A static theory of promises

A static Theory of Promises Jan A. Bergstra ∗ Informatics Institute, Uni versity of A msterdam Science P ark 904, 1098 XH Amsterdam The Netherlands Mark Burgess † V isiting Informatics Institute, Uni versity of Amsterdam Abstract W e discus s the conce pt of promises within a frame work that can be a pplied to either humans or technolo gy . W e compare promises to the m ore establish ed notion of oblig ations and ﬁ nd promise s to be both simpler and more eff ecti ve at reduci ng uncerta inty about multi-ag ent beha viou r . Contents 1 Intr oduction 2 2 A motivation 3 2.1 Promises about a piece of technology . . . . . . . . . . . . . . . . . 4 2.2 Range of promiser types . . . . . . . . . . . . . . . . . . . . . . . . 5 2.3 Promises and reduction of uncertainty . . . . . . . . . . . . . . . . . 5 2.4 The logical status of promises: trust/reputati on backed action . . . . . 6 2.5 Extending promise to prediction . . . . . . . . . . . . . . . . . . . . 7 3 The concept of a pr omise 7 3.1 Implicit promise versus e xplicit promise . . . . . . . . . . . . . . . . 8 3.2 Existing views on promises . . . . . . . . . . . . . . . . . . . . . . . 8 3.3 Promises in distributed computing . . . . . . . . . . . . . . . . . . . 9 3.4 Promises subject to product/process ambiguity? . . . . . . . . . . . . 10 ∗ Email: j.a.ber gstra@uva.n l,janaldertb @gmail.com . † Mark Burgess was supp orted by th e EC IST -EMANI CS Ne twork of Ex cellence (#2 6854) . Email: mark.burgess @cfengine.c om 1 4 Modeling the structur e of a promise 10 4.1 Promises as documented apparent intentions . . . . . . . . . . . . . . 12 4.2 Intentions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 4.3 Commitments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.4 Expressing intentions . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.5 Promises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 4.6 Deceptions – non-intended promises . . . . . . . . . . . . . . . . . . 18 4.7 Positive promises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 4.8 Promises as speciﬁcations in technology . . . . . . . . . . . . . . . . 19 4.9 A technical reﬁnement of the example . . . . . . . . . . . . . . . . . 20 4.10 About the philosophical literature on promises . . . . . . . . . . . . . 21 4.10.1 Promises in the philosoph y of Law . . . . . . . . . . . . . . . 22 4.10.2 Promising in anthropology . . . . . . . . . . . . . . . . . . . 22 5 Pr omise valuation, assessment, reasoning, and logic 22 5.1 Promise v alue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 5.2 Promise strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 5.3 Assessments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 5.4 Promise related reasoning processes . . . . . . . . . . . . . . . . . . 24 5.5 Promise related informal logic . . . . . . . . . . . . . . . . . . . . . 25 5.5.1 Speciﬁc links between promises and informal logic . . . . . . 26 5.5.2 Nonlogical processing of promise bodies . . . . . . . . . . . 27 6 Pr omises versus obligations 27 6.1 Obligation strength . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 6.2 Potential prominence of promises . . . . . . . . . . . . . . . . . . . 29 6.3 For and against the primacy of obligations . . . . . . . . . . . . . . . 30 6.3.1 In fa v our of obli gations . . . . . . . . . . . . . . . . . . . . . 30 6.3.2 Against obligations . . . . . . . . . . . . . . . . . . . . . . . 30 6.3.3 Promises precede obligations ; locality precedes globality . . . 32 7 Conclusions 32 1 Introduction This p aper 1 is about the usefulness and imp ortance of promises as a concept. For something that is so abundant in daily life, the concept of a promise has been giv en 1 This paper is a revision of the p aper arXiv ed on http:// arxiv.org/ab s/0810.3294 v4 . In this version re marks on the in stitution of prom ising in the context of Philosop hy of Law hav e been added, as well as a reference to work on the an thropo logy of promising . V arious small mistakes were remedied as well. 2 only a limited treatment in academic literature, and then mainly in the areas of philos- ophy , law , and economi cs. Promises give way more often to th e idea of ob ligations, indeed many authors quickl y do away wi th promises and replace them with apparent obligations as if thi s were a necessit y . No doubt this tendency orig inates i n hi storical matters, but it is unfortunate as promis es hav e a plaus ible and practical value both i n their mundane meaning for humans a nd in a n extended interpretation that can a pply to machines and other inanimate objects. The concept of a prom ise is not a difﬁcult one – in fact we claim it to be a si mpler concept th an an obligation – but it requires some care, and it leads to some compli- cations. In this paper we p resent ou r understanding of promises and explain why in some appl ication areas there are both practical and theoretical advantages to t heir use over obligations. T o make our case unamb iguously , we ask the forbearance of readers in presenting the fundamentals and motiv ations a t som e length. W e assume the existence of an emerging topic of promise theory , to which we hope to make a con tribution with our paper . The development of promise theory is stil in an initial phase. W e position promise theory as a branch of informal logic. W e menti on [23, 20, 5] and [18] as i ntroductions to i nformal lo gic and [39] for an example of ho w informal logic may be applied in computing. Promise theory has bee n speciﬁcally de veloped with appl ications i n distri buted systems in mind, but not limited to applications in informatics. Promise theory may th erefore be viewed alt ernativ ely as a topic in theoretical i n- formatics wi th foundation s in informal l ogic and with a p otential for applicati on in management sci ence. Promise theory presuppo ses a theory of agents. It draws u pon the classical noti on of reciprocity , that is agents mod eling one another recursiv ely (see [16]). For the notion of an agent we refer to [26]. 2 A motivation For surely all readers the no tion of a prom ise will be q uite familiar . Our e xperience of promi ses might be both posit iv e and ne gative, for m any promises that are made are nev er ke pt. Below are examples of the kinds of statements we intend to call promises. Let us begin wit h everyday statements and prog ress gradually to t he kin ds of abstract promises that we would like to use in a variety of technical scenarios. • I prom ise you that I will walk the dog. • I prom ise you that I fed your cat while you were a way . • W e promise to accept cash payment s. • W e promise to accept validated c redit cards. • I prom ise to lock the door when I lea ve. 3 • I prom ise not to lock the door when I lea ve. These examples are quite uncont rover sial. They are easily found in every day life, spoken by humans or posted on sig ns. W e no w want to ar gue that it is useful to e xtend the notion of promises to inanim ate o bjects that have been designed or programmed to beha ve i n a certain manner . Thi s is not a very large s tep, but it is easier to m ake with some examples. Consider the following promises that might be made in the world of Information T echnology: • The Internet Service Provider promises to deliver broadband Internet at a ﬁxed speed for a ﬁxed monthly payment. • The security ofﬁcer promi ses that the system wi ll conform to security require- ments. • The s upport personnel promise to be a vailable by pager 24 hours a day . • Support staff promise to reply to queries within 24 hours. Again these are straightforward prom ises, which c ould be described further t o be more speciﬁc. The ﬁnal promi se could als o be restated in m ore abstract terms, t ransferring the promise to an abstract entity: “the help desk”: • The com pany help-desk promises to reply to service requests within 24 hours. 2.1 Promises about a p iece of t echnology This latter example i llustrates t he way that we transfer the intenti ons of promises to ‘entities’ that we consider to be respon sible by association. It is a small step from th is transference to a more general assignm ent of prom ises to individual component s in a piece of technology . For example, we can document t he properti es of the fol lowing tools and technologies in the spirit of this argument: • I am a logic gate and promise to transform a TRUE si gnal i nto a FALSE signal and vice versa. • I am a command line interpreter and promise to accept input and execute com - mands from the user . • I am a router and p romise to accept packets from a list of authorized IP addresses. • I am a compliance monit or and promise t o verify and automatically repair the state of the system based on this description of system conﬁguration and policy . 4 • I am a high av ailabili ty server and I promise you service del iv ery with 99.9999% a vailability . From these example we see that the range of appl ication of p romises is quite gen- eral. Indeed such promises are all around us in everyday life, both in mu ndane cloth - ing as well as in technical disciplines. Statements about engineering speciﬁcations can also proﬁtably be consi dered as promises, even though we might not ordinaril y thi nk of them in this way . When an electronics eng ineer l ooks i n a component catalogu e and sees ‘resistors’ for sale promi sing to have resist ance of 500 Ohm s to with in a tolerance of 5 %, we do not ar gue about who made this promi se or whether the resistor is capable of i nde- pendent th ought. The coloured bands on t he com ponent are a sufﬁcient expression o f this promis e, and we accept i t by association. By this reasoning, we propose that the concept of a promise should be understood in a way which allows fo r all of these uses. 2.2 Range of promiser types W ith the expansion of the use of promise as just adv ocated goes a rather lar ge range of types of promisers that we will admit, we mention: • Obj ects (tools, instruments) without intelligence, • Non -human animate agents, • Intell igent artiﬁcial agents, • Hum an agents. The deﬁnition s belo w are int ended to apply wit h intellig ent artiﬁcial agents 2 acting as promisers and a lso in the case of promising hum an agents. The extension of t hese deﬁ- nitions to the cases of ob jects without intelligence and to non-human ani mate agents is not carried out in detail, b ut the idea is that in such c ases an intelligent agent (eith er hu- man or a rtiﬁcial) is imagi ned which promises on behalf of these objects or agents . The details of the concept of promising o n behalf of another entity will not be consi dered below . 3 2.3 Promises and r eduction of uncertainty Promises are enti ties that we use to form expectations of the behaviour of all manner of things. Such expectations cont ribute to reducing our uncertai nty about their behaviour , 2 About appro priate deﬁnition s of the co ncept o f an (inte lligent) ag ents a signiﬁcant literature has been accumulated in the ﬁeld of artiﬁcial intelligence. 3 Dev elopin g a deﬁnition of promise that applies uniformly for a wide range of promiser types is left as a concep tual problem for futur e w ork. 5 and thi s can apply as much to techno logy as to hum ans. W e therefore take it as giv en that the concept of a promis e is a useful one, althou gh s till in need of a st able deﬁni- tion t hat is suit able as a basis for promis e theory , and w e consider next h ow one can formalize promises in the simplest and least assuming way . Producing a promis e may be more effecti ve in reducing uncertainty than putti ng forward an assertion that is stated wit h more certaint y . Indeed i f an expectation about a piece of technology or about an hu man or artiﬁcial agent is asserted with absol ute certainty , or merely wit h some quantiﬁed probabili ty o f being valid, the question im- mediately arises ho w that kno wledge has been obtained, t hereby pos sibly increasing uncertainty rather than reducing it. Such existential questions do not arise wi thin a commu nity of users for a piece of technology that has been delivered in combination wit h a bundle of promises to its user , who may simply react to disappointin g performance with a reduction o f trust in the promiser . Future promises from that same so urce will be recei ved with less optimis m. Con ve rsely , if the piece of equipment outperforms the promi sed performance that fact may lead to increased trust in the original promiser , causi ng increased conﬁdence in forthcoming promises by the same agent. 2.4 The logical status of pr omises : trust/rep utation backed action Deﬁning what a prom ise is can hardly be do ne witho ut examining deﬁnitions or at least approximate descripti ons of se veral similar notions. Related b ut d iffe rent notions are these: • T aking a decision (role backed action). • Belief based predicti on (a belief backed action). • Science b ased prediction (a science backed action). • Ann ouncing a legal cl aim (legal theory and case history backed action). • Ann ouncing a mathematical conjecture (reputation/trust backed action). • Speciﬁcation of properti es of an artifac t (reputation /trust backed ac tion). • Exp ression of a mathematical fa ct (a lo gic backed ac tion). • Exp ression of a quantiﬁed expectation (re putation /trust backed ac tion). A promise is an action wi th a primary and a s econdary ef fect. T he primary ef fect concerns prom isees and other observers who m ay adapt their actions and plans in accordance with becoming aware of the cont ent of a promise. A s econdary effec t reﬂects u pon a promiser and impacts on the t rust or reput ation that the promiser enjo ys among those who no ticed a promise being issu ed. Promising is an action backed by credibility , and trus t, the latter mediated by reputation. 6 Other forms of backing for an action or an utt erance exist as well: if a mathe- matical theorem with it s proof is communicated and the theorem is found to lead to inconsistencies with existing theory that fact will im pact the trust one has i n the proof rather than in the agent who puts forward the proof. Similarly if a sci entiﬁc claim is found to contradict ne w ﬁndings the science that justiﬁes the claim will be scrutinized once more, rather than its originator . 4 The s eparation of a not ion of promise from related noti ons li ke prediction (wit h and without quantiﬁed expectation), conjecture, legal claim, and decision will require sy s- tematic work that depends on th e deﬁnitions of tho se other not ions. That d ependency makes it infeasible to analyze these matters in det ail within our p aper . An exception is t he noti on of a prediction. Predictions can be understood liberally (e.g. predictions made by artiﬁcial agents) just as promises. A prediction diff ers from a promise in that its ass essment inv olves no past, present, or future activity or p articipation of its origi- nator . A predictio n cannot create an obligation for its creator to comply with the body of the prediction, t hough issuing a prediction may create a responsibility to rem edy expected adverse consequences of the predicted event or state of affairs. Predictions that don’t materialize are li kely to be held against the reputation of their origi nators, howe ver . 2.5 Extending promise to pr ediction When designi ng a theory of promi ses m eant for use in informatics some dis crepancy with t he concept of promise as used in th at theory and the collo quial (or rather philo- sophical) use of the term prom ise may be acceptable. Once a t heory of promis es takes a formali zed form dedicated to a s peciﬁc application area, such as for i nstance distributed computing, the emergence of a gap between a mainst ream informal under- standing o f promise and its meaning in that particular theory is almos t unav oidable. W ith this in mi nd it may be acceptable in principl e that a theory of promis es includ es includes predictions under prom ises as well in spit e o f t he fact th at a meaningful dis- tinction can be made. In this way a to ur opera tor can promise (rather than predict) a client that a trip ne xt week will proﬁt from good whether conditions, 3 The concept of a pr omise Because of their ove rriding ubiquity , and practical importance, one would like to ha ve an a ccount of promises t hat ca ptures their key properties a nd explains related concepts 4 Of co urse rep eated ﬁnding s of p roblems with mathematical o r scientiﬁc results of the same agent will in v oke questions about that agent’ s credibility , and so will the discovery of methodo logical ﬂaws in that agen ts working proc esses. The po int we make is th at in lo gic o r scien ce backed ac tion th e logical or scientiﬁc backin g of an action itself can be scrutinized , an op tion that is a bsent with trust/rep utation backed actions. 7 such as commitment , o bligation and in tention. There is a s urprising lack of discussion about the meaning of promises in the li terature as far as w e can t ell. Although the concept or its relatives hav e been mentioned in such diverse areas from l ogic, law and philosophy t o economics, information science and com puting, there is no agreement on what constit utes the semanti c cont ent of the terms or if there is even more than a tacit relationship between promise, commi tment, obli gation etc. The most attention has b een giv en to the concept of obligatio ns especially in the area of d eontic logic. W e believ e on the other hand that th e philosophical implications of promises are far wider than is generally assum ed and that there is both a need and a practical i mportance to clarify them once and for all. Indeed, we will show that notwith standing ou r lengthy deﬁnitions the concept of a promise is simpler than that of an obligation . 3.1 Implicit promise v ersus explicit pr omise By far t he m ost ubiqui tous usage of the term prom ise is in ph rases like “the promise of s olar energy” or “the promis e of nano-techno logy”. Such prom ises are impli cit i n the sense that the existence of a promis er is not assumed and the word promise means approximately: expectation of usabili ty and impact, where an author using the term usually intends to contribute to an assessment of the mentioned expectation. In contrast an explicit promise comes about from an actio n performed by a promiser in an appropriate con text. W e will on ly focus on explicit p romises below and we wi ll assume by default that promises are e xplicit . 3.2 Existing views on pr omises W e wi ll survey some vi e ws that oth er authors have put forward o n promises, henceforth understood as explicit promises. Atiyah [1] suggests that any p romise leads to an obli gation to kee p that prom ise that is motiv ated by the threat of tit for tat reprisals. Reciprocation is thus coupled to the idea of promi ses i mmediately , which seems to h op ove r fundamental deﬁnitio ns directly t o a discus sion of the economics of keeping promises. The obligati ons are to a void injury and to reciprocate goodwi ll. It might be discus sed whether incentiv es are the same as obligations. At iyah points out howe ver that promisin g som ething cannot be necessarily used to create obligation at wil l. Promises might cause obli gations but they can als o represent obl igations that already exist, i .e. to show commi tment to an existing obligation to pay th e price o f s omething. e.g. I promise to pay the bearer the sum of 1 pound (in gold). This is only an existing admis sion o f m oral obligation . Atiyah maintains, plausibly , that the motiv ation for promising has changed throug hout history . When people m ake promises, th eir intentions are culturally bound. Thus a V ictorian gentlem an’ s conception of a promise might not ﬁt with that of a present-day child who promises to be home in time for dinner . Cartwright takes Atiyah’ s view and asks what might be the point of promi ses if not to generate t he assumed obligation [10]. Wh y do p eople bot her to m ake promises 8 about thin gs to which they are already obliged? His answer incl udes the i dea that it is a f ace-sa ving measure: to mitigate their humility , suggesti ng that an obligation is interpreted as a kind o f att ack or levy o f force? Alternativ ely , p erhaps t he obligati on to keep one’ s promi ses weighs heavier t han the o riginal obligati on (I promise you my word as a gentleman not to kill you, even though the law s ays I am forbidden). Referring to Fried [14 ], Cartwright points out t hat the economics o f contractual ti t- for -tat su ggested by Atiyah is tied to promises and no t to the ob ligations they m ight confer . The idea that prom ises are an economic d riv er of contracts or agreements as b ilat- eral e xchanges of promises is continued in the w ork of Gilbert [15]. Then Carrillo and De watripont have argued that promises can best be understood as a market m echanism for reducing the uncertaint y in a mo ral-hazard gam e [9]. Thi s work does not seem to hav e b een pu rsued. Does a prom ise in crease the likelihood of voluntary cooperation? A numb er of ot her works ment ion the concept of promi ses in t he context of game theory also. In t hese, t he concept of a promis e is tacitly assumed to be related to the probability of choosing a particular game strategy . Scanlon [30] meti culously analyses how and t o w hat extent prom ises give rise to obligations under a variety of combi nations of addit ional assumptions . In h is analysis morality plays a important rol e and it is impli citly ass umed that prom iser and promisee are human beings capable of moral reﬂection. Zhao et al. [40] p rovide a com prehensiv e modal logic incorporating b eliefs, ca- pabilities and p romises. Unfortunately i t is dif ﬁcult from that work to extant a clear intuition of the concept of a promise that th e authors had in mind. It s eems that this dif- ﬁculty is in part caused by the form alist approach tak en. In Framinan and Leisten [13] order promis ing is dis played as a standard technical term in industrial workﬂow man- agement, whil e at the sam e time that u se of th e term promis e is considered lacking a sufﬁ ciently clear deﬁnition. Promises are o ften t reated in the context of agreements, see for instance Schein- man [31]. Agreements may be constructed out of mutual promises. But promises may exist without being part of agree ments. 3.3 Promises in di stributed computing More recently , a different m otiv ation for promises was introduced by Bur gess in the context of dist ributed m anagement [6]. Rather t han focusing on m orals or even eco- nomics as the principal moti vator , Burgess uses the promi se as a measure of ‘v oluntary cooperation’ as a way o f c ircumventing what we con sider to be fundamental problems with logics of obl igation for determining system behaviour . V oluntary cooperation is seen as a way of simplifyi ng constraints and av oiding many-worlds paradoxes. He pursues the argument further by emphasizing the role of auto nomy of the parts, and ar- gues for a ‘promi se t heory’ in which e very com ponent in a system that can have uni que information or independent action should be viewed as axiomatically autonomous [ 7]. Any coop eration or even s ubordination of th e parts t hat comes about in an o r ganized 9 system m ust then be understo od as the result of ‘ho nouring’ pu rely voluntary promises to do so. Bur gess argues th at no matter what one beli e ves about the power to oblige (e ven so ldiers can refuse to follow orders), voluntary cooperation can be used as a prag- matic eng ineering m ethodology for mappi ng out the com plexity of a control problem in a way that is in v ariant with respect to ce ntralization or decentralization o f systems. In computer science, particularly the ﬁeld of Mu lti-Agent System s th e concept of commitments has been used for some time [37, 3 8]. It h as b een sug gested that promises and com mitments are the same. Howe ver , we s hall show that this is not the case. More seriously , the sense in which the term commitm ent is used in such discussions is more stylized than pu rposely cons idered and can only beneﬁt from the di scussion in this paper . 3.4 Promises subject to pr oduct/p rocess ambiguity? Issuing a promis e i s an eve nt, and one may say t hat A promised when A issued a promise. More common is to view a prom ise as t he outcome, that is a description of what h as been prom ised, rather than as an event. Nevertheless to some degree the notion of a promise is subject to p roduct/process ambiguity as described in [17] and further references cited in that paper . Although [17] discusses the noti on of an ar gument, its conclusions may hol d in the case o f p romising as well. In particular the result of issuing a promise m ay be th e awareness that a promis e has been issued, which is rather remote form a con ventional understanding of “product”. The ambig uity is perhaps even deeper because after h a ving been i ssued a prom ise creates a process of credibi lity and trust assessment and various other reasoning processes, of whi ch a textual d escription of “what has been prom ised” cons titutes merely an abstraction. If aw areness is understood as a p rocess rather t han as a state the matter becomes quite comparable to the case of an ar gument. 4 Modeling the structure of a pr omise Consider the foll owing intuitive idea of what a promi se mig ht be: A pr omise is an announcement of f act or behaviour b y a promiser to a pr o misee, observed by a number of witnesses (r eferr ed to as the scope of the pr omise), whose out come has yet to be assessed. The promiser and promisee are b oth as sumed to be ‘agents’, i.e. h umans or inan- imate objects to which we attach id entity in the story of prom ises. This general de- scription ﬁts the examples that we have already given and give s some clues as to the constituti on of a prom ise, but it also o pens up a n umber of qu estions that need an- swering. Already we can see that thi s apparently basic deﬁnition rests on a num ber of assumption s: that we can observe th e outcomes of beha viours and that the outcome of a prom ise is clear at some single m oment o f time in the future, to be measured and v er- iﬁed by an observer . A full account of this mig ht includ e a theory of measurement, but 10 we wish to a void thi s lev el o f detail as it b inds us to too many details that ha ve nothing directly to do with the is sue. Let us ins tead t ry t o understand to essential characteristics for promises and con sider what d istinguis hes a promise from related matters, such as obligations , com mitments and other terms. The model world in which we fo rmulate promises must have the follo wing charac- teristics. • There mu st be agents in order for promises to exist. • There mu st be a promiser (or source agent). • There m ust be a prom isee (or recipient agent) which mig ht be t he same as the source. • There mu st be a body which describes the nature of the promise. W e might summarize these attrib utes with a notation as in [6]: pr omiser body − → pr omisee (1) • W e can leav e the body unspeciﬁed, but it must consist of a quality (a type, topic or sub ject for the promise) and a quantiﬁer (which indicates how m uch o f the realm o f possibi lity for that subject is being promi sed). For example: promis e quality: “travel to work”, promise quantity “on Monday and Friday each week”. Finally , what is im plicit in the above is that a promise requires the transmis sion of a message, or at least d ocumentation in s ome kind of phy sical form, e.g. a speech act, or a writ ten statement, else it cannot be made known to anyone except the promi ser . A promise must therefore ha ve documentation that is made intentionally or otherwise. What t hen is a promise before we write it do wn? W e s hall refer to this as a po ssible intention . An intenti on is th e basi c formu lation of a course of behaviour , whi ch is made internally by (or on behalf of) an age nt. When an intention is made public, it becomes a promise. If an intention is documented or leaked in some way then anyone has a right to assume it is a promise. W e take it as giv en that t here has to be a source for e very promi se. A p romiser does not hav e to reve al its identity of cou rse, so wit nesses to the promise migh t not know its source e.g. consider the anonymous threat. There is no reason to deny the existence of a source ho wev er . The lack of such information about a promiser is simply a defect in the k nowledge of the receiving agent, but on e would normally prefer to assume a consistent picture of prom ises and i nfer the e x istence of an anonymous prom iser . This justiﬁes our postulati ng the source. 11 4.1 Promises as documented appar ent intentions Stoljar [34] writ es that a promise is an announcement of an intention. and he argues that any offer must be a promise as we ll. Indeed a key characteristic of a prom ise is that it documents an intentio n, so let us explore the i dea of intentions in more det ail. In- tentions turn out to be a lo west common denominator for all of the concepts discussed in thi s paper and thu s have a special i mportance. Unfort unately from a do cument one cannot conclude to wh at extent the intention expressed in it is re al. That indeterminac y giv es rise to an additional complexity of ou r deﬁnitions. Since prom ises in volve communication we require a no tion of the sp read of i nfor- mation amongst the agents. W e use the term scope (as used in com puter science) for this. A scope is si mply deﬁned to b e a set of agents. For example, t he scope of a promise would typi cally refer to the promiser and a list of witnesses to the p romise, e.g. those who heard to utterance or those who saw the written document. 4.2 Intentions In t he realm of all pos sible formu lations about agent behaviour t he concept of an in- tention stands out as an important foundation. T erm 1 (Curr ent intention of an agent A ) A curr ent intention of an age nt A i s de- scription of a possibl e behaviour , or goal, or obj ective, or st ate of affairs, that is con- templated by A with the understanding that i t can be and pre ferably (for A ) will be br oug ht to r eali zation. T erm 2 (P ossible intention for an agent A ) A possible intentio n fo r an agent A is a description of a possible behavi our , o r goa l, or objective, or state o f affairs, tha t may but need n ot curr ently (at the time o f quali fying th e descrip tion) be contemplat ed or pr eferably brought to r ealization b y A , and which might be in som e (poss ibly differ ent) cir cumsta nces a curr ent intention of A . Obviously a current intention is also a possible in tention. But if an ut terance of A announces a possible intention that need not be a current intention, it m ay only appear to be a current intention . The compon ents of an int ention are as foll ows: a source agent who formulates the intention, a t ar get agent if the in tention is directed at a potenti al sub ject, and a body whi ch explains th e quali ty and quanti ty of the intention. Only the s ource of an intention knows about the int ention, i .e. the scop e of an intenti on is the source only . There are no witnesses. Now we must be careful: t he set of all possi ble intention s should be distinguis hed from actual instances of intentio ns selected b y an particular agent. W e s hall so me- times use the phrase “pos sible intention s” t o mean th is full set of abs tract entities to emphasize when we wish to signi fy a general des cription of beha viour rather than an individual agent’ s decisio n. 12 4.3 Commitments T o commit to something is t o make a decision in fa vour of it. Th e issue of comm it- ments is therefore about the fav ourizati on of intentio ns. Comm itment is a personal decision and has nothing to do with physical representation or communication, thus the issue of commi tment precedes any discussi on of promis es. A commitm ent has a source, a t ar get and a body , i.e. it is an intention . Like an in tention it has no p hysical representation and does not hav e a non-tri vial scope. Once an i ntention becomes a commit ment we often assume th at som e poin t of no return has been passed i n the act of commi tting (decidin g) about the particular in ten- tion. i.e. adding the int ention to a list o f commitments. For example, in a game of chicken in which two cars drive towards each oth er to see which one will swerve off ﬁrst, a drive r has com mitted t o not swerving wh en the decisi on to not back down has been m ade [32]. This m ight hav e certain irrev ersible consequences, b ut it i s difﬁcult to generalize the idea of irrev ersibilit y in examples o f this kind. What comm itment essentially boils down to is the elev ation of some intention beyond an arbitrary thresh- old. In other words, in th e universe of i ntentions there is a subset of t hese which we may call commitments . T erm 3 (Commitment) Commit ments ar e curr ent intentions that we ar e committ ed to. W e may call them intended intentions, or equivalently r eal i ntentions, intent ions that we hol d, or committed i ntentions. The commi tment o f an i ntention exceeds its mer ely being curr ent in that it is s table and p ersists i n t ime until some achievement of the intention will take pl ace or until some overruling con siderations in validate the commitment. When passing by a shop one may feel a current intenti on to buy a n ice g adget, only to b e reliev ed of th at intenti on (or rather i ts currency) after noticin g its price. If howe ver , t he pri ce is quit e go od, but th e shop is closed at th e ti me of passing along, then a current intent ion t o buy the same i tem can become activa ted with the statu s of a commi tment, only t o be termi nated when th e i tem has been acquired or w hen unexpected problems turn out to stand in the w ay . 4.4 Expressing in tentions Making intentions k nown to other agents is the essence (meta-intention) of promising. W e will dev elop some terminology for such acts o f expression. An intention is not necessarily announced by the agent h olding it to any other agents. Indeed, we may now deﬁne any intent ion that is announced to be a promise. Con versely we no tice that any prom ise t hat has not been announced i s merely a pos sible intention. Some intentions are desirable while others are absolutely und esirable and an agent m ight nev er choose them, y et they are possible i ntentions ne vertheless. The fac t that such beha viours can be intended is enough for them to qualify as possibl e intentions. 13 Intentions must always be tho ught of as belo nging to a speciﬁc agent. Th ose in- tentions which are actual plans of the agent are ca lled its commitm ents. Other phrases for a commitment that we may use are: intended intenti on, or real intention. Due to the static nature of our account we pay no attention to the process by which an intention might become a commitment or vice versa. T erm 4 (Intention utterance) An age nt a pr oduces an intention utterance if A pr o- duces an e xpr essi on of a description of a possible intentio n. What m atters for our d iscussion on promises is intention utterances that seem to be real. That leads to the idea of an apparent intention utterance. T erm 5 (A ppare nt Intention utterance) An utt erance expr essing a poss ible inten- tion (of a prin cipal agent) with the contextual appearance of an i ntention. Appare nt intentions, may be drawn fr om the following rang e: Real intention: (alterna tively: commit ment, true in tention, or intended intent ion) what is announced corre sponds to wha the agent expects that will happen, or that (s)he will do, or what holds or what will hold. In other wor ds the appar ent intention is re al if it is a c ommitment (and ther efor e curr ent). Incidental intention: (alt ernatively: non-committing curre nt intent ion) what is an- nounced corr esponds to why the agent e xpects t hat will happen, or that (s)he will do, or what holds or what will hold, b ut only a the time of expr ess ion. Indiff erent intention: (alterna tively: quasi-i ntention) t he issuer h as no current inten- tion corr espon ding to the utterance , and no curr ent conﬂicting int ention either . An indiffer ent intention is curr ently contemplated as a possi ble behaviour , goal, objective, or stat e of affairs, b ut its bri nging about is not pr eferred, and thus an indiffer ent intention is not a curr ent intention . Deceptive intention: (also: mis leading i ntention) the an nouncement might s eem to be r eal fo r an audience in scope b ut it is a lie fr om the perspective of the pr omiser . A deceptive intention is incompatibl e r e gar d ing r ealization with a curr ent inten- tion, though this may be only known to the principal agent. In valid intention: (alternatively: mani fest lie) all obs ervers may notice a discr epancy between what is announced and the facts. The in vali dity of an in valid intention will become clear to agents in scope of that utterance. 14 The id ea of an apparent intention is that at face value it is like an intention from the perspecti ve of an observer but there is a c onsiderable de gree of freedom in connec- tion with a so-called underlying intent ion, the existence of which we postulate in t he following deﬁnition. T erm 6 (Underlying intention (of an appar ent i ntention utterance) ) Given an ap- par ent intent ion utt erance of an agent, ther e is an underlying intention (whic h need not be comprised i n the same utterance) as well. W e will distingu ish ﬁve cases, corr e- sponding to the case distinction of intention utterances: Real intention: The underl ying intention of a r eal intention is that same intenti on. Incidental intention: The underlying intention of an i ncidental intention is that same intention which is known to be consistent but is non-committing as well. Indiffer ent intention: The underlyin g intention of an indiffer ent intention is empty . Deceptiv e intention: The underlying intention of a deceptive intention differ s signiﬁ- cantly fr om the (deceptive) intention . In valid intention: The underl ying i ntention o f an in val id i ntention di ffer s noticeably (for observing agents) fr om the (in valid) intention. W e wil l assume the agents keep underlying intentio ns pri vate. Otherwise new le v- els of complexity emer ge as un derlying intent ions may turn out to split over the s ame distinctio n of four cases recursiv ely . 4.5 Promises A promise i s the physical publication of an int ention within a certain scop e. This suggests that there must be so me agent to observe the prom ise and i ts outcome wh ich in turn requires t he expensive not ion of a t heory of obs erv ation so we sh all tackle th is issue separately (see section 5.3). T erm 7 (Pr omise) A promise is an apparent intention of an a gent, (the pr omiser or pr omi sing agent) t he u tterance of which has been documented within a s cope that goes be yon d the pr omiser . Accor ding t o the deﬁnit ion of intenti on ut terances, a pr om ise bri ngs with it an appar ent intention and an underlying intention, and ﬁve cases can be disting uished for pr omis es: r eal, incidental, indiffer ent, deceptive, and in valid. Promises th us hav e scope. Formally intenti ons also have a scope, but the scope of an i ntention held by an agent is by deﬁnition limited to the agent (source) itself. An intention could b e leaked deliberately (e.g. to t he p ress, in order to inﬂuence someone). This might be a form o f le verage, or an attempt to impose an oblig ation on some party 15 in scope. Howe ver , at the i nstant an i ntention expands in scope to encom pass more agents it becom es a promise. A so-called letter of i ntent, for example, is a promise rather than merely an intention. The time aspect of promises presents further challenges. Intentio ns can b ecome outdated b y e vents. An event which is found to fulﬁll an in tention documents the implicit promise, since one must admi t t o the in tention in a wider scope. Con versely , the documentation for a promise does n ot hav e to l ast for ev er; if documentation of a promise disappears completely , it reverts to being an intention. A promis e to oneself is in excess of a mere intention bec ause i t has been documented. The distinction between the promisee and the scope of the prom ise is key t o under- standing promises. Suppose someone intends to arrange a surprise birthd ay party for their friend. Initially this i s an intention . The int ention is written in a diary or m en- tioned to a third p arty and it becomes a promise. The target is not in th e scope o f the promise howe ver , so t he promi se remains un known to the jubil ant. Ho wev er , su ppose that before telling anyone el se, the promiser destroys all evidence of the prom ise by tearing out the page of the diary , effecti vely withdrawing knowledge of the promi se, then the promise rev erts to being just an in tention. But as long as knowledge of the promise remains “out there” in the world, it remains a promise that has been made. T erm 8 (Kee ping a promise (r elative to an observing agent)) A pr omi se i s kept, fo r the perspective of an agent A (e.g. the promisee), if an action is p erformed or a state of affairs is r eached th at complies, accor ding t o A ’ s assessment o r observati on, with the body o f the pr omise. It is plau sible but not strictly needed to r equ ir e that once a pr omi se is k ept some causal r elat ion with pr omi ser behaviour can be found. Although in a static theory of promises t he keeping of prom ises is not center -stage, having some some characterization of it is helpful. Criti cal to the deﬁnition of keeping a promise i s relativity w .r .t. the observing agent. If a promi ser assesses a promise p as having b een kept while a promis ee B , p erhaps supported by other agent s in scope of the promis e doesn’t agree, that state of af fairs casts n o d oubt on the status of the promise p as such. On the contrary thi s situation may hav e been precisely what A had in mind when issuing promise p . T erm 9 (Kee ping a promise (in absolute terms)) A pr omise is kept in ab solute terms, if it is kept accor ding to all agents in its scope . T erm 10 (Br oken pr omise) A pr omis e is br oken i f after it has been issued a state is r eached a t which the pr omise ha s not been kept, fr om any conceivable ( though r easo n- able) perspective, and fr om which it will certain ly n ot be kept anymor e either , again fr om any conceivable (t hough r easonab le) p erspective that an agent in its scope mint have. W e will now discuss the relation between promises and obligation in some detail. 16 T erm 11 (Pr omissory obligatio n) W ith each pr o mise of an agent A an obligat ion is connected, the so-called pr o missory obliga tion. It is that obligation to w hich the a gent has become obliged by making the pr omise. Promissory oblig ations are an i mportant tenet of the philosophy of prom ises, and we do no t deny t heir existence. This d eﬁnition al lows signiﬁcant freedom. It can be read as follows: a promi se theory wil l allow for a not ion o f a prom issory obligation . That notion may be doe weak that it provides empty promisso ry oblig ations onl y . It is no t assumed that a promissory obligation e xplains a promise in terms of a philosophi cally grounded concept of an obli gation. Howe ver , we will oppose to what we call obli ga- tionism. T erm 12 (Obligationism) W i th obl igationis m we denote th e viewpoint that (i) promises ar e characterized by a u nique capacity t o (auto)generate an obli gation (speciﬁcally t he pr omi ssory obligation) for t he pr omising agent, and tha t (ii) t he essence or content of a pr omi se is fully captur ed by its pr omis sory obligation. An argument in fav o r of obligationi sm is found if on e assumes that t he promi ssory obligation captures the essence o f the obligation s o that the concept of a promise can be reduced to that of an obli gation. Ob ligations being known from d eontic log ic and form various strands of phi losophy , on e may consider this reduction to constitu te a worthwhile simpl iﬁcation rendering the cons truct o f a promise redundant. W e don’ t support this argument. T erm 13 (Non-obligationism) W ith non-obligat ionism we denote the belief that obli- gationism is false. T erm 14 (Str ong non-obligationism) W ith str ong n on-obligati onism we denote the belief that ob ligationi sm is false and th at in addi tion the concept of pr omis e may be accounted for without making use of the concept of an obligation . Preferring non-obligati onism ove r ob ligationism , and convinced of t he relev ance of strong non-obli gationism, we will propose a strongly non-obl igationist conceptio n of promises. Arguments for (strong) non-obligationism ha ve been detailed in [6]. T erm 15 (Restricted str ong non-obligatio nism) W ith r estricted non-ob ligationi sm we denote the belief that ob ligationi sm i s fa lse, and that for a lar ge class of pr omises, though not f or all, the concept of pr omise may be accounted for without making use of the concept of an obligation. If i n court a witness prom ises to tell the truth and noth ing b ut the truth, an obligation is created. Remarkably that ob ligation can only be created by m eans of a promise. Restricted strong non-obl igationism does n ot claim that t his p romise, though b eing prior to it s promissory obligati on, can be fully un derstood without contempl ating the concept of an obligatio n. 17 A recent argument can be found in [3] where payments in a peer -to-peer system for ﬁnancial transfer an s tore of value are considered autonomous actions to the ex- tent that such payments cannot be ob liged by deﬁnition. Payments can be promised howe ver , wit h th e correspon ding p romissory ob ligation cons tituting no more t han ac- quiring some certainty that th e payment could be effec ted if th e agent wishes to do so. 4.6 Deceptions – non-intended promises Although we hav e di stinguis hed ﬁ ve cases for promis es, comm itments stand out as most important im mediately fol lowed by deceptions, with the ot her classes entering the picture in order to hav e a complete story . W e add some furt her comments on deceptions. Understanding deceptions (or lies) is als o an im portant step in clarifyi ng t he rela- tionship betw een int entions and p romises, because it is po ssible for an agent to h a ve two diff erent i ntentions in pl ay at the s ame tim e: a commitm ent and an announced intention (i.e. a promise) w hich are n ot com patible. Incompatibi lity means that striv- ing for bot h int entions sim ultaneously i s fruitl ess because their realizatio ns cannot be combined. In a deception, there is always a source and alwa ys a t ar get a nd the target cannot be the same as the source, as an agent cannot (int entionally) deceiv e its elf. Furthermore, we m aintain that the target of a d eception mus t be in scope, so there m ust be a physical documentation and hence a deception necessarily in volves a promise and not merely an intention. T erm 16 (Deception) A deception consist s of two intentions: a documented intenti on (i.e. a pr omise) and a non-documented commitment, which ar e incompatible. The non-documented commitment will also be called the hidden intention. In a deception the hi dden int ention is more important than t he witnessed on e one and we might refer to it as the domin ant intent ion. This simply expresses that it is a commitment while the promise contains merely a “poss ible int ention”. It is the r eal intention of the agent (“intended in tention”), while the intent ion in the promise can merely be described as non-r eal . If the dom inant intenti on should be rescinded, a deception will re vert to being a promise, b ut this is only known to the source. 4.7 Positiv e pr o mises A deception is the augment ation of a prom ise wit h an incompatibl e i ntention. W e should l ike a name for the augmentation of a promis e wit h a positive intention. W e might call this a positively intended promise, or more brieﬂy a posit iv e promise. From these slightly strained terms, we can no w appreciate wh y t he concept of a promise is i n fact so important. A promise is simply a p romise (th e documentat ion of 18 an i ntention), regardless of what lies behind it. Any internal priorities or considerations are hid den from the v iew of other agent s and cannot be ob served. Thus, promises are an in dependently import ant concept because we can (indeed must) talk a bout promises without discussing the basis on which they are made. When a promis e is issued, we are neither required nor able to confront the trut h or falsity of the promise. Indeed, as soon as we ask su ch qu estions, new issues su ch as trust a nd a pl ethora of other subjecti ve issues come int o p lay . Such issues are probably un-resolvable in a logical sense. Howe ver , what we assume is here that no matter ho w trustworthy a promise m ight be, it can increa se or dec rease our certainty of a promised outcome and thus it bears an inﬂuence . 4.8 Promises as speciﬁcations in technology Suppose a user commu nity t akes an int erest i n m achines M th at com pute a function f from a ﬁnite domain D to a ﬁnite range R , with the only constraint that for each d ∈ D , the result r = f ( d ) satisﬁes some criterion φ ( r , d ) . Thus for f we h a ve an impl icit d eﬁnition that is li kely to be satisﬁed by m any d iff erent implem entations. A provider P m of these machin es may deliver the good (say machine M ) in a box equipped wi th the following promises: (i) t he expected time of computin g f ( d ) i s below E t ( d ) , (ii) th e distribution of runni ng t imes is normal, (ii i) the spread is s , and (iv) energy consumption when running is below e (per unit of time). Are these four assertions a bout M best vie wed as prom ises, or is com puter science able to provide a better termi nology in this case? W e ﬁrst notice that it is certainly conceiv able that th e s etting is so com plex that it is implausibl e t hat any assertions about M like t he mentioned promises c an be prov en abo ut it as correctness assertions . This may b e the case in spi te of t he fact that each of these assertions are in essence well- deﬁned mathematical, logical, or combinatorial, statements. No amount of testing can compensate for that difﬁculty . The subjective status of such prom ises when understood as propos itions or ass ertions abo ut th e behavior of M wi ll no t go aw ay s imply by not thinking about them in terms of promises. A very important ﬂexibility results from viewing these assertions as clai ms about M that hav e been cast as promises. In order to appreciate this ﬂe xibil ity we will make the plausi ble assumption that the provider of M has arriv ed at thos e claims after due experimentation, thus giving them ini tially a status com parable to that of a s cientiﬁc hypothesis, at least from the provider’ s perspecti ve. W e consider the circums tance that the us er communi ty of m achines of type M and of simi lar competing equipment that has bee n provided by competing providers in combination with different pac kages of promi ses fee ls di ssatisﬁed with M in the li ght of P m ’ s promises. Now the members of the user community ha ve the option to update their trust in P m , rather than to contempl ate mod iﬁcations of th e promises that P has made. It s hould be no ticed that, apart from forgetting enti rely about P m ’ s promis es regard- ing M , the only alt ernativ e way to accomm odate their dissatis faction that i s open for 19 the members of the user community , is to threat thes e prom ises as scientiﬁc hypot heses which must now be up dated in a syst ematic fashion on t he basi s of new inform ation gained d uring usage. But foll owing t hat s ystematic path of renewe d hypothesis for- mation is impossib le without being given full info rmation on how P arrived at these opinions in the ﬁrst place. And such informatio n is l ikely t o be unav ailable to t he members of M ’ s user communit y . These remarks serve as an illustration of the p rinciple m entioned in Paragraph 2.3 above. 4.9 A technical reﬁ nement of t he example The setti ng j ust s ketched may seem artiﬁcial, but with the fol lowing example we will demonstrate that it m akes practical sense. Let 2 n represent t he set of b it sequ ences of length n . SHA-256-512 is the secure hash algori thm SHA-256 appl ied to 2 512 (see [12] for an authorit ativ e d escription and [4] for an alternative form alization). W e wil l ab- bre viate S HA-256-512 to h . As a domain D we choose D = 2 256 × 2 256 , and the range R is 2 256 . The criterion φ ( d ) = φ ( d 1 , d 2 ) works as follo ws: φ ( d 1 , d 2 , r ) ⇔ bs2n ( h ( d 1 || r )) ≤ bs 2n ( d 2 ) . Here || denotes string concatenation and bs 2n con verts elements (bit s equences) of 2 256 interpreted as naturals in binary not ation with big e ndian representation to natural numbers. 5 ASIC based machines that com pute functi ons f according to th ese concrete spec- iﬁcations, m odulo som e details, are now on the market as so-called Bitcoin (see [25]) mining t ools. 6 W e claim that viewing claims about these machines put forward by their providers as promises is con vi ncing. SHA-256-512 is presented by FIPS (see [12]) as a secure hashi ng functi on wh ich is for that reason su pposed to be resistant against so -called ﬁrst pre-image attacks: giv en r ∈ 2 256 it is very difﬁ cult to ﬁnd a s tring d ∈ 2 512 with h ( d ) = r . Althoug h expressing th e algorithmi c di fﬁ culty of ﬁnding d give n r as a property of SHA-256- 512 is a mathematical issu e in princip le, no indication of a proof o f the dif ﬁculty of that problem h as b een found . Ne vertheless, the assertion about h that the menti oned problem is compu tationally difﬁc ult to solve can be qualiﬁed as a scientiﬁc hy pothesis maintained b y the computer science commun ity at lar ge rather than as a promise made by FIPS. Nakamoto’ s design o f Bitcoin turns SHA-256-512 into a p arametrized prob lem (the Bitcoin mini ng probl em: computing f ( d 1 , d 2 ) , necessarily fast in order to win 5 Ignor ing any knowledge of SHA-2 56-51 2 and assuming th at it work s rando mly an estima te of the probab ility that some r satisﬁes the men tioned condition is bs2n ( d 1 ) · 2 − 256 . This provid es an e xpec- tation of the number of inp uts that m ust b e tried b efore a bit sequence r that “solves the comb inatorial problem ” is found . 6 Such equipm ent is being sold for instance unde r the trademarks A valon and Butterﬂy . 20 a competition and preferably at low energy cost s on cheap but dedicated machines in order to make a proﬁt as well ) whi ch is also assumed to be d ifﬁ cult to s olve, and which is assumed to gro w exponentially more difﬁc ult when the number of leading zeroes of d 2 increases. Otherwise Bitcoi n m ining w ill eventually be cracked so to say . In the absence of a thoroug h scient iﬁc analy sis of the Bitcoin mining problem the difﬁculty of t hat problem is probably adequately viewed as a promise made by Nakamoto and reproduced by the open source dev eloper community around Bitcoin. 4.10 About the philosophical literatur e o n pr omises Ha ving worked out a v iew on promises in th is Section, we are in the p osition to make some furt her remarks concerning the connectio n of non-ob ligationis t promissory the- ory as advocated above and th e plurality of views on promise that has been developed in philosophy . W e wil l use the illum inating entry on promi ses in the Stanford Ency- clopedia on Philo sophy as a guide (see [19]). W e refer to that survey paper only and don’t repeat references cited there. Our no n-obligationi st prom ise theory (NOPT) takes accounting for promissory obligations and measuring their force i n compariso n to other ki nds of obli gations for an issue of secondary pragmatic importance only . NOPT may be consid ered a branch of act-consequentialism , and also as a brach of th e expectational t heory of promises. In mechanical terms NOPT is connected to speech act theory and promises may be consi derd a sublcass of Searle’ s com missio ns, and of the illocut ions of Austin. Expectation alism portrays a vi ew of promises not unlike that of NOPT , where trust and m odiﬁcations of trust are th e key parameters against wh ich promises and promis e keeping are measured. NOPT seems to differ from expectationalist views in not payin g attention to whether or no t failing to liv e up to a prom ise and t hus failing to meet an expectation created by and trusted in connec- tion with a promise is wrong. Rather than thinkin g i n terms of wrongdoings th e future impact of not keeping a promise on the interaction between promising and trust main- tenance and its secondary effect o n the calculation of expectations from a promisee’ s perspectiv e is considered central in NOPT . NOPT h as been worked out in su ch a way as to be relev ant for automated, or arti- ﬁcial, agent communities as well as for c ommun ities of human agents. For that reason NOPT has a bias, at least in principl e, tow ards (i) quantitative aspects of the creation and maintenance of expectation, (ii) logical aspects (informal logic) of reasoning on the basis of expectation and trust, (iii) trust m aintenance and the i nteraction between trust and reputation dynamics, (iv) temporal as pects of a prom ise l ife-cycle, (v) i nter- action between diff erent promis es. The above focus of NOPT is at cost of a di minished focus on the following aspects: (i) moral aspects and ethical valuation of actio ns, (ii) t he special st atus of prom issory obligations (whi ch seem to be rather useless in a world of artiﬁcial agents), (iii) the ramiﬁcation of speech acts (though NOP T must view a promise as an act of communi- cation), (iv) promises in relatio n to contracts (for whi ch the im portance of obli gations 21 is n ot deni ed), and with various forms of contractuali sm and contractarian theory , (v) relations with Law . 4.10.1 Promises in the philosoph y of Law There is a lar ge body of literature in the philoso phy of Law about the role of promises in general, and in the cont ext of contracts in particular . In [29] t he questio n why and to what extent keeping a promi se ought to be legally enforced is dis cussed and sev en differ ent v iews on th e matter are distingui shed. What matters for us is t hat the author takes promising as a giv en instit ution and understands the concepti on of legal instruments as a variable, rather than the other way around. Ho wev er , though out t his paper prom ises in var iably give ri se to expectations, whereas the status of prom issory obligations is much less ob vious, though such obligations if an y are alwa ys connected to promissory expectations, which c onforms t o our o wn viewpoints. In [11] th e case is made that philosoph ical explanations of why promises are bin ing are less critical to the development of contract law than many authors seem to defend. Instead a notion of ef ﬁciency is brought forward which might hav e greater e xplanatory power reading the role o f promising. 4.10.2 Promising in anth ropology in [24] th e ob serva tion is docum ented that in some comm unities promising does not exist b ecause prom ises are not considered as creating (bindin g) promissory obliga- tions, which is take n to be an essent ial aspect of the institution of promising. W e might turn the message of the paper around and draw the conclusion that there exists anthro- pologic evidence for the feasibility of human com muniti es with a non-obli gationist conception of the institu tion of promising. 5 Pr omise valuation, assessment, r ea soning, and logic In this s ection we di scuss that a promi se may have tome value to its pro miser as well as to it s promisee. Furth er the s trength of a p romise is on d isplay as well as as v ariety of assessment method s and ﬁnally an embedding of the tropi c in “Informal Logic” is giv en. 5.1 Promise v alue Promises are valuable to agents, because they help to deal with uncertainty . Because certainty is key , a promi se is worth not hing unless there is trus t. Zero t rust m akes promises worthless. Tr ust might be based on a hist ory of keeping prom ises o r , in our termi nology , on a hi story o f p ositive assessments about a succession of promises. 22 So there i s a symmetry between trus t and prom ises that m ust be bro ken t o solve the dilemma. If there is trus t, a promise about future behaviour does not need t o be perceiv ed as an oblig ation on the p romiser but as an i ndication that b est effort will be respected. If a gi ven agent X does not trust t he promiser ho wever , it might assume that the promise implies an ob ligation on the prom iser . Th is perception of obl igation by X does no t of course impl y an obl igation perceiv ed by th e promis er . There is a fundam ental subjec- tivity in these perceptions. The value of a prom ise to any agent in its scope is an expectation of th e e ventual beneﬁt. Suppo se, then that A promises B 400 dol lars p er year . B prom ises to wash A ’ s windows at this price. Both are satisﬁed with the v alue the y get from thi s arr angement and prefer not to question it too much as this c ould unleash all kinds of consequences. Observer C can see th at t he values are quit e m ismatched, or that A is gettin g a poor deal by it s judgement, b ut C also cannot d eny that the relati onship is stable because both A and B are happy . 5.2 Promise str ength The above example il lustrates that once promises are made and hav e b een noticed by an audience consist ing of agents in its scope, these agents engage in a process o f trust management. The strength, or force, of a prom ise, viewed from the perspective of an agent in it s scop e, quantiﬁes the expectation that i t will be kept. The same promise may ha ve dif ferent strength f or diffe rent members of its audience, and the e volution of that strengt h may differ between agents with differe nt experiences. Promise strength combines (i) a promisee’ s prom ise independent trust in the prom iser , (ii) the credibi lity of the p romise (given the promiser), and (iii ) the p romise type related trus t of the promisee in the promiser . It mus t be stressed, howe ver , that the promiser , the promisee, and all agents in scope are individually respons ible for their management of the strength they attribute to a promise. Aggregate trust management may be used, but is is necessarily a v oluntary matter for all agents in volved. 5.3 Assess ments The noti on of whether prom ises are kept or n ot is central to their s ustained usefulness in any appli cation domain, thus we need to make mention of how this com es about in a theory of promises. It would be poss ible to go overboard and d elve i nto th e complexities o f observa tion and measurement tin an attempt to provide a satisfactory answer but that is no t in the spi rit of this p aper . W e seek instead a s impler notion whi ch is at the same l e vel o f abstraction as the concepts of promise and intention that we hav e introduced thus far . W e call this the concept of assessment . 23 T erm 17 (Assessment) An assess ment is a subjective statement made by an agent about whether the intentions of itself or of another agent wer e ful ﬁlled. Our no tion of a n a ssessment is more generic and less quantitativ e than a veriﬁcation. It is both subjectiv e and not a priori linked to observation. In a static theory of promises and intent ions the existence of i ntentions as w ell as t he value of assessments is l inked to st ate parameters l ike time. Thu s, for an intent ion of agent A , i n existence at t ime t , and k nown t o be by way of a prom ise that A had is sued before t , it m ay be th e case that agent B ’ s assessment, made at time t ′ > t is posi tive (or negative) . At this l e vel of description, we need n ot say any more abou t it than t his. What is important i s ho w a witness t o the promi se assesses the fate of the prom ise. Such an ass essor may or m ay not feel obliged to assess a resul ting state or activity in a particular way , might promise to conform to certain criteria, and s o on. What matters is only t he ass essment, which might or might no t be ratio nally obt ained. W e b eliev e that this is a fair model of the w orld in which we live. Assessment inv olves a variety of possibl e rout es to inference, i.e. there are differe nt kinds of assessment. This is a subj ectiv e issue, but t his should not be viewed as a weakness of our theory , it is an imp ortant feature of the subjectiv e nature of indi vidual assessment. 5.4 Promise r elated reasoning pr oces ses Promise descriptions are symboli c representati ons of actio ns, and events and states of affa irs that allow s ystematic transformatio n p rocesses. T ransforming t he physical representation of a symbolic description of a promise might be considered an instance of reasoning. The l ogical complexity of promises becomes apparent by taking into account the div ersity of reasoni ng processes related to a sin gle promis e th at are performed by promiser , promi see, and agents in scope of that promise. The fol lowing reasoning processes can be distinguis hed in relation to a single promise p with promiser A , b ody X , promisee B and with agent C in its scope ( C may b e equal to B ). W e assume that some representation of P and its compo nents, or of components i n a stage of preparation of P i s av ailable to an agent that performs reasoning about it. No analysis i s suggested of the means of representation used, but this range will i nclude m ental representation as i t occurs within an animate agent and also formal representation amenable for some formal reasoning process as it may be implemented within an inanimate intelligent agent. Pr omise preparation: a reasoning process performed by A leading to the desig n, tim- ing and issuing of p by A . Cr edibility analysi s: a rea soning process where agents C i n scope of p determine th e credibility t hey assign to A ’ s p romising p in the l ight o f general facts k nown of 24 A (but excluding speciﬁc historical information about the individual behavior of A as a member of its agent class.) Prior trust determination: reasonin g processes performed by C (provided t hat C i s in the scope of p ) a imed at (i) determining the trust C h ad in A before bec omin g aw are of p ( C ’ s p rior trust in A ), and (ii) specifying which expectations are generated by taking notice of p . Counter -pr omise deliberation: reasoning processes performed by B concerning plau- sible counter -promises that may be iss ued i n by B reply of taking notice of p (provided B is in scope of p ). Pr omise impact prediction: (this may be d one with the understanding that B has is- sued one or more plausib le coun ter -promises) ( i) r easoning processes performed by B (when i n s cope of p ) and by C (any agent in scope of p ) aimed at determin- ing the (change of) expectation (as viewe d by B resp. C ) that p creates in B (and that A intended t o generate), (ii ) and reasoning p rocesses aimed at m odiﬁcation of plans (held by B or by C ) giv en the change of expectations held by each of them that was brought about by taking notice of p . Pr omise ass essment: reasoning process es performed by C concerning (i) the way in which C will assess wheth er o r n ot A ’ s promise is kept, (ii) assessing th e l atter by means of the preferred assessment method. Pr omise withdrawal monitoring: A may at some later stage issue another promise, say q , the keeping of whi ch is n ot compatible with keeping p . It that case q qualiﬁes as a withdrawal of p . An agent C appli es a reasoning process that monitors and ev aluates subsequent pro mises issued by A to determine whether these may be viewed as breaking p to an extent that p may be considered ha vin g been withdrawn. T rust updating: reasoning processes in place for each agent C in scope of p . The process aims at updating C ’ s prior trust in A in accordance with the result of the assessment C makes concerning the degree of keeping p that is achiev ed by A . Reputation pr ocessing: a reasoning processes performed by each agent C i n scope of p aimed at e xchanging the effec ts o f t rust updates between different agent s. Reputation ﬂo w permits an agent C whil e having n o prior trust assessment of an agent A to acquire an initial trust of A by taking notice of evidence gathered about A b y other agents. 5.5 Promise r elated inf ormal logic Connections between prom ises and informal logi c are of a dive rse nature. Such con- nections adm it a general classiﬁcation. Here are four forms of s uch con nections t hat 25 may b e imagined, the ﬁrst three o f which may be consi dered relev ant i n t he case of promises: Pr omises as a contribution to inf ormal logic: Promises considered an additi onal en- try in the catalogue of entities that are central for informal logic. V iewpoints from inf ormal logic supporting pr omise theory: Deﬁnition and demar- cation of th e concept of a promise proﬁting from existing dev elopments in infor - mal logic. Methods of informal logic applied to pr o mise dynamics: Promise related reasoning mechanisms ha ve in common that a stri ct formalization is im plausible. A signif- icant variation between promise related reasoning patt erns for individual agents is plaus ible. This variation is a feature of informal logic rather th an of form alized logic. Pr omises as an alternative t o (parts of) informal logic: W e are not aware of parts of informal logic that would proﬁt from being rewritten i n the language of promises. 5.5.1 Speciﬁc links between promises and inf ormal logic Concerning th ese three forms of connections that matt er in the case of promises the following more speciﬁc remarks can be made: 1. Promi ses are ut terances l ike remarks, questions, announcements, replies, and obligations . 7 Special of promises i s the dynamic in terplay with credibil ity and trust. Issuing a promise differs from producing a st atement in d iscourse (dialecti- cal use), or from m aking an ar gument (alethi c use), or from rhetorical language (though rhetoric m ay in volve p romises and deceptive promises). W e mention [27] for an applicatio n of this di stinction. In that paper reason for action are studied in detail, and reasons for issuing a promise can be considered a subtopic within that theme worth of independent in vestigation. 2. Credibil ity assessm ent may in volve reasoning by analogy (simi lar agents have kept simil ar promis es in si milar circumstances and t herefore there promise is credible), and ar gum ents from authority (another h ighly regarded agent claims that the prom ise is credible). Pattern matching techni ques may be used to su p- port an agent in ﬁnding s imilarities that allow for comparativ e assessment of credibility . The form of reasoning is similar to “story si milarity” as discussed in [36]. 7 While a n obligatio n is not a speech act and not the result of a speech act a promise is likely to be the result o f a speech act. Th e d ynamos o f o bligations and prom ises differs. Such differences an b e fo und with utterance types remark, question, annou ncement, reply , and (to a lesser extent) prediction as well. 26 3. Trust ass essment may in volve ar guments from aut hority and as well as various forms of p robabilistic and plausible reasoning . W e mention [33] which deals with systemati c reasoning give n the absence of evidence, a very plausible con- dition for agents w ho need to assess trust in a prom ising agent. Many examples of trust based reasoni ng are given in [28]. Althou gh that paper focuses on trust as a human behavioral or mental feature most e x amples make sens e in a context of artiﬁcial examples just as well . In [35] one ﬁnds an analysis of detecting and a voiding circular reasoning when establishing or updating trust. 4. Both trust and credibility are con veyed and created throu gh a combinatio n of network bas ed reputation development, which is an instance of distributed rea- soning from authority , and distributed plausi ble reasoning. 5.5.2 Nonlogical proce ssing of promise bodies Agents aw are of promises that hav e been issued will perform processing on the basis of their own local representations of promise statements. From a symboli c perspective this in volve s symbo lic processing on (representation s of) promise types and promi se bodies. But there s eems t o be no need f or a notion of validity for the processing of p romise statement representation s that i s performed by agents in s cope of a prom ise. Wh ether or not promisi ng works out well to the extent that it is of sus tained use i n some context is a learning outcome by agents who may act as promisers and promisees. Besides alethic use, di alectical use, and rhetorical use, a beha vioral use is conceiv- able. Behavioral use of p romises does aw ay with a principled approach to the promise types and t he meanin g of prom ise bo dies as well as with valid reasoning on the basis of these ing redients and admits any form of associative proce ssing b y agents in volved that suits their objectiv es. 6 Pr omises versus obligations The importance of the notion of promi ses may be questioned by those who consider promises as a concept second ary to obligations . If one vi e ws obligations as the pri- mary concept, the v al ue or i mportance of prom ises unav oidably shrinks. W e shall now survey adv antages and disadvantages of obligations as an alt ernativ e cornerstone of a theory of multi -agent systems. Th e d iscussion w ill b e s omewhat asymmetric b ecause we will not base our comparison on a proper deﬁ nition of the concept of a n obligation (which we cannot fathom). Ha ving explained intentions and promises, l et us n ow try to describe th e notion of obligations in the same manner . The intuitive notion of an obligati on seems straight- forward, but it proves to have dif ﬁcult prop erties. W e refer t o Jackson & Alth am [22] for a con vincing exposition of these complications. 27 W e migh t try to think of obligations in a straig htforward way , for instance: an obligatio n is an intention that is per ceived to be necessary by an agent . Th is certainly captures som e o f the characteristics that we u nderstand by the term, but it also leav es many quest ions unanswered: is the feeling of the necessity voluntary o r forced, a matter of surviv al or simply an authoritarian con vention? An obligation, or more precisely th e impact of an obligati on in general, as felt by an agent within it s scope, falls into the category o f p ossible int entions. W e wi ll speak of an induced i ntention, the intention being achieving compliance with the obli gation. The induced intentio n m ust ha ve s ource, a target and a body , a nd the body must ha ve a quality and a quantity . Th e source and target are no w some what di f ﬁcult to underst and howe ver . Unclear is furt her to what extent an indu ced intentio n is current, and if so if it is a commitment. Can it be indif ferent, or ev en a dec eption or simply in valid? Beyond this, we shall not attem pt t o deﬁne obl igations m ore carefully in this paper . W e shall merely state some assumptions about them. An obli gation can b e impo sed by external con ditions, e.g. by the expected be- haviour of external agents, by la ws threatening sanctions etc, or it can be self-imposed by codes of personal beha vio ur whi ch an agent holds to be nec essary . But this i mposi- tion s uggests th e actio n of a force wh ich attemp ts to induce a comm itment in another agent (or itself). An obligation is a possib le intenti on which may or m ay n ot be current and may or may not h a ve the status of a comm itment. In any case t he agent is aware of any compelli ng reasons to include the int ention in the po rtfolio of commitments, either from within itself or without due to external forces. It seems nat ural then to refer to the s ource and tar get of the in duced i ntention as being the agent in which one attempts to induce the in tention, and the recipient of the intention respecti vely . Howe ver , the source of the obligation itself migh t n ot be an agent at all, but merely a set of external conditions, no rms, experiences or other information acquired by the agent that lead to a percei ved priority . Note again that e ven ‘forced’ behaviour can be classiﬁed under the r ealms of (pos- sible) intent ions since all behaviour can be intended. Again, we emphasize that this does not i mply t hat a coerced agent holds the i ntention that is being forced upon it. Nor does it say anything about wh ether th e agent i s able to resi st the force or not, or whether it matters if an obligation is self-imposed or e xternally impos ed. V iewed from th e perspectiv e of an agent, the notion of an obl igation im mediately seems signiﬁcantly more complicated than an int ention or a promise and does not seem to be close to the notion of either promises or intentions . W e hold that obligations are f ar from being a reliable tool for ensuring com pliance. If a law-gi ver wanted to ensure the compliance of an agent, a better strategy would be t o obtain a promise from th e agent, and to con v ince it to view the intention as a commitment since the law-giv er could never know whether t he agent had indeed committed to the body of the obligation. 28 6.1 Obligation strength Obligations can be assigned a m easure or de gree of strength from the perspective of an audience j ust as promises. When assess ing t he strength of an obl igation an agent will ﬁrst of all estimate the degree to which an obligation applies to its own c ontext, rather than the degree to which the obliging entity is entitled to do so. The div ergence between promi se strength and obli gation strength app ress strik - ingly when one appreciates th at, unlike promise strength, obligation strength has little to do with expectations and is n early i ndependent of future action of any agent in- volv ed. 6.2 Potential pr ominence of promises Our belief in restricted strong non-obligati onism im plies t hat conceptuall y separat- ing promises from obligation s is a s tarting point o f our work. V iewed as utterances promises mu st further be s eparated from other utterances such as in formativ e utter - ances, questions, predictions, and commands. If an agent A expresses a fact, say F , meant to be informative for an audi ence, the similarity with A prom ising F is signiﬁcant. Howe ver , when stating F , agent A m ay not be able, or have been able i n the past, to exercise it s i nﬂuence to bring about that F hold s true. Whi le i t is al ways acceptable that an agent acts in such a way t hat its promises are kept, the same is not al ways valid for informative statement s. For ins tance: an agent telling an employer that “he wi ll not b e present tomorrow because of a ﬂu” need not be absent if the ﬂu fails to hit as expected. Such a message implies no si ngle action that i ncreases its p lausibilit y . What may get lost when viewing informative utterances as promises is ob jectivity . This provides a clear separation, and at the same t ime it ind icates that in syst ems that do not measure or assess agent objectivity it is unproblematic to treat each informativ e utterance as a promise. When agent A asks question q to another agent , say B , in the scope of an audi- ence, then the correspondi ng utterance need not be considered a promise. Indeed for understanding the notion of a question one has no need of promises. In other words as concepts promises are not prior to quest ions. Ne vertheless it is quite plausible to re- place the question by an ut terance where A promi ses B som e reward upon B providing an answer to q . Finally , if agent A issues a command c to agent B in scope S , then l ike with questions as jus t mentioned, in order to und erstand what is go ing one has no need of the concept of a promise. Having said that, howe ver , again it is plausible that A ’ s command is replaced by a promise that A wil l provide some rewa rd if B acts as if it puts command c in to ef fect. Suggesti ng this replacement is plausible in particular i f one has no interest in obligatio ns that might result from the issuing of a command. W e ﬁnd th at promis es stand out as th e utterances from which all other com muni- cations can be derived. This give s prom issory utterances som e conceptual prominence of other utterances. 29 6.3 F or and against the primacy of obligations In variably , howev er prom ises are linke d wi th obli gations. By t he sheer weight of tradi- tion, obligations domi nate discuss ions of behaviour . Our ob ject is to p osition prom ises as a major tool for the speciﬁcation and the regulation of behavior . For that reason we make an attempt t o com pare p romises and obligatio ns as conceptu al too ls for dis - tributed systems design. 6.3.1 In fav our of obligations W e begin wi th lis ting so me arguments i n fa vor of obligations as regulatory mechanism when compared with promises. 1. Some p eople mig ht th ink t hat a promi se is an obligat ion because it seems to create o ne, and might therefore be considered equiv al ent to that obligati on. (Th is is a version of obligationism to which we oppose.) 2. Obl igations are a well kno w n conce pt from de ontic logic. There is an adv antage to to reducing the less well-known concept of promi ses to one t hat has been studied for more than ﬁfty years. (This is true, but it i mplies no more than that promises are worth studying.) 3. Obl igations have a formal status i n stat e laws and regulations. Th ere is no such public body of promises. (Promises are a dynamic phenomenon concurrent with autonomous action, listing promises globally and statically is not plausible.) 4. M any obligations give ris e to promis es which occur in th e process of fulﬁlling an obl igation. E.g. the cat mu st get fed while o wner is on holi day , the o wner is obliged to get the cat fed (by law forbidding cruelty to anim als). A friend promises to help in the ful ﬁllment of the obl igation. (This is true but it does not imply that obligation s are prior t o promises in general.) 6.3.2 Against obligations When balancin g obligati ons and promises as a regulatory m echanism for distributed systems signiﬁcant ar guments in fa vor of promises can be mentioned. 1. Obl igations have their own com plex log ic, that is on philo sophical grounds som e combinations of obligat ions are less plausi ble. Seemingly s imple deontic log ic still captures (or hides) a wealth of m oral dilem mas and ethical mysteries. W e refer to Holbo [21] for a strategies for dealing with inconsistent obligations. 2. M uch mo re than obli gations promises p rovide a to ol for th e self-organization of a community of aut onomous agents. At any stage, and concerning any fu- ture cooperative activity , agents can start exchanging promises thereby creating an increasingly complex and functional promis e graph which m ay e ventually provide a basis for usable expectation assessments. 30 3. Once is sued a prom ise admits a dynami cally maintained equilibrium between the expectation of t he promise being kept and with credibility of and trust in the promiser . Such b alancing tradeof fs are absent with obligations. 4. It is easy to imagine a sign iﬁcant plu rality of prom ises constituting a promis e graph re lated to a sin gle planned activity . It is count erproductiv e to explain such promises in terms of equally many obligations. 5. Suppo se one has the con cepts of promise av ailable, and n ow reﬂects upon ob li- gations. An agent P issue : ob issuing an obligati on with body b ob to all agents Q in a scope S , migh t be understood as simul taneously promi sing all members Q of S that P issue : ob will act i n su ch a way as if P had recei ved a prom ise with body B ob from Q . This is a reasonabl e explanation of what m ay happen when issuing an o bligation and it explains why obl igations seem to b e complex entities or e vents from the perspectiv e of promises. 6. If a future promis e (e.g. the promise to feed the cat in t he future) is in fact a deception then this falsiﬁes th e necessit y o f a relationship between promises and obligations . In other words, all intention s occurring as apparent intentions in promises cannot be induced obli gations because some promises can be decep- tions and these cannot be (easily) understood as induced obligations. 7. Sim ilarly , not all prom ises are about future actio ns, so there cannot be an i mplied obligation capturing its essence for al l prom ises. E .g. I prom ise th at the cat got fed. Indeed the o wner might actually be displeased that the cat was fed if i t was supposed to be dieting. 8. Th ere is a dissimilarity between promi ses and obligations. They are quite di f- ferent entities. Obligations m ay ca use promises and promises may cause obliga- tions, but prom ises have a ph ysical reality as e vents in space and time, whereas obligations do not. Obligations are at a diff erent le vel of abstraction altogether . 9. Promi ses are made on a voluntary basis. F or obligati on ho we ver , the concept of voluntarity is almost irrational. In any case it might be v oluntary to imply an obligation on someone els e, b ut engaging in a promise you may f ace an in volun- tary obligation o r a voluntary o ne. V oluntarity is therefore natural for promises but is quite problematic for obligati ons. 10. Promise announcement constitutes positive extensions of user behaviour , wh ereas obligations primaril y const itute a n egati ve constraint on the degrees of freedom of the obliged party . 31 6.3.3 Promises p reced e obli gations; locality pr ecedes globali ty If one would choose between promises and obligations, it seems ab un dantly clear that promises are the sim pler concept, or perhaps the more primitive concept. It is hard t o imagine a world without promising whi le one may imagine a w orld without ob liga- tions. In addition the concept of a promi se seems mo re natural in t he t echnological world: since computers cannot feel ethical responsibilit y , the reduction to promises to obliga- tions seems to be neither philosophi cally satisfactory nor technically correct. So far we hav e argued that promises are d iffe rent, simpler and can be analyzed independently of ob ligations. There is one more point that is o f p rincipal practical importance. Promises are local constructio ns, whereas obli gations are n on-local . The source of a promise i s localized in a sing le ent ity that has all of the in formation and self-control to be av ailable to resolve conﬂicts and problems w ith m ultiple promises. The sources o f o bligations howe ver are di stributed amongst m any individuals and the obliged p arty does not have the access to resolve the con ﬂicts w ithout maintaining a voluntary dialogue with all of these multip le parties. From a practical vie wpoint, obli gations are sim ply less effecti ve at reducing un- certainty because they tend to i ncrease uncertainty no t reduce it. Indeed, obligat ions can be inconsist ent, but promises cannot. Mo re p recisely: consi stency of prom ises is a m atter that can be veriﬁed at the level of sources only . Promises made by different agents canno t be i nconsistent if promis e bodi es are dealing with action s or states of af fairs about which the promiser has an ove rriding control against other agents. If dif- ferent agents iss ue conﬂicting promises at least one of these fails to have the e xpected degree of control. In any case on ce such conﬂicting prom ise are n oticed the t rust in both issuers needs reconsideration, and a decrease in trus t in at least one o f the prom ise agents is likely to occur . Preferably then one would not use obl igation as a coordination principle if a mech- anism based on promises can be used instead. 7 Conclusio ns W e ha ve restricted our attention to s tatic scenarios in this work, as the matter of change is a complex one. An account of how promises arri ve, persist and are removed again is forthcom ing. Some work has already been don e i n thi s area, howe ver , i n [2] and in [8] b ut scope for embelli shment is vast, as is the number app lications for the concept of promises. In t he latter reference, the mat ter of o r gani zation i s related to promises, as a form of cooperation between individuals or autonomous agents. Although we have av oided describing speciﬁc and detailed scenarios or using the notion of agent autonomy in t his work, on e of u s has wri tten extensiv ely on the v iew- point that promises represent a basis for a sound engineering d iscipline, somewhat analogous to an “atomic theory”, where arbitrary system s can be reduced to indepen- 32 dent agents that i ssue promises abo ut their behaviour [6]. This sets abou t a process of documentation of i ndependent prop erties that can be h elpful in detailin g one’ s under- standing of obs erved ph enomena. Indeed, in computin g in particular t here is a genuine case to be m ade for vie wing a c omput er as a number of independent electrical compo- nents that make quite speciﬁc promises to one another . Giv en th e role promises have i n inﬂuencing certainty , an exploration of the re- lationship to causation and fault propagation networks seems fruitful. Sim ilarly the beneﬁts in economics are obvious. Outside computing undoing promi ses from their obligation generating effects can be a helpful to strengthen t he role of prom ises as an organizational princip le. This may provide a fruitful direction for management science re search. W ithou t attem pting to s uggest applications in any ﬁeld, we have ar g ued for the usefulness of promises as an independent and practical concept, whether in p hiloso- phy , economics or technolo gy . W e have comp ared promises to the more usual idea of obligations and have concluded unequivocally that promises are a sim pler th eoret- ical notion and a more practical to ol than obligation s i n the reduction of an agent’ s uncertainty about the beha viour of other agents. W e sh ow that intention s, promis es, and commitm ents can be int roduced and t o a signiﬁcant extent explained i n the absence of an understandi ng of obligations . Fur- thermore, although i t seems to be a common view t hat obligations are a m ore pri mitive concept than promises, our paper sugg ests the contrary . Promises need not b e vi ewe d merely as proxies for obligation s; if prom ises giv e rise to obligat ions, this can i n fact be studi ed purely in an expositi on based on promises, i ntentions and comm itments. Indeed more often than not promises are m ade by agents who would not be able to explain the extent to whi ch their prom ises might lead t o obl igations or not , or to what extent such obligat ions would b e more s igniﬁcant than the promises from which th ey arose. References [1] P .S. Ati yah. Promises, morals, and law . Clarendon Press, Oxford, 1981. [2] J. A. Ber gstra, I. Bethke, and M. Burgess. A promis e alg ebra framew ork for promise theory . T echnical report, CoRR abs/0707. 0744, 2007. [3] Jan A. Bergstra and Karl de Leeuw . Bitcoin and Beyond: Ex clusive ly Infor- mational Money . http://arx iv.org/abs /1304.475 8v2 [cs.CY] , 2013. [4] J. A. Ber gst ra and C. A. Middelbur g. Instruction s equence expressions for the secure hash algorithm SHA-256. http://ar xiv.org/ab s/1301.3297 [cs.PL] , 2013. 33 [5] J. Anthony Blair . Govier’ s “Informal Logic”. Informal lo gic, 33 (2) pages 83-97, 2013. [6] Mark Burgess. An approach to understandin g pol icy based on autonom y and voluntary coop eration. In IFIP/IEEE 16th international workshop on distributed systems operations and manageme nt (DSOM), in LNCS 3775 pages 97–108, 2005. [7] M. Burgess. Promise yo u a rose garden. http://research.iu.hi o.no/papers/rosegarden.pdf . [8] M. Bur gess and S. Fagernes. Laws of system ic organization and collective be- haviour in ensembles. I n Pr oceedings o f MA CE 2007 , volume 6 of Multi con Lectur e Notes . Mult icon V erlag, 2007. [9] Juan D. Carrillo and Mathi as De watripont. Promises, promises,... The Economic J ournal, 118 (531) pages 1453–1473 , 2008. [10] J.P .W . Cartwright. An evidentiary theory of promises. Mind (New Series) , 93 (370) pages 230–248, 1984. [11] Richard Craswell. Contract law , default rules, and the p hilosophy of promising . Michigan L aw re view , 88 (3) pages 489-529, 1989. [12] FIPS 180-3. Secure Hash Standard. NIST (2008). [13] Jos e M . Framin am and Rainer Leis ten. A vailable-to-promise (A TP) syst ems: a classiﬁcation and framework for analysis . Int. J ournal o f Pr oducti on Resear ch, 48 (11) pages 3079–3103, 2010. [14] C. Fried. Contract as pr omises . Harv ard University Press, 19 81. [15] M. Gilbert. Is an agreement and exchange of promises? Journal of P hilosophy , 90 (12) pages 627–649, 1993. [16] Piotr J. Gmytrasiewicz and Edm und H. Durfee. Reasonin g about other agents : Philosophy , theory , and impl ementation. in: Pr oceedings of the Interna tional W orkshop on Distributed Artiﬁcial I ntelligence, Hidden V alle y , P A. 1993. [17] G.C. God du. Is ‘argument’ subj ect tot t he product/p rocess ambiguity? Informal Logic, 31 (2) pages 75-88, 2011. [18] L. Groarke. Informal Logic. The Stanf or d Encyclopedi a of Phil osophy (Spring 2013 Edition), Edwar d N. Zalta (ed.) , plato.stan ford.edu/ archives/spr2013/entries/ logic-info rmal/ 2 013. 34 [19] All en Habib. Promi ses. The Stanfor d Encyclopedi a of Philosop hy , Ed war d N. Zalta (ed) , plato.stan ford.edu/ archives/win2008/entries/promises/ , 2008. [20] Ben Hamby . T oulmi n’ s “ An alytic Arguments”. Informal Logic, 32 (1) pages 116-131, 2012. [21] Joh n Holbo. Moral di lemmas and the l ogic of obl igation. Ameri can Philosoph i- cal Quarterly , 39 (3) pages 259–273, 2002. [22] Frank Jackson and J.E.T . Altham. Understanding the Logi c of Obligations. Pr o- ceedings of the Aristot elian Society , Supplementary V olumes, 62 pages 255–283 , 1988. [23] R.H. Johnson. Making sense of ‘Informal L ogic”. Informal l ogic , 26 (3), pages 231-258, 2006. [24] Ferd K orn and Shul amit R. K orn. Where peop le don’t promise. Et hics, 93 (3) pages 445-450, 1983. [25] Satoshi Nakamoto. Bitcoin: a peer -to-peer electronic cash system. http://Bit coin.org/ Bitcoin.pdf (2008). [26] Pietro Panzar asa and Nicholas R. Jennings. The organisation of so- ciality: a m anifesto for a new science of mul ti-agent systems. in: 10th Eur opean W orkshop on Multi-agents systems (MAA W A W -01) http://htt p://eprin ts.soton.ac.uk/256150/ , (2001). [27] R.C. Pinto. A r gumentati on and the force of reasons. Informa l Logic, 29 (3) pages 268-295, 2006. [28] R.C. Pinto. Govier on trust. Info rmal Logic, 33 (2) pages 236-291, 2013. [29] Roscoe Pound. Promise or b ar gain? T ulane Law Review , 33 pages 455-472, 1959. [30] T . Scanlon. Promises and Practices. Ph ilosophy and Public Af fairs, 19 (3) pages 199-226, 1990. [31] H. Sheinman. Int roduction: promis es and agreements. in: ed. H. Scheinman, Pr omis es and Agr eements, Oxford U niversity Pr ess , 3-57 (2011). [32] T .C. Schelling. The Strate g y of Conﬂict . Harv ard University Press, Cambridge, Mass., 1960. [33] C. S tephens. A Bayesian approach t o absent e vidence r easoning. Info rmal Logic, 31 (1) pages 56-65, 2011. 35 [34] Samuel J. Stolj ar . The ambigui ty of prom ise. Northwestern University Law Revie w , 47 (1) pages 1–20, 1952. [35] Doug las W alton. Epistemic and dialectical models of begging the question. Syn- these, 152 (2) pages 237-284, 2006. [36] Doug las W alton. Story similarity in ar guments from analogy . Info rmal L ogic, 32 (2) pages 190-221, 2012. [37] Feng W an and Munindar P . Singh. Commitments and causali ty for mult iagent design. In Pr oceedings of th e 2nd Int ernational Joint Confer ence on Autonomous Agents and MultiAgent S ystems (AAMAS) , 2003. [38] M. W ooldri dge. An Intr od uction to Multi Agent Systems . W iley , Chichest er , 2002. [39] T angmin Y uan, T . Kelly . Ar g ument schemes in computer safety angineering. Informal logic , 31 (2), pages 89-109, 2011. [40] Xinyu Zhao, et . al. T oward a formal theory of b elief, capability , and promise incorporating temporal aspect. In: CEEMAS 2005, LN AI 3690 , pages 296– 305, 2005. 36

A static theory of promises

Original Paper

Comments & Academic Discussion

Leave a Comment

Original Paper

Related Papers

Comments & Academic Discussion

Leave a Comment