EUMAS 09. Session 6. Negotiation, Dialog and Laws

Managing quality in agent dialogues
Josep Puyol-Gruart

Measuring the quality of answers in agent’s conversations. Quality measures:
precision (smaller interval) and certainly (how close is to true or false). They are related: precision is more interesting when certainly is close to true or false. Talking about absolute (values for the facts) and relative (external view) quality. But explaining all these thing he’s run out of time, so he can’t explain how agents can use this quality measures.

Designing Automated Agents Capable of Efficiently Negotiating with People – Overcoming the Challenge
Raz Lin

It’s very difficult to design domain independent agents that negotiate with other agents or with people, and this is the goal of the paper. HE establihes the negotiation environment, how the agent isdesigned and shows some samples in games andother environments: Diplomacy, autONA, Cliff-Edge… finishing with the KBAgent, which includes all the characteristics developed in the previous ones

  • Generic agent / domain independent
  • Qualitative decision making
  • Non deterministic behavior / randomization
  • Incorporating data from past interactions

And now something about validation. It is a problem because it is no standard fto do that. What is a ‘good’ agent? maximal payoff/maximal social welfare/end with agreement/pass Turing test? This is an open question.

A good question from Ingrid Nunes abuot emotions, because their influence in human negotiations. But at the moment they are not taken into account

Blogged with the Flock Browser

EUMAS 09. Session 3. Self-* and organization

Topology and memory effect on convention emergence
Daniel Villatoro

Sorry, I’m a bit late, but this is the ideas that I’ve seen before, so I hope I’ll be able to complete this part later.

Measures of Context-Awareness for Self-Organizing Systems
Andrei Olaru

Points in common of selforg and ambient intelligence (a boring part about reactive and congnitive agents that I’m not going to comment) and now a bit about context-awareness. Agents connected in agrid with 8 neig (similiar to Kleinberg structure). Two measures: (i) pressure: how important a piece of information is; it represents urgency and determines how important the information is and how it is propagated; and (ii) interest (similar to attention focus) related with data, agents and facts. Showing how information is propagated.

Dynamic Evolution of Role Taxonomies through Multidimensional Clustering in Multiagent Organizations
Ramon Hermoso (a.k.a. Dani Mateo :-)

Related with adaptive organizations. When an agent arrives to a organization, it has to choose a counterpart to interact with. They propose a mechanism to create role taxonomies and to allow them to evolve (that is, to modificate them and, sometimes, to create new ones).

Some comments: Are always new roles specializations of existing ones? Are new roles based on existing actions? Removing roles? Has it any sense? (better remove it when there is no agents in it? who decides the role fora existing agents (supervision??

A Taxonomy of Adaptive Systems
Jan Calta

Talking about self-* properties:

  • self-stabilization (SSS) = closure +convergence
  • selg-organization (SOS) = adaptability +decentralization + local knowledge + homogeneity
  • self-management (SMS)  = configuration + optimization + healing (fixing)+ protection
  • self-adaptation (SAS) = can be considered as a generalization of SSS, SOS and SMS.

Conparing these three approaches in terms of adaptation, architecture and the approach used in their design. Interesting to describe correctly consensus networks. I think that is a SOS+SSS system (importance of the convergence)

Blogged with the Flock Browser

Tags: , , , ,