Aligning Law and Action - Chapter 7. Positioning agents

of 23

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
23 pages
0 downs
Aligning Law and Action - Chapter 7. Positioning agents
  Chapter 7 Positioning agents In order to model complex institutional scenarios (including non-compliance), thecomposition of mere institutional roles is not suf  fi cient. We need to enrich institu-tionalroleswithanintentionalcharacterization, soastocreateanexplicitconnectionbetween  motives  and institutional aspects. This requirement leads to an explorationof the practical reasoning of intentional agents. Interestingly, a similar analysis ap-plies to collective intentionality, and therefore to normative positions. 1 7.1 The “contract” for living In §4.4.3, we observed that the Hohfeldian potestative prism could be used to qual-ify aspects of physical power: power, negative power, and disability can be put incorrespondence to attraction, repulsion and independence. An ampliative questionnaturally follows: is there a more general analogy between the  correlativeness  of legalpositions holding between two parties, and the correlativeness of an agent with hisown environment? The latter is more fundamental because, even if we could imag-ine a situation in which no legal relation holds (e.g. being stranded on a uninhabited,uncharted island), we cannot imagine an agent not embedded in an environment. If such an analogy holds, then we should recognize for each institutional position acorresponding  agentive   position. Agentive potestative positions.  The exercise starts smoothly with the power po-sitions. As institutional ability (in the sense of legal capacity to modify the institu-tional reality) can be put in correspondence with physical ability, i.e. the ability of the agent to modify the environment, institutional disability can be associated withphysical disability (cf. §5.3.1). In this context, we will prefer the term  affordance   toability or power, in order to insist on the cognitive aspect of power rather than theformulation given in physics, engineering. 1 The material presented in this chapter re fi nes and extends elements presented in  [ Sileno, Boer,and van Engers, 2015c ] . 109  110  Chapter 7. Positioning agents perspective institutional position agentive positionperformer power affordancedisability disabilityrecipient liability susceptibilityimmunity no-susceptibility Table 7.1: Correspondences between institutional and agentive  potestative   positions Concerning the recipient perspective, depending on whether the agent does ordoes not react at some level to changes in the environment, we will refer to the po-sitions of susceptibility and no-susceptibility. 2 Table 7.1 illustrates these correspon-dences. Agentive teleological positions.  Unfortunately, mapping the positions of the fi rstHohfeldian square in agentive terms is less obvious. In these conditions, proceedingmetaphorically may be of some use.In which sense would the environment be in ‘duty’ towards an agent? Living inthe world, we base our conduct on many assumptions, often implicit, about howthe environment reacts to our actions. For instance, we all know that if we let goa stone from our hand, it will fall on the ground. We  expect   that, given a certaincondition, a certain consequence will hold. Such expectations do not bear the forceof necessity: things may go differently than what was expected, and this is analogousto legal claims. The existence and the possibility of capturing regularities in the realworld can be read as the presupposition to build a kind of ‘contract’ between agentand environment, which is necessary not only for rational behaviour (in the formsthatweknow), but for theveryprocess ofliving. Iftheenvironment wassuch thatnoexpectation could be formed, there would be no basis to construct cognition, neitherfrom an epistemic nor from an ontological, e.g. biological, point of view. 3 On the correlative side, what can we—in the role of observers—expect from anagent? Taking an  intentional stance  , we can attribute to the agent beliefs, desires, andintents; and considering him as a rational entity, i.e. one which “will act to further itsgoals in the light of its beliefs”  [ Dennett, 1987, p. 18 ] , we can construct, with an ade-quate approximation, what he will do next. If we inverse this point of view, entering 2 In §4.3.2, we have observed that the normal use of the notions of institutional power and liabil-ity is narrower than the general case, and we suggested the term susceptibility for the institutionalcounterpart as well, as a more general category than liability. 3 From an analytical point of view, cf. the cybernetic  law of requisite variety  [ Ashby, 1956, p. 206 ] :a system consisting of a controller and a controlled component can be stable only if the number of states of the controller is greater than the number of states of the controlled; in other words, if thecontroller cognitively mirrors what should be under control. From a probabilistic point of view, seeSolomonoff’s hypothesis  [ Solomonoff, 1997 ] : organic evolution resulted in a reasonably good initialbuilt-in probability distribution (possibly modi fi ed during life experience in higher living forms) thatorganisms use to make predictions and decisions, enabling them to sustain their living in the world.  7.1. The “contract” for living   111perspective institutional position positionbene fi ciary claim expectationno-claim no-expectationaddressee duty commitmentliberty no-commitment Table 7.2: Correspondences between institutional and agentive  teleological   positions the agent’s mind, the same mental states can be seen as manifestation of the driversforhisconduct. It isto  satisfy hisdesiresandintentsthattheagentbehavesinacertainway. 4 We call this category of generic motivational components ‘ commitments ’. 5 De-scribed in those terms, commitments and expectations share the  teleological   natureof duties and claims. Table 7.2 summarizes all the correspondences. Extrinsic and intrinsic components of agency.  Despite such analogies, the inher-ent distinction between institutional and agentive categories can be traced back toa dichotomy between  extrinsic   and  intrinsic   components of agency. Extrinsic com-ponents are the result of social, normative forces,  external   to the agent: he cannotchange those; or, better, he may change them only trough intervening at the sociallevel, thus requiring adequate recognition from the social body. This does not meanthat there is a relation of control from social to individual. The agent may be notaware of certain norms, or he may still neglect or overlook them. In contrast, thesrcin of intrinsic components is  internal   to the agent, i.e. it is the result of internalprocessing and deliberation. Proposition 11.  Where institutional positions identify extrinsic commitments, expec-tations, abilities and susceptibilities, agentive positions identify intrinsic commitments,expectations, abilities and susceptibilities. Agent-based modeling revisited.  This epistemological leap provides us with an al-ternativestartingpointforagentmodeling, whichbringstwoimportantinnovations.First, proceeding similarly to the analysis resulting in the Hohfeldian prisms, we in-troduce the negative characterizations of agentive positions, which, to our knowl- 4 It is not important to acknowledge whether such mental states actually exist or are epiphenomenaof other unconscious activities of the brain. The principle of responsibility in our legal systems isbased on the assumption that they are signi fi cant to reason about human behaviour. 5 In Latin,  committere   means initiating an action (e.g.  committo me itineri ,  proelium committo ).The term has been also chosen to make explicit the connection with the notion of   social commitment  :an agent socially commits to another agent to perform a certain action if the  fi rst agent intends to doso and the second one has interests in this intention. Duty can be seen as the legal aspect of socialcommitment. Similar connections can be found in the works of e.g. Castelfranchi  [ 1995 ]  and Singh [ 1999 ] , distinguishing between  internal  ,  social   and  collective   commitments.  112  Chapter 7. Positioning agents edge, no one has yet explored in the literature. Second, we draw attention to sus-ceptibilities, which are generally neglected or given as implicit in current agent archi-tectures. Both aspects go towards requirements of behavioural modeling which arehighly relevant in the legal activity. For the  fi rst aspect, consider  negative action —inthe two forms of lack of action or actively preventing another outcome— and howit is relevant to attribute responsibility (see e.g. Lehmann, Breuker, and Brouwer [ 2004 ] , elaborating on Hart and Honoré  [ 1985 ] ). For the second aspect, susceptibil-ities play the role of anchors for reactivity, and, for these reason, correctly designingsusceptibilities is part of a general discourse about  evidence  . 7.2 Representational requirements Consideringacertainaction(e.g. “XpaysYacertainamount”)wemaythinkofcasesin which this action is missing (“X doesn’t pay Y”) or performed differently (“X paysY another amount”). Similarly, considering a certain intent (e.g. “X wants to payY”), we may consider cases in which this intent has not been formed (“X doesn’twant to pay Y”), which include cases in which there are  contrastive intents  (“X wantsto avoid paying Y”, “X wants to stop paying Y”). As these examples show, ‘negative’characterizations of action and of intent exist and we actually use them. They alsoplay an important role in the attribution of responsibility: they can clearly be usedto address a misalignment with expected behaviour (what  should   have been done),and more generally, with normative goals. But these are just two of the dimensionson which negation introduces new perspectives to the matter. A mythological example.  Concerned as we are in this chapter with intentionalcomponents, we don’t need to limit ourselves to legal scenarios. Mythology, forinstance, is a rich repository of stories constructed on interesting intentional con fi g-urations. For example, Even if he knew that all sailors who had done it went lost into the open sea,Ulysses wanted to hear the voices of the Sirens. To achieve this goal, the saildirection set, he put some wax in his companions’ ears and asked them to bindhim to the mainmast with the strongest rope. He also ordered not to follow anyof his requests before reaching their destination. Ulysses eventually succeeded in his goal of hearing the Sirens, and, listening to thisstory, people are able to grasp why. The scenario refers to notions such as:  con-ditional persistent commitment   (Ulysses desiring to jump off towards the Sirens af-ter hearing their voices, and insisting on trying even though strongly bound to themainmast);  positive expectation  (about the fact that the Sirens were along that spe-ci fi c path);  negative affordance   (associated with the plan preventing the Sirens’ effect); disability  (Ulysses bound to the mast);  negative susceptibility  (the sailors to Ulysses’srequests);  no-susceptibility  (the sailors to Sirens’ voices). Only the fi rst two have more  7.2. Representational requirements  113direct correspondences to the components used in most known  belief-desire-intention (BDI) architectures. The purpose of the present chapter is to identify and considerthese ‘neglected’  positions  as  fi rst-class citizens.As we saw with the Hohfeldian prisms ( §4.4), two types of negation are at stake(carrying  negative   and  null   polarizations). Furthermore, beyond an informationaldimension (beliefs, knowledge, expectations), and a teleological dimension (desires,intentions, commitments), Ulysses’s story proves that to fully capture a model of anagent’s behaviour we also have to consider ascription of affordances (as abilities of intervention on the environment) and of susceptibilities (as abilities of sensing theenvironment, operationaling corresponding reactions). How do we deal with thesecategories? What constructs are required to express a story like the one given abovein agent-based modeling? 7.2.1 Assessment of existing agent-based architectures In the last decades, software engineering has been moving from machine-orientedviews of programming towards concepts and abstractions that more closely re fl ectthe way in which humans conceive the world. In particular,  multi-agent systems (MAS) have been introduced to develop and implement distributed algorithms andcomplexservices, characterizedbydiffuseinteractionsbetweenindividualautonomousentities  [ Wooldridge and Jennings, 1995b ] . Then, exploiting  theories of mind   elabo-rated in psychology, philosophy, and cognitive sciences  [ Bratman, 1987; Dennett,1987 ] , paradigms such as  belief-desire-intention  (BDI) have been taken as a base to im-plementagentsexhibiting rationalbehaviour  , byusingthesamecategoriesthatweusetypicallytoaddresshumanbehaviour. 6 AfastglanceoverMASdesignmethodologies(e.g. GAIA  [ Zambonelli, Jennings, and Wooldridge, 2003 ] ) makes already explicitthat this  fi eld is primarily concerned with arti fi cial computational systems. Never-theless, because their functioning mirrors architectures inspired by theories of mind,they could, in principle, be applied to model human behaviour; for instance, simi-lar models have started to be exploited in  intention recognition  applications  [ Sadri,2012 ] .To ensure a practical operational result to our research, rather than diving intothe several semantics (and variations) presented in the literature, 7 we decided to startfrom existing agent platforms, assessing from the ‘inside’ the modeling tool box pro-vided by the underlying agent language. We have considered  fi ve agent-based lan-guages / platforms, three based on a BDI architecture and two on logic programming.Most of these, namely 2APL  [ Dastani, 2008 ] , GOAL  [ Hindriks, 2009b ] , ALP [ Kowalski and Sadri, 1999 ]  and DALI  [ Costantini and Tocchio, 2008 ] , refer only to default negation : when the agent is not able to conclude a certain fact, then that fact isfalse. Without the difference between  null   polarity and  negative   polarity, all negative 6 A recent overview can be found in  [ Baldoni et al., 2010 ] . 7 As observed by Neumann  [ 2010 ]  for normative architectures, the number of conceptually ori-ented articles far exceeds the number of existing execution models.
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks