Another good question, thanks Alan.
Goals are not sufficient for autonomous agents. Another essential ingredient
of autonomy is beliefs. An autonomous agent acts on the grounds of own goals
and own beliefs. To state it differently, autonomous agenthood requires that
two filters are applied to external inputs: belief filter, which decides
whether and what input to accept as a believe, and goal filter, which
decides whether and which input to select as a goal.
ross
Il 23-01-2007 11:34, "Alan Penn" <[log in to unmask]> ha scritto:
> OK - so I take it that having a goal is necessary for agent autonomy. Is it
> sufficient?
> [Alan Penn]
>>
>> In the mail "Re: [SIMSOC] Newbie on the list - working on emergence of n",
>> Alan Penn wrote:
>>> A quick question. For an agent to be autonomous must it have a goal? In
>>> other words is it possible to imagine a simulation with autonomous social
>>> agents in which individual agents do not possess 'goals'.
>>
>>
>> Alan,
>>
>> But what would "autonomous" mean in that context ?
>>
>> Briefly, it seems to me impossible to define "autonomy" as an objective
>> and
>> absolute notion. One cannot be autonomous per se, but only with respect to
>> a given set of dependencies (relativity), and an observer (subjectivity).
>>
>> These dependencies can be broken down in two sub-categories : constraints
>> and objects. Constraints can be seen as the "laws" of the environment in
>> which the subject acts ("Autonomy is freedom under laws", Jean-Jacques
>> Rousseau), and may include as well other agents' actions. Objects are the
>> "things" with respect to which the subject (or the agent) can be described
>> as autonomous by the observer. And these "things" can either be goals (if
>> they are explicitly manipulated by the agent) or "tasks" (in which case,
>> the goal might be implicit and buried in the definition of the task, but
>> nonetheless still exist).
>>
>> Don't know if I made myself clear enough. Anyway, the sentence "this agent
>> is autonomous" (or not) does not possess any meaning by itself. The
>> correct
>> way to put it would be : "under these constraints, and with respect to
>> this
>> goal/task, this agent can be described by this observer as autonomous".
>> So,
>> defining agents, for example in a social simulation, as autonomous,
>> without
>> defining their goals appears to me as an ontological impossibility (but I
>> can be wrong).
>>
>> Cheers
>> Alexis
>
|