Practice: Trust-based Interaction Design

Design interactions in multi-agent systems that can be evaluated with trust models and agent decisions that use the trust values thus created to make the system more robust and efficient.

Relationships

Content References

How to adopt the Trust-based Interaction Design practice

Key Concepts

Trust Model

Tasks

Bibliography for Trust and Reputation in Multi-Agent Systems

The Trust Lifecycle

Inputs

Purpose

Trust-based interaction design enables the agents in the system to determine and select trustworthy interaction partners with a high likelihood of successful completion of an interaction. It helps make the system more robust against unintentional and malevolent interaction behaviour and can even enable more efficient problem solving.

Background

Trust in multi-agent systems is often defined as the expectation that an interaction partner behaves favorably. It is useful in open systems in which agents engage in critical, contract-based interactions. Trust has been investigated intensively and a number of models have been proposed that allow the quantification of this expectation. Trust can be the basis of a reputation system.

Main Description

The key principles of trust-based interaction design are:

Evaluate interactions for their favorability. As agents interact with each other, they accumulate data about each other that can be used in decision making.
Create a model of other agents' behaviour to make predictions about them. A good trust model allows an agent to make a prediction about the behaviour of a potential interaction partner in future interactions.
Use knowledge about other agents to minimise risk and interact more efficiently. The knowledge accumulated in the system can help in making decisions that minimise risk by choosing interaction partners with a high probability of a favourable interaction outcome and increase efficiency by reducing the need to repeat interactions or use corrective mechanisms.

The concept of an interaction can here be regarded very generally. Not only are communications with other agents an interaction but also querying of sensors, the use of environmental data, and others. All these interactions are sources of uncertainties that can be mitigated by trust, as described in Trust in Multi-Agent Systems.

As part of this practice, the stakeholders need to perform the following tasks in order to customise the trust lifecycle for a specific system:

Find sources of uncertainty that can be mitigated by trust (cf. Identify Sources of Uncertainty)
Select trust models and metrics (cf. Define Trust Model)
Devise agent decisions based on trust (cf. Define Agent Decisions based on Trust Values)
Design observation of interactions (cf. Design Trust Measurement Infrastructure)
(Optional:) Incorporate reputation system (cf. Design a Reputation System)

How to read this practice

The best way to read this practice is to first familiarize yourself with its overall structure: what is in it and how it is organized.

Next, review the key concepts for the practice. An especially important concept is Trust in Multi-Agent Systems. After you understand the basic principle of trust, turn your intention to The Trust Lifecycle that describes the relationship between interactions and trust. At this point, you might also want to read some of the seminal papers on trust that are pointed out in the supporting material. Then review the tasks, accessing guidelines and tool mentors associated with each task as needed. You can also access the guidance provided by the practice directly through the Guidance folder.

For step-by-step instructions on how to adopt this practice, see How to adopt the Trust-based Interaction Design practice.

This material is made available under the Creative Commons—Attribution-ShareAlike License v3.0.

Parts of this material is based on the Eclipse Process Framework Practice Library, distributed under the Eclipse Public License V1.0.