheduling, can often be solved optimally in practice, with performance guarantees for large-scale instances. This is a great achievement; these problems are usually NP-complete, which is a strong theoretical certificate for intractability. Continuous computational challenges are intractable in practice even on moderately sized scales. We want to show that many algorithmic challenges are complete for the complexity class ER. This captures the aforementioned intractability phenomenon from a theoretical perspective, in the process, unveils deep relations to real algebraic geometry and implies that discretizing the problem is unlikely to be possible. One such continuous intractable problem is motion planning (MP). A practical example of MP is the control of a robot arm to perform a given mechanical task without collisions. The continuous nature of movement in MP is the intuitive reason why methods, which have been applied to problems in NP, fail for MP. Showing ER-completeness makes this intuition a tangible fact. Awareness of this relation is crucial for algorithm designers as it is used to identify tractable cases. In a recent paper, Mikkel Abrahamsen, Anna Adamaszek and I could show that the Art Gallery Problem is ER-complete, settling the complexity of this long-standing open problem and implying that all current practical approaches are non-optimal. In a subsequent paper, in preparation, I designed a provably correct algorithm avoiding the lower bound in a careful way. A student is currently implementing and testing the algorithm.

Since the 1990s, the number of websites has increased extensively and online information seeking (IS) has become an integral part of our daily activities. In parallel, the population aged over 65 is increasing in all OECD countries, a trend that is expected to continue in the coming decades (OECD, 2009). Accessing the Web for older people is a very important challenge, as it helps them to foster independence, reduce isolation, and to increase communication and wellbeing. Despite its importance, there is still a lack of focus on improving the use of Internet by older people by developing detailed process models that account for age-related individual differences in information seeking. The aims of this research project are: (1) To develop a new computational model of information seeking taking into account individual differences related to aging. (2) To create tool to support older people while seeking information on the Web. Firstly, the development of such a model is required for a better understanding of the IS activities and for supporting older users. To this end, the current project aims to determine the effect of cognitive processes and abilities related to fluid and crystallized intelligence on IS while considering the age of participants (old vs. young adults) and type of documents to be interacted with, during different stages of IS, i.e. from the formulation of the query with a search engine to the processing of and navigation through web pages. We will compare search and navigation performance and strategies of younger and older web users to the performance obtained by two complementary models of search and navigation: SNIF-ACT and CoLiDeS+Pic. Secondly, the objectives are to achieve a better understanding of cognitive difficulties experienced by older adults, and to improve our model of IS activity and on basis of that to develop tools to support this IS activity. The project falls within the fields of Cognitive Psychology (and Ergonomics) and Science and Technology Studies. Two methods will be used: (1) Experimental method, using various techniques depending on the objectives of the study (analysis of online seeking behavior and eye tracking); (2) simulating and modelling activity. The project is divided into four parts with different objectives and experiments. The experimental studies aim to determine the effects of age (younger vs. older adults), the level of prior domain knowledge (high vs. low), the relevance of pictures and specificitiesof seek tasks to be performed (difficulty, preciseness, complexity) on IS activity. In parallel, CoLiDeS+Pic and SNIF-ACT models will be developed to simulate users performance and behavior. More precisely, Part A establishes baseline performance in seeking information from formulation of queries to navigation in websites. The experiments will be carried out in an ecological context, i.e. with real search engines and websites. Part B addresses the role of aging, pictures, prior domain knowledge, search difficulty and complexity during formulation of queries and selection of links visited with experimental search engines. Part C studies the role of aging, pictures, prior domain knowledge, and search difficulty during navigation in experimental websites. Part D aims to develop a new model of IS taking into account participants age. Results from each stage of the project will lead to an assistive tool that supports older people to more effectively seek online information.

Network science is in great need of algorithms that can analyze increasingly larger networks fast. However, this ambition is undermined by the recent theory of fine-grained complexity. It predicts tight (conditional) lower bounds on the complexity of graph distance, counting, and enumeration problems that underlie network science. These lower bounds, while polynomial, are too high for the staggering size of modern data sets. Fortunately, the lower bounds may be circumvented by parameterized algorithms. Still, we lack systematic studies into the effectiveness of this approach. In particular, commonly studied parameters are linear in the input size for standard models of networks. Hence, current parameterized algorithms might not yield the urgently needed improvements to analyze current and future real-world networks. The proposal aims to design new parameterized algorithms to enable the analysis of huge networks. The proposal will initiate a design cycle to discover, analyze, exploit, and validate new parameters geared towards the graphs and problems commonly encountered in network science. This will be achieved through interaction with domain experts, the analysis of data and mathematical models, and building the required algorithmic knowledge of parameterized computation within P. The project will yield a parameter ecology for polynomial-time problems and implementations of the discovered algorithms. In doing so, the proposal seamlessly integrates fundamental research and practical considerations, and may impact the many application areas of network science.