A different approach to parameters

The framework introduced in this thesis to model dynamic communication with unreliable agents was based on some predefined parameters. We for instance assumed that the network is asynchronous, meaning that agents do not observe calls that are being made in which they are not involved, and we explored two different attitudes of agents: the presumption of innocence– and the see for yourself attitude. We also say that agents with this first attitude are naive.

Additionally, we assumed that in calls only information regarding phone numbers and secrets is shared.

In this section we will approach the parameters differently. We will explore the idea that the parameters are treated as dynamic quantities rather than static or the idea to require agents to share more information in calls.

Dynamic parameters

Treating the parameters as dynamic quantities is motivated by the fact that it may be unrealistic to assume that the parameters do not change while the network evolves. In our framework, agents have a certain type and property and follow a certain, not necessarily the same, protocol. Until now, we have treated these characteristics of agents as static, but what happens if agents change their behavior based on the information available about the network? For instance, if agents find out that all their connections are unreliable, they might decide to start following a different protocol or take on certain properties. In particular, she might decide that she will change her behavior toward asubset of the agents in the network—namely toward those agents that she considers unreliable.

These are two ideas: parameters such as agent types, properties or protocols are dynamic so that they can change during the execution of a call sequence, and agents behave differently toward different agents. That is, an agentain a networkG= (A, R, f, N, S) may be described by the way she acts upon another agentb. That means that for each agent a∈A, a can be described by a set of types{tb

a}, properties{propba}and protocols {Pab} wheretba,propba andPab denote the type, property and protocol agentaacts out toward agentb

By treating the parameters as dynamic, real-life situations can be better approached. This is because it is quite natural for agents to change their behavior once unreliability is detected. Note though that also here, changing your behavior by taking on the property of being (strong) Trumpian may not be the desirable approach to deal with unreliability with regards to the network as a whole.

In order to account for dynamic parameters, we need to change the data struc- ture accordingly so that each agent is described by the triple ({tb

a},{propba},{Pab}). It then remains to define how agents change their behavior, with options ranging from doing so depending on your own information or randomly by means of a coin toss.

Communicated parameters

Another approach to the parameters of dynamic communication is by requiring more information to be shared in calls. In addition to sharing phone numbers and secrets, we can require agents to share agent types, properties and protocols. This allows agents to calculate individually which calls are likely to take place based on their knowledge about the network. If in addition we let agents communicate the calls they have been involved in in the past and the calls that they heard to have taken place, we create a situation in which the agents “gossip about gossip”. That is, the subject to the gossip is the gossip itself. Intuitively, this

will make it much easier for agents to observe unreliable agents. This is because when all information about calls and the information shared in these calls is shared, each agent can individually determine the setsZa of agents that are considered unreliable by an agenta. We will elaborate on “gossip about gossip” in Section 7.2.4 where we discuss the application of this idea as an alternative to blockchain technology.

Requiring agents to share more information is a very realistic requirement. In communication, we might namely want more information than just phone numbers and secrets in order to decide whether another agent is reliable. In particular, we could let agents decide themselves what information they want to share.

Generally, we expect that storing more information about the network will make it harder for agents to be unreliable without being detected. Therefore this approach is also interesting to investigate from a security point of view. Research Questions:

• Are there easy adaptations of our framework so that the parameters can be treated dynamically or more information is required to be shared?

• In our framework we consider that agents only share information about phone numbers and secrets. The question is now whether including more information to be shared makes it harder for the unreliable agents to remain uncovered. Or do then more sophisticated unreliable agents emerge?