State 3: The agent is known and trusted
8.8 Additional Future Work
During the course of this thesis, there have been many places where future work has been mentioned. This chapter has discussed much of that work. There are omissions, however. In the text, there have been places where we discussed problems with, for example, the current formalism, which would have to be addressed in the future. Rather than repeat these discussions, table 8.1 provides a page reference to each aspect.
Page No. Description
64 Fix problem of zero importance. 80 Strategies for altering trust. 81 Trust in DAI agents.
105 Agent-subjective estimates in complex situations. 113 Estimations of utility.
120 Fixing evolution problem in PlayGround. 122 Societies of communicating agents. 125 Formation of societies of agents.
Table 8.1: Other work for the future
8.9
Summary
The formalisation of trust presented in this work is not complete. It is a piece which is in flux, and this is as it should be (Popper, 1967; Popper, 1969). In addition to being able to work further on the formalism itself, we have identified several areas where a knowledge of, or use of trust would be practicable. Amongst these are CSCW and DAI.
The identification of these extra avenues for work illustrate the strength of the formalisation, but also the all-pervasive nature of trust. It is, indeed, present in all systems which have a social aspect. Whether or not an explicit understanding of it is necessary depends on the system, but in order to determine that, the system has to be addressed.
The next chapter draws some general conclusions and summarises the work that has been presented here.
Conclusions
9.1
Introduction
The thesis introduced a formalism for trust which was useful in clarifying discussion of the concept, and was extensible to take into account further work in the area. In addition, the formalism was implementable, and provides the basis for the first implementation of trust in an intelligent artificial agent. Further, experiments were carried out using the implemented formalism. The following sections briefly summarise the chapters which make valuable research contributions.
9.2
The Formalism
The introduction of a formalism for trust is of unquestionable worth in the under- standing of the concept. To date, discussions of trust have suffered from vagueness and the lack of an agreed definition. The lack of a definition has sprung from the lack of an agreed method for discussion. The formalism presents a means of establishing a clear, precise, and easily understood language for that discussion. It does not stop at discussions of trust, however. The methodology used in the thesis — that of a top- down approach to the development of a simple formalism — can be of use in many spheres of sociality, where human emotions play a part. This is because, although we may be unable to deconstruct the precise method by which, for example, trusting decisions are made, we can provide a suitably close approximation to the final result which enables us to provide a path back to the initial considerations made by the agent. The formalism presented in chapters 4 and 5 allow such work to be carried out.
Chapter 6 substantiates the claim that the formalism can be used to discuss the concept of trust by providing several brief discussions of several aspects of the concept, such as transitivity and rationality.
9.3
Implementation
A major contribution of the thesis is the presentation of a formalism which isimple- mentable. The formalism itself provides a tool to social scientists for the discussion of
the concept. From the point of view of further work in DAI and intelligent systems, an important aspect is its inherent ease of implementation. Being based on simple linear mathematics, it provides the ideal tool for artificial agents in making reasoned decisions about the world in which they exist. Since trust has proved so successful for human agents, there is no reason to assume that it should not prove equally successful for artificial agents (Marsh, 1992).
Chapter 7 provides a discussion of implementations of the formalism, and experi- ments which have been performed on a simple testbed using the Prisoners’ Dilemma. Whilst acknowledging the limitations of the Prisoners’ Dilemma as a tool for simulat- ing ‘real life’ (Argyle, 1991), the findings of the chapter are exciting and illuminating. The most important contribution of this chapter is proof that the formalism can be implemented successfully, and that artificial agents can make trusting decisions. This is important because it allows us to study their behaviour under various conditions, with various other agents (as in Axelrod’s PD tournament (Axelrod, 1984)) and as- certain the true utility of trusting decisions in agents.
9.4
Further Work
Whilst the formalism presented here is of use in itself, particularly with regard to artificial implementations, it is of great importance as an indicator of work which could be done. There is much more work which can be done with the idea of trust discussed in the thesis, particularly as regards using the formalism, or developments of it, in other social spheres such as CSCW, HCI and intelligent networks.
Chapter 8 presents detailed discussions of many aspects of much of the further work which is possible, both using the formalism and using a knowledge of the workings, or the concept, of trust. One of these areas, that of CSCW, has already proved fruitful in terms of the applicability of the formalism, and work is ongoing in this area (Thimbleby et al., 1994; Joneset al., 1994).