Principal and Helper: Notes on Reflex Responsibility in MAS

What justifies -in the head of another agent different from the one actingthe obligation to compensate is the fact that the principal agent has lengthen its own action through the implementation of a foreign activity for its own interests. We present two basic modal operators for representing, respectively, intentions in the interest of another agent and agency in the interest of another agent. They appear useful enough for characterizing the notion of reflex responsibility in a multi-modal multi-agent system (MAS) context. 1 MOTIVATION AND AIMS As pointed out by Chopra and White [1], theorizing in domains such as legal and cognitive status of agents is crucial for designers of agents, especially, for the design of “on demand” MAS. Within such engineering account, a legal question arises: do the designed agents are to be autonomous enough to have rights and responsibilities? (By being autonomous we at least mean that agents act to achieve their own goals cf. Conte and Castelfranchi [2].) Most works on the topic are centered on “contractual issues” (see e.g. [3,4]). Chopra and White point out four approaches as one moves up the sophistication scale of agents: three “weak” positions, based on (i) the idea of agents as mere tools of their operators, (ii) the unilateral offer doctrine (a contract formed by a party ́s offer plus an acceptance, stipulated in the offer), and (iii) the objective theory of contractual intention (a contract –usually wordsis an obligation which is law to the parties, who have the intention to agree), plus a fourth, radical one, which involves treating artificial agents as the legal agents of their operators. There is also a fifth position that postulates the legal systems treating agents as legal persons. In this work we focus on the legal binding between a principal agent and a dependent agent. Particularly, we are interested on the dependent’s performance that has its origin in extra contractual situations e.g. factual and/or occasional situations, trust, or courtesy. Examples of such bindings occur e.g. between the owner of a car -or any other deviceand the one who drives it with the owner ́s authorization (and without a proper title for using it), blog activities such as twitting in the name of another, or bidding in an auction in the interest of another: performed in the interest of a principal agent. In these situations is enough that the principal wills to be bind to third parties through the helper’s or dependent’s performance. 1 Universidad Nacional de La Plata, and FACEI, Universidad Católica de La Plata, Argentina. [email protected]. We therefore keep apart from our analysis situations in which performance in the interest of another has a contractual basis. This because, in contracts, function for another one and subordination may be rather straightforward to identify, mainly because there is a notion of obligation involved. If an agent gives some explicit orders or instructions to another agent which acts as his helper, or if and agent is obliged through a contract in the interest of another agent (even when agents voluntarily engage into contracts because of their own utility), or if an agent h forms part of agent p’s business organization, subordination is somehow established. We therefore exclude here cases such as mandates and any conferral of a power of representation accompanied by an obligation of representation in certain ways (e.g. a cheque is a mandate from the customer to its bank to pay the sum in question.) We give a definition for the concept of reflex responsibility between a principal agent and a helper agent, mainly inspired on general provisions settled by Italian and Argentinean provisions for persons. We indeed use the terms “does”, “performance”, and “action” as referring to persons, although it is not entirely clear for us if it is meaningful to speak of the actions of devices, and artificial agents within highly automates systems; possibly a term as “executes” sounds more suitable. In what follows, “does” has the usual expected anthropomorphic meaning. The definitions we give may be useful as a step towards a specific notion of reflex responsibility of artificial agents. Article 1113 of the Argentinean Civil Code states that “The obligation of the one who caused damage is extended to the damages caused by those under his dependence.” In its turn, art. 1228 of the Italian Civil Code settles that “except a different will of the parties, the debtor who profits from the work of a third party for fulfilling the obligation is responsible for malicious or negligent facts carried out by that third party.” According to general doctrine and jurisprudence related to such articles, reflex responsibility has a subjective basis. This is a reason that makes reflex responsibility challenging to represent: if it had an objective basis, checking the standard legal extremes would be sufficient. Requisites for reflex responsibility are: i) the existence of a dependence relationship between principal and helper (or dependent), ii) the successful performance of an illegal action carried out by the helper, iii) that such performance was carried out while exercising a subordinate incumbency, iv) that such performance provoked a damage or injury to a third party, and that v) there must be an efficient causal relation between the helper’s act and the damage caused. Regarding the formal framework, we use as a basis a BDI multiagent context for dealing with agents’ attitudes, extended with generic obligations, as in [5]. A = {x, y, z...} is a finite set of agents, and P = {p, q, r, ...} is a countable set of propositions. Complex expressions are formed syntactically from these, plus the following unary modalities, in the usual way: Goalx A is used to mean that “agent x has goal A”, where A is a proposition. Propositions reflect particular state-of-affairs cf. B. DunninKepliçz and R. Verbrugge [6]. Intx A is meant to stand for “agent x has the intention to make A true”. The doxastic (or epistemic) modality Belx A represents that “agent x has the belief that A”. The deontic operator O represents generic (legal/lawful) obligations, meaning “it is obligatory that” [7]. The operator Doesx A represents successful agency in the sense given by D. Elgesem, i.e. agent x indeed brings about A [8]. For simplicity, we assume that in expressions like Doesx A, A denotes behavioral actions concerning only single conducts of agents such as withdrawal, inform, purchase, payment, etc. (i.e. no modalized formulas occur in the scope of a Does.) As classically established, Goal is a Kn operator, while Int and Bel are, respectively, KDn and KD45n. O is taken to be a classical KD operator. These are all normal modalities. The logic of Does, instead, is non-normal [8,9]. The rest of the paper is organized as follows. Section 2 addresses one possible characterization for a certain notion of dependence which happens to be complex enough and central to the lawful concept of reflex responsibility we deal with. We attempt four subsequent definitions, each of which improves the previous one. We go through them by using several examples. A relativized modality is introduced for dealing with oriented, coordinated intentions: an agent intends to become true a state-ofaffairs A in the interest of another agent. We introduce in Section 3 another modality, a directed agency operator that binds a helper agent h to the principal agent p, and to the “oriented” action or situation A that h carries out in the interest of p. In Section 4 we formally define reflex responsibility of p regarding h with respect to an action or state-of-affairs A when: there is dependence between p and h w.r.t. A, h succeeds on carrying out A on account of p, such action constitutes an illegal act, and there is a damage a third agent t suffers which is attributable to h ́s performance of A on account of p. Section 5 presents the underlying logical structure and the corresponding semantics. Conclusions end the paper.