Page 79 - Trust in ICT 2017

Trust in ICT 1

• Decision Theoretic Approaches: These approaches are built upon a strong foundation of probability
theory and so their trust evaluations are compatible with standard statistical decision theory. That
is an agent which can calculate its expected utility directly using the output of the model.
8.3.2 Develop a decision making algorithm for policy decision and enforcement [81]

• Exploration and Threshold
Griffiths et al. [82] employ a simple, threshold-based decision-making model. To this end, they
define the concepts of untrust and undistrust (in addition to trust and distrust) to represent the
notions that a degree of trust may be insufficient for deciding to delegate (or not, in the case of
undistrust). Agents who are ‘untrusted’ are only considered for interaction if no explicitly trusted
alternatives are available. The eventual decision to interact is made if the degree of trust exceeds a
pre-defined threshold, provided by the system designer. In initial cases, the authors require that all
trustors participate in a ‘bootstrapping’ phase of a fixed duration, whereby agents explore the
society before beginning to use their trust models. While the particular exploration strategy is not
discussed, Griffiths states that any partner has an equal chance of being selected during the
bootstrapping phase.
The SULTAN model was developed primarily with a view to supporting secure interactions in
internet applications, in the domain of trust management. These works can be distinguished from
other works by their focus on security and implement ability within enterprise systems. Typical
decisions necessitating trust, in this context, may be the decision to allow a user access to a sensitive
or restricted system resource, or the decision to accept a user’s authorisation key. Trust is generally
specified as rules (or policies) provided by users, stating the preconditions of trust. By taking a
probabilistic view of the possible contingencies, the authors quantify risks in terms of Expected Loss
and Maximum Allowable Loss. The decision to trust is made using the policies together with a risk
threshold, here an interaction will be considered too risky.

The FIRE model Huynh et al. employ a more sophisticated variant of the most-trusted strategy for
selecting interaction partners which includes exploration. The decision mechanism of FIRE consists
of two stages, and can be summarised as follows. The set of potential partners is initially divided
into two subsets, based on the ability of the trustor to produce evaluations for those partners. These
sets are termed hasTrustValue and noTrustValue. The most trusted candidate from the
hasTrustValue is advanced to the exploration stage. In this secondary stage, the Boltzmann
exploration strategy is used to make a decision between selecting the most trusted agent, or a
random one from the noTrustValue set. In this model, the trustor always chooses to delegate.

The Boltzmann exploration strategy is useful for decision-making when nothing is known about the
candidate set. Given an agent has a choice between a number of actions (i.e. delegation candidates)
(a1, a2, …, an) with expected utilities (u1, u2, …, un), the Boltzmann strategy assigns a probability to
each action according to the distribution in the equation:

• Decision Theoretic Approaches
Matt et al. present an approach which combines probabilistic measures of trustworthiness within
the context of a logical argumentation framework. In this work, the authors assume the existence
of contracts which specify certain guarantees about the interaction outcomes that can be expected.
The probabilistic representation of trust is based on the model of Yu and Singh. An agent deliberates
by advancing arguments regarding service parameters (e.g. reliability, security) that either attack or
support a proposition T, representing the assertion that a particular trustee is trustworthy. A second
kind of argument (called a mitigation argument) attacks contract arguments that support T. These
arguments represent claims that a particular agent usually violates a contract clause which supports
T. The decision to trust is eventually made on the basis of whether the proposition T is supported
beyond some cautiousness parameter, which is equivalent to the trusting threshold of Yu and Singh.

71