A Fuzzy Implementation for the Socio-Cognitive Approach to Trust - Trust Theory: A Socio-Cognitive and Computational Model - page 329

how much time is spent in analyzing offers and delegating, i.e. how much offers an agent
collects before choosing. 9
While in the preceding experiments the agents collected all the offers before deciding, here
an agent can delegate when it wants, saving time. This situation is closer to a real MAS
situation, where agents act in real time and sometimes do not even know how many agents will
offer help. How much time is spent in delegation depends on the strategy and on simulation
constraints. The random trustor can always choose the first offer it has, so it results in being
the quickest in all cases. If there is a fixed number of agents and the guarantee that all of
them will offer, best ability, best willingness, the cognitive trustors and the statistical trustor
can build and use an ordered list of the agents: so they have to wait until the offer from the
pre-selected agent arrives. In the more interesting MAS scenario, without a fixed number of
offering agents, each incoming offer has to be analyzed and compared with the others. In
order to avoid waiting ad infinitum, a maximum number of offers (or a maximum time) has
to be set.
However, in this scenario there can be other interesting strategies, such as the first trustful
trustor, who does not wait until all six offers are collected, but delegates when the first 'good
offer' (over a certain threshold) is met; this can lead to more or less time saved, depending
on the threshold. Here we present the results of EXP7 (250 simulations, 100 tasks); in this
case all agents wait for exactly six offers (and compare them) before delegating, except for
the random trustor (who always delegates to the first one) and the first trustful trustor who
delegates to the first one that is over a fixed threshold. Figure 11.14 and Figure 11.15 show
the results for credits (as usual) and time spent (analyzed offers).
The first trustworthy trustor still performs better than the statistical trustor, saving a lot of
time. Depending on the situation, there can be many ways of aggregating data about credits and
time. For example, in a limited time situation agents will privilege quickness over accurateness;
on the contrary, in a situation with many agents and no particular constraints over time it would
be better to take a larger amount of time before delegating.
Experiments with Costs
In our simulations we assume that the costs (for delegation, for performing tasks, etc.) are
always the same; in the future it would be interesting to introduce explicit 'costs' for operations,
in order to better model real world situations (e.g. higher costs for more skilled agents).
The costs are introduced as follows. When an agent sends their 'proposal' they quote a price
that averages to their ability and willingness (from 0 to 90 Credits). 10 Half the cost is paid
on delegation, the second half is paid only if the task is successfully performed. If a task is
re-delegated, the delegator has to pay again to delegat. For each successfully performed task
the delegator gains 100 credits. There may in the future even be costs for receiving the reports,
but at the moment this does not happen.
In order to model an agent who delegates taking into account the gain (and not the number
of achieved tasks) we have used a very simple utility function, that simply multiplies trust
and (potential) gains minus costs. It has to be noticed that on average a better agent has a
9 There are other possible parameters, such as time spent in reasoning or in performing a task. However, we have
chosen only the parameter which is more related to the Delegation Strategy; the other ones are assumed to have fixed
values.
10 We generate randomized values over a bell curve which averages it (ability + willingness)/2.