The workshop aims to exploit the knowledge and experience of multi-disciplinary scientific community and to extract a set of fundamental concepts describing a phenomenon of dynamic decision making with interacting imperfect selfish participants. Devices (e.g. robots), computer algorithms (e.g. controllers), humans (e.g. experts) and their combination will be considered. This challenging aim requires a deep understanding the main discrepancies between normative and descriptive theories as well as the imperfection phenomenon.
The proposed workshop grounds on the findings from the preceding workshops, goes beyond them and focuses on the following aspects:
A1. How to describe and control deliberation effort (operational cost) of an individual participant?
Optimal decision making requires performing intractable computations within a limited amount of time and hence any intentionally rational behaviour is behaviour within constraints, which can be unfeasible in complex domains. One of the key problems addressed is how to find a proper balance between deliberation effort spent and quality of the obtained solution.
This aspect can be characterised by the questions like:
- Does H. Simon’s infinite regress represent a real barrier?
- Do any-time algorithms provide a remedy?
- Do a priori prepared systematic sequential stopping rules serve the purpose?
A2. How to formalise cooperation of individual participants (fusing individual DM goals and models; creating a coalition; resolving competitive DM goals)?
Many of goal-directed participants know the goals/models of other participants in their neighbourhood, and exploit this knowledge to fulfil their selfish DM goals, which makes their design and control challenging. The solution of this problem is addressed in the ECML context, i.e. no human moderator can be considered in this role. Typical questions to be addressed are:
- How to approach and evaluate non-moderator situations?
- How to learn individual goals, their hierarchies, their commonalities, compromises?
- What methodology, common language, quantitative expressions are to be used?
A3. How to describe and to influence the resulting collective behaviour?
Research addressing emergent behaviour concerns any large multi-participant system without centralised communication and control. Examples of such systems are present everywhere: WWW, social networks communication, population biology, economical markets, society of robots, etc. The solution of the problem addressed requires inner view, adequate methodology as well as advanced ML techniques. Typical questions to be addressed are as follows:
- Can tools relating micro and macro worlds (statistical-physics or macro-economy) be applied generally?
- How to select the global features serving to subsequent decision making?
- What are the possible influential inputs for an efficient control of such collectives?
Inspiring insights how are these problem addressed by natural/societal systems are very welcome.