Trust, Control and Autonomy: A Dialectic Relationship - Trust Theory: A Socio-Cognitive and Computational Model - page 203

7.2.1 The Notion of Autonomy in Collaboration
For the purpose of this topic we use a practical and not very general notion of autonomy. 18 In
particular, we refer to the social autonomy in a collaborative relationship among agents. We
distinguish between:

a meta-level autonomy that denotes how much the agent is able and in condition to negotiate
over the delegation or indeed to change it (in this regard, a slave, for example, is not
autonomous: he cannot negotiate or refuse);
a realization autonomy , that means that the agent has some discretion in finding a solution
to an assigned problem, or a plan for an assigned goal.

Both are forms of goal-autonomy, the former at the higher level, the latter at the sub-goals
(instrumental) level. For definition of different kinds of autonomy, including some of the
dimensions we consider, see also (Huhns and Singh, 1997).
The lower the control of the client/trustor (monitoring or intervention) on the execution,
the more autonomous is the contractor. In this context then, autonomy means the possibility
of displaying or providing an unexpected behavior (including refusal) that departs from the
requested (agreed upon or not) behavior. The autonomous agent can be either entitled or not
to perform such an unexpected behavior . 19

7.2.2 Delegation/Adoption Theory
We introduced the delegation notion in Section 2.9. Here we use that concept, integrating it
with the notion of adoption and developing the theory of adjustable autonomy.

18 We do not consider here some important aspects of autonomy (that could be adjusted) like the agent's indepen-
dence or self-sufficiency. For an analytical discussion on the notion of autonomy in agents and for a more principled
definition, see (Martin and Barber, 1996), (Castelfranchi, 2000b), and (Castelfranchi, 1995).
19 In this topic we do not discuss in detail another very important distinction between:
being practically in condition of doing something (refusing, negotiating, changing and doing something else), i.e.
what we would like to call practical possibility ;and
being deontically in condition of doing something, i.e. to be entitled, permitted in the strong sense, i.e. the deontic
possibility

.

An agent can have the former without the latter, or vice versa (see (Castelfranchi, 2000b)). In fact, there are two kinds
of lack of power (hence, of dependence and autonomy): one based on practical conditions, the other based on deontic
conditions. In deontic autonomy, an agent is permitted to do/decide/ interpret/ infer/ etc. Not only is it practically able
and in condition to, but it can do this without violating a social or legal norm, or the user/designer prescriptions. As
there are two kinds of autonomy there are two kinds of 'empowerment' (giving autonomy): deontic empowerment
versus practical, material empowerment (Jones and Sergot, 1996). Therefore, an additional dimension of adjustment
should be taken into account that is, the deontic one. The delegator (or the delegee) can attempt to modify (restrict
or enlarge) either what the delegee is practically and actually able to do independently of the other, or what it is
entitled to do. For example when a delegee re-starts negotiation, instead of directly modifying the task, it is implicitly
asking some sort of permission, or agreement. Obviously enough, in strong delegation (contract relation, see later)
the assignment of a task τ to the delegee implicitly entails giving it the permission to do τ . Adjusting the entitled
space of freedom, or adjusting the practical space of freedom, is an interesting difference, but we cannot examine it
in this topic. Notice that this theory would imply the same plan-based dimensions of delegation and help.