- James P. Delgrande, Birte Glimm, Thomas Meyer, Miroslaw Truszczynski, Milene Santos Teixeira, and Frank Wolter. Current and Future Challenges in Knowledge Representation and Reasoning (Dagstuhl Seminar 22282). In Dagstuhl Reports, Volume 12, Issue 7, 62-79, Schloss Dagstuhl - Leibniz-Zentrum für Informatik. February 3, 2023
Knowledge Representation and Reasoning (KR) is the field of Artificial Intelligence (AI) that deals with explicit, declarative representations of knowledge along with inference procedures for deriving further, implicit information from these symbolic representations. Research in KR as a mature area of AI is commonly taken as being marked by an Artificial Intelligence Journal Special Issue on Nonmonotonic Reasoning in 1980. In 1989 the Principles of Knowledge Representation and Reasoning Conference was founded, providing a dedicated, specialised forum for research in the area. While KR is one of the oldest and best-established areas of AI, it has continued to grow and thrive over the years. Most of the original research areas have evolved significantly, and have matured from the discovery and exploration of foundations, to the development and analysis of systems for emerging or established applications. Yet other areas, such as argumentation, arose much more recently, and are now thriving areas of KR.
While progress in KR has been steady and often impressive, it has not kept pace with the recent significant successes in AI in the use of statistical techniques and machine learning (ML). As a result, much of the work in AI, and much of the public perception of AI, centres on machine learning and on statistical applications. Nonetheless, we take it as given that KR is a vital, essential area of AI, and that research and development in KR remains necessary. Indeed, despite the unquestionable successes in machine learning and statistical techniques, limitations of these approaches are now emerging that, we believe, can only be overcome with advances in KR. Indicative of this is the recent interest in "Explainable AI“, which requires a reference to declarative structures and reasoning over such structures. Furthermore, and in common with the majority opinion in AI, cognitive science, and philosophy, we take it as given that symbolic, declarative representations of knowledge are essential for any ultimate, general theory of intelligence.
For all of these reasons, a reassessment of the area of Knowledge Representation was a very timely undertaking of the Dagstuhl Perspectives Workshop 22282 "Current and Future Challenges in Knowledge Representation and Reasoning“. During the seminar, the participants assessed the current state of KR along with future trends and developments. A questionnaire, which had been earlier distributed to the participants, helped in this assessment. Alltogether, the seminar served as a basis for developing an innovative agenda for the next 10-20 years of KR research. Key findings are measures to support a synergistic relationship with other subareas of the rapidly-changing field of AI and of computer science as a whole, e.g, through tutorials at the major KR conference, through new conference tracks and updated reviewing guidelines. The seminar further identified research areas for emphasis, assessed prospects for practical application of techniques, and considered how KR may address limitations of statistical techniques and machine learning.
The program comprised invited talks, panel discussions, working groups, and general dicussions. While the invited talks were agreed upon beforehand, the topics of the working groups (apart from Day1) were decided interactively with all participants to allow for flexibility and reacting to the talks and the triggered discussions. Day1 started with a short welcome and participant introduction session, followed by an assessment of the past and present of KR in the form of two invited talks by Anthony Cohn and Thomas Eiter. The remainder of Day1 was dedicated to presenting the questionnaire results, which also prepared for the first working group on rethinking the call for papers (CfP) for the main KR conference, which not only served as rethinking the CfP, but also steered the working groups into thinking about the definition of KR as an area. The day closed with a report from the four working groups and indeed identified changes for the CfP, but also for the track structure and the recruitment and instructions for reviewers.
Day2 focussed on the relationships of KR with four neighboring areas. For each sub-areas we began with a short invited talk (20 min) followed by a commentary (5 min), also invited, and a short general discussion (5 min). The function of the commentator was to look at the area from a different angle or give another perspective to avoid a too personal or narrow a perspective. The four talks addressed "KR and AI“ (Ian Horrocks, commentator: Sébastien Konieczny), "KR and ML“ (Francesca Toni, commentator: Ana Ozaki), "KR and Information Systems“ (Diego Calvanese, commentator: Meghyn Bienvenu), and "KR and Robotics“ (Gerhard Lakemeyer, commentator: Michael Beetz). Working groups on research challenges for these subareas concluded the day.
The third day began with a short talk on "Handling Uncertainty“ (Jean Christoph Jung), for initiating a panel discussion on this topic. The morning concluded with a continuation of the working groups on sub-areas of KR from the previous day. The afternoon was dedicated to hiking and biking in smaller groups.
Day4 started with short invited talks on "Applications of KR“ (Esra Erdem, Thorsten Schaub, Michael Tielscher). The remainder of the day was dedicated to working groups on assessing the state of the art in sub-areas of KR and to expanding KR. For this latter group, we discussed the fact that geographically KR is stronger in Europe than in other parts of the world. As well, we considered how to attract new talent and how to reach out to disadvantaged groups, along with thinking of new forms of events such as hybrid conferences or virtual seminar series.
The final day of the seminar looked at strengthening the interaction between sub-areas of KR and wrapped up with statements of the participants regarding their personal impressions and "take-home“ messages. This has, for example, already led to the creation of a novel KR discussion channel (on a Discord server). Key findings include that KR applications are very important to make the field visible and that applications are to be made more visible, e.g., through a journal special issue. Another outcome includes measures to reach out to other areas of AI, in particular machine learning and statistical techniques, where symbolic approaches can make contributions, e.g., for general intelligent agents. A separate Manifesto will provide an assessment of the area, and will give a set of recommendations regarding the future of KR and its promotion
Knowledge Representation and Reasoning (KR) is the field of Artificial Intelligence (AI) that deals with explicit, declarative representations of knowledge along with inference procedures for deriving further, implicit information from this knowledge. It has evolved significantly over the last 40 years, and research in many subareas of KR has matured from the exploration of foundations, to the development and analysis of systems for emerging or established applications. However, while progress in KR has been steady and often impressive, it has not kept pace with the recent successes in AI in the use of statistical techniques and machine learning. Indeed, much of the work and focus in AI has shifted to machine learning and statistical applications in areas like vision, natural language understanding, and big data. Nonetheless, we take it as given that KR is an essential area of AI, and that research and development in KR remains necessary for any ultimate, general theory of intelligence.
With this belief as a key motivation, this Dagstuhl Perspectives Workshop aims to assess the current state of KR along with future trends and developments, and to develop an innovative agenda for the next 20 years of KR research. Among its goals are identifying areas for emphasis, assessing prospects for practical application of techniques, and considering how KR may address limitations of statistical techniques and machine learning. Because KR is strongly interdisciplinary, another important goal is to develop strategies for fostering links between KR and other areas of AI and computing science.
Tentatively, the workshop will begin with introductory and overview talks on select topics, followed by short position statements by participants. A major part of the workshop will be given over to working groups, reports from these groups and general discussion. The final morning will discuss meeting outcomes, next steps, and planning for the future. The workshop will be described in a Dagstuhl Report, as with all seminars. Two further outcomes will be a Dagstuhl Manifesto that will describe the suggested initiatives and goals resulting from the meeting, as well as a more focused and technical document written for researchers in the field.
A key condition of success for the workshop is that the participants reflect the diversity of the field, represent all its main areas of research, offer balance between theory and practice of KR, and come from a broad range of geographical areas. As an additional prerequisite for success, the workshop seeks to attract participants from areas adjacent to KR, such as natural language understanding, planning, and constraints, to learn about their views on KR and on effective ways KR and their areas may develop critical synergies.
The attendance in the workshop is limited. Therefore, active participation and direct contributions from all attendees are necessary for the workshop’s success. To create an atmosphere of creativity and a sense of a shared vision, all participants are expected to attend for the full duration of the workshop, to contribute during the workshop, and to get involved in the preparation of workshop documents once the meeting is over.
- Michael Beetz (Universität Bremen, DE) [dblp]
- Meghyn Bienvenu (University of Bordeaux, FR) [dblp]
- Piero Andrea Bonatti (University of Naples, IT) [dblp]
- Diego Calvanese (Free University of Bozen-Bolzano, IT) [dblp]
- Anthony Cohn (University of Leeds, GB) [dblp]
- James P. Delgrande (Simon Fraser University - Burnaby, CA) [dblp]
- Marc Denecker (KU Leuven, BE) [dblp]
- Thomas Eiter (TU Wien, AT) [dblp]
- Esra Erdem (Sabanci University - Istanbul, TR) [dblp]
- Birte Glimm (Universität Ulm, DE) [dblp]
- Andreas Herzig (Paul Sabatier University - Toulouse, FR) [dblp]
- Ian Horrocks (University of Oxford, GB) [dblp]
- Jean Christoph Jung (Universität Hildesheim, DE)
- Sebastien Konieczny (University of Artois/CNRS - Lens, FR) [dblp]
- Gerhard Lakemeyer (RWTH Aachen, DE) [dblp]
- Thomas Meyer (University of Cape Town, ZA) [dblp]
- Magdalena Ortiz (TU Wien, AT) [dblp]
- Ana Ozaki (University of Bergen, NO) [dblp]
- Milene Santos Teixeira (Universität Ulm, DE)
- Torsten Schaub (Universität Potsdam, DE) [dblp]
- Steven Schockaert (Cardiff University, GB) [dblp]
- Michael Thielscher (UNSW - Sydney, AU) [dblp]
- Francesca Toni (Imperial College London, GB) [dblp]
- Renata Wassermann (University of Sao Paulo, BR) [dblp]
- Frank Wolter (University of Liverpool, GB) [dblp]
- Artificial Intelligence
- Logic in Computer Science
- Symbolic Computation
- knowledge representation and reasoning
- declarative representations
- formal logic
- applications of logics