Künstl Intell (2010) 24: 159–160 DOI 10.1007/s13218-010-0024-4
I N T E RV I E W
Interview with Paul Davidsson Andreas D. Lattner · Ingo J. Timm
Published online: 21 May 2010 © Springer-Verlag 2010
Paul Davidsson is a well-known expert in the field of Artificial Intelligence and Logistics for more than two decades. After his studies at Lund Institute of Technology, Paul Davidsson received the Master’s degree (1989) in computer science, and then Teknologie Licentiat (1994) as well as the Ph.D. degree (1996). After positions as assistant professor (1996–2000) and associate professor (2000–2002), he was appointed as Professor of Computer Science in 2002 at Blekinge Institute of Technology in Ronneby, Sweden. Here he founded and is head of the Distributed and Intelligent Systems Laboratory (DISL). Furthermore, Paul Davidsson has co-founded the company Noda Intelligent Systems AB in May 2005 which has received numerous awards and nominations. His research interests include the theory and application of agent technology, information systems, and machine learning. Current application areas of his research include traffic and transport systems, logistics and supply chain management, and district heating systems with the goal to both improve efficiency and reduce environmental impact. He has published more than 100 peer-reviewed articles in international journals, conference proceedings, and books. Since 2009 Paul Davidsson is also Professor of Computer Science at Malmö University, Sweden.
A.D. Lattner · I.J. Timm () Information Systems and Simulation, Goethe-Univesität Frankfurt, P.O. Box 111932, 60054 Frankfurt am Main, Germany e-mail:
[email protected] A.D. Lattner e-mail:
[email protected]
KI: Paul, you started your research on Autonomous Agents more than 15 years ago—in the early days of agent research. Nowadays, you are a well-known expert for logistics and social simulation using agent models. You started with pretty formal approaches to autonomous agents, esp. their knowledge representation. Can you please describe the development of your personal research? What have been particular reasons for shifts in your research agenda? Davidsson: Basically, I wanted to see my research results used in practice. To do theoretical research is rewarding, but to experience your work being deployed in actual application is even more satisfying. However, there is often a tradeoff between practical relevance and scientific “contribution”. The more useful a research result is for practical applications, the less scientific significance it tends to get credited. Moreover, during the last 10–15 years or so, it has been difficult to get funding for basic research. . . KI: From your perspective, what is the role of multiagent systems (MAS) and multiagent-based simulation (MABS) in logistics? Davidsson: As I see it MAS mainly provides means for coordinating logistical activities as well as enable interoperability between (legacy) information systems. The role of MABS, on the other hand, is mainly to provide tools for evaluating and testing different logistical strategies before using them in practice. KI: Could you provide us with some examples where coordination by MAS is beneficial in logistics? Davidsson: In the planning phase, agents can help in coordinating production and transport in order to achieve, e.g., just-in-time behavior, or the coordination of intermodal transports. I am here assuming a distributed MAS where the different actors in a supply or transport chain have agents acting on their behalf making local decisions based on the current situation and negotiating with each other. In the operational phase, agents may for instance be embedded in “intelligent” or “smart” goods, interacting with the vehicle
160
that transports it, or with other smart goods also being transported by the vehicle. KI: Together with your former Ph.D. student Lawrence Henesey, you evaluated strategies in container terminals. What are the major insights of this research? Davidsson: We found that MABS could be a useful tool to evaluate different strategical decisions (e.g., about infrastructure investments) as well as operational policies (e.g., regarding the order in which ships should be served) within a container terminal. KI: Did your work lead to concrete developments or an adaptation in the “real world”? Davidsson: Yes, the simulation tool was used as decision support for making strategical decisions in an Indian port, and it is currently being used at the company (TTS Port Equipment) where Dr Henesey now works. KI: How would you describe the overall development of the research field in the last decade? Davidsson: Increased maturity, in the sense that we now are taking into account more and more of the requirements of the real applications than before. Although one could wish for more deployed systems. . . KI: This is an interesting aspect: if we consider early OR models, there are only few requirements. However, the developed algorithms are widely used in practice. Do you think that we model too complex systems for development of MAS systems in logistics? What degree of abstraction could be sufficient? Davidsson: I would rather say that MAS model makes it possible to take into account more details as processing and decision-making may be distributed... Of course, this comes at the expense of providing solutions that may not be optimal. KI: Your answer also implies that there are not enough deployed systems. What are the main barriers for real-world applications? Davidsson: It could be a question of trust. Companies want robust predictable systems and often get hesitant when you talk about “autonomous” decision-making etc. Another explanation for the relative few existing deployed MAS systems in logistics may be that MAS could be very useful in
Künstl Intell (2010) 24: 159–160
the design / modeling phase, but that more traditional methods and technologies are used in the implementation phase. KI: You are also researching on social simulation. What are the connections to applications in logistics? Davidsson: The resulting logistical solution is typically a result of interaction between different actors and the decisions made by these actors. These actors, e.g., transport service providers, product suppliers, 3 PL, consumers and so on, are in the end human beings, and can thus be seen as social entities. KI: On an abstract level: How would you describe the role of Artificial Intelligence in logistics? Davidsson: AI methods provide decision support in order to make logistics more efficient. KI: Which areas of AI in your opinion are most promising for application in logistics? Davidsson: This is difficult, perhaps MAS and machine learning. On the other hand, these are the areas that I’m most familiar with. KI: How is your experience w. r. t. reservations of applying AI systems, e.g., autonomous systems, in logistics companies? Davidsson: Companies are typically hesitant towards autonomous decision-making. They often want a human in the loop. Thus, limiting the AI systems to provide decision support (rather than automation), or at least have adjustable autonomy, is usually wise. KI: In many scientific fields, benchmarks for comparing approaches have been established. Could you imagine a scientific “grand challenge”, e.g., similar to the DARPA Grand Challenge for autonomous vehicles, for AI in logistics? What would be necessary characteristics for such a challenge to your point of view? Davidsson: One such challenge could be totally autonomous transport planning. The providers of transport services should announce (on the Internet) what services they offer and the users should have AI-based software that find the best combination of services for a particular transport according to the user’s requirements. Also automatic booking of these services could be included in the challenge. KI: Thank you very much for this interview!