Centralizada Se diseñará una base de datos en ACCESS que contenga toda la

Having explained how these different conflicts may occur within our multi-agent con- text, we will now proceed to detail a number of different approaches that agents may use to overcome them. Specifically, when an agent encounters a conflict with another it may choose one of three possible paths to overcome this conflict (refer to Section 1.2). First, it may choose to manage this conflict through a process of interaction, arguing and negotiating with its counterpart and thereby attempting to resolve their differences

1_{Theoretically, it is possible to introduce imperfections to all aspects of the agents’ knowledge (i.e.,}

the task parameters, the capability parameters, and the counterparts known within the society). However, since the objective of our experiments is to prove the concept of how arguments can resolve conflicts, in- stead of designing an exhaustive implementation with all possible imperfections and arguments, we chose to concentrate on resolving conflicts that arise due to imperfect knowledge about their social influences. In particular, we concentrate on the imperfections that arise due to the lack of knowledge about the first two premises in the schema Act(ai, ri)and Act(aj, rj)(refer to Section 3.1). Thus, conflicts may arise

due to the agents’ lack of knowledge about the role they and their counterparts enact within the society. Increasing the imperfections would merely increase the reasons why a conflict may occur, thus, bringing more arguments into play. However, this would have little bearing on the general pattern of the results. Section 4.2.2 presents a more detailed discussion on how such conflicts arise within the context.

and reach a mutual agreement. However, not all conflicts need to be resolved. There- fore, secondly, when faced with a conflict, an agent could find an alternative means to work around the situation; therebyevading the conflictrather than attempting to resolve it. Third, in addition to either evading the conflict or arguing and resolving it, an agent could also attempt to re-plan and alter the means by which it intends to achieve the objective so that the conflict situation is removed. The following explains how we map these three techniques into our argumentation context.

1. Argue: Use argumentation-based negotiation to resolve conflicts

In abstract, when an agent requires a capability from an acquaintance, it generates a proposal and forwards it to an agent that has that capability. Once received, the agent evaluates the proposal and decides whether to accept or reject it.2 The agent will then communicate its decision, either as an acceptance or as a rejection, to the original agent. If it decides to accept, the interaction ends in an agreement. However, if the decision is to reject, the onus is transferred back to the original buyer agent to generate and forward an alternative proposal. To help this interaction process, the seller agent will accompany its rejection with two additional forms of meta-information (arguments) that it will convey back to the original buyer agent:

• Reasons for refusal: This details the reason that prompted the refusal. In our system, seller agents reject due to two types of reasons. First, the agent may be fully committed to a prior arrangement in the requested time slot, so it returns an argument indicating that the reason for rejection is because it isunavailable

at that time slot (rather than the offer price being too low). Second, the offer value may not be sufficiently valuable to the agent, in which case it will return an argument accompanied with its rejection indicating the minimal threshold

that must be exceeded before the proposal will be considered. The return of such arguments should assist the buyer in its attempt to choose the next pro- posal to forward. For example, if the reason is unavailability, the buyer would not make an increased value proposal since doing so would be futile. On the other hand, if the threshold is returned as reason for refusal, the buyer can use this to gauge whether to make another proposal to that agent and if it does then it also indicates the value that should be used in such circumstances. These form of arguments are analogues to the types of meta-information exchanged in [Jung et al., 2001].

2_{Refer to Sections 4.2.4.1 and 4.2.4.2 for a detailed discussion on how agents generate and evaluate}

• Alternative suggestions:If the seller is willing to work for the suggested value of the offer, but not in the proposed time slot, it will send a number of its neighbouring time slots as alternative suggestions. These alternative suggestion provide additional meta-information to the buyer with respect to its current work schedule.3 _{This meta-information helps the buyer agent in finding agents}

for those future time slots. For example, assume that in the attempt to find a partner fort1, agent B indicates to A that it is willing to work for t2 as an

alternative. If agent A requires the same capability for the same price (the price offered when it got the alternative) int2, before requesting other random agents,

A will first ask B who has already expressed its willingness. Thus, alternatives provide agents with information about their partners’ schedules, which they will, in turn, use to selectively choose the sequence (instead of strictly adhering to a random one) in which they request their partners.

If any such proposal results in an agreement the argue method is said to have suc- ceeded in its objective. However, if all possible proposals fail to make an agreement, the argue process ends in failure.4

2. Evade: Find an alternative method to achieve the same plan

Unlike the previous method, here the agents do not attempt to use ABN to resolve their conflicts. The buyer agent will only make a single proposal. This is to estab- lish the willingness of the potential partner. If that offer is rejected the agent will not attempt to convince the non-willing partner, but will move on to the next known acquaintance, which has the required capability. However, in this scenario the buyer chooses to offer the maximum price it can in its single proposal. The rationale for this choice is to maximise the chances of success of its single proposal, thus this rep- resents the maximally effective evade strategy. Since the sellers are always motivated to accept higher offers (equation 4.2), making the highest offer possible maximises the chances of success in its single proposal. If the seller refuses this proposal the evade method fails. On the other hand, if it accepts, then the evade method succeeds.

3_{It is important to note that this information provides only a non-binding indication to the buyer about}

the seller’s current availability. Thus, in certain situations when the buyer decides to request it to act upon the suggestion, the seller may refuse to do so. This sort of situation arises when the state of the society changes during the time lapse (the lapse of time from the seller making the suggestion and the buyer deciding to request the seller to enact its suggestion). For example, another agent might have formed a more profitable agreement during this time or a change in the market conditions might have rendered the previous suggestion unprofitable.

4_{Clearly, this is toward the simpler end of the possibilities in argumentation. However, our purpose}

here is not to exhaustively cover all forms of argumentation. Rather we seek to evaluate the trade-offs involved in engaging in argumentation and concentrating on the simpler models provides an initial point of departure. In Chapter 6, where we considerhowagents can argue in a multi-agent context, we analyse a number of more complex methods of argumentation.

It is important to note that in this form of exchange the seller agent also does not provide any additional arguments explaining its reasons for refusal or suggesting al- ternative time slots to induce the buyer into an agreement. Thus, neither the buyer nor the seller attempt to argue.

3. Re-plan: Change the original plan

When a conflict arises at a particular time slot, the buyer agent simply places a delay slot in its schedule and tries to arrange for the desired capability to be scheduled to the next time slot. This delays the whole sequence of remaining activities and, thereby, extends the task’s overall duration by one time slot. While the argue and evade methods remain the main methods in our strategies, re-plan represents the fall back option (refer to Section 5.1). Thus, re-planning through delays (since theo- retically an agent can delay forever) will always ensure success in overcoming any specific conflict. However, delays may cause subsequent conflict situations to arise and will render the task less rewarding via penalties.

As introduced in Section 1.2, Even though our argue method can be effective at resolv- ing conflicts, there are a number of overheads associated with its use. In more detail, it takes time to persuade and convince an opponent to change its stance and yield to a less favourable agreement. Furthermore, it takes computational effort for both parties of the conflict to carry out the reasoning required to generate and select a set of convincing ar- guments, and to evaluate the incoming arguments and reason whether to accept or reject them. Thus, given these overheads of argumentation, and the alternative methods avail- able for overcoming conflicts (evade and re-plan), we believe it is important for agents to be able to weigh up the relative advantages and disadvantages of arguing, before at- tempting to resolve conflicts through argumentation. To this end, Chapter 5 presents an empirical study ofwhenand under what conditions agents should use argumentation techniques to resolve conflicts within a multi-agent context.

Now, in the event that agents do indeed choose to argue and manage their conflicts, the next important issue that comes into contention is how to argue. In other words, what issues must agents take into consideration and what strategies should they employ? To this end, Chapter 6 presents an empirical study on how agents can use our ABN framework specified in Chapter 3 to argue, negotiate and resolve the various different types of conflicts that may arise within the above context. Next, however, we present a more detailed system specification of our argumentation context explained above.