An explanation-oriented inquiry dialogue game for expert collaborative recommendations

A gumen & Compu a ion 15 (2024) 355–390 355 DOI 10.3233/AAC-230010 IOS P ess An explana ion-o ien ed inqui y dialogue game o expe collabo a i e ecommenda ions Qu a -ul-ain Shaheen a,c,∗, Ka a zyna Budzynska band Ca les Sie a a aIIIA, CSIC, Ba celona, Spain E-mails: [email p o ec ed],sie [email p o ec ed] bLabo a o y o The New E hos, Facul y o Adminis a ion and Social Sciences, Wa saw Uni e si y o Technology, Wa saw, Poland E-mail: Ka a [email protected] cSo wa e Compe ence Cen e Hagenbe g, Hagenbe g, Aus ia E-mail: qu a [email protected] Abs ac . This wo k p esen s a equi emen analysis o collabo a i e dialogues among medical expe s and an inqui y dialogue game based on his analysis o inco po a ing explainabili y in o mul iagen sys em design. The game allows expe s wi h di e en knowledge bases o collabo a i ely make ecommenda ions while gene a ing ich aces o he easoning p ocess h ough combining explana ion-based illocu iona y o ces in an inqui y dialogue. The dialogue game was implemen ed as a p o o ype web-applica ion and e alua ed agains he speci ica ion h ough a o ma i e use s udy. The use s udy con i ms ha he dialogue game mee s he needs o collabo a ion among medical expe s. I also p o ides insigh s on he eal-li e alue o dialogue-based communica ion ools o he medical communi y. Keywo ds: Inqui y dialogue game, collabo a i e decisions, expe decisions, explainable a i icial in elligence, human-cen ed compu ing 1. In oduc ion As he human socie y has become mo e digi ally connec ed, i has de eloped an inc eased app ecia- ion o in e disciplina y collabo a ion [23]. Heal hca e is one such domain which has a long adi ion o in e disciplina y collabo a ion amongs di e en medical expe s [67]. Imagine a dis ibu ed heal h ecommenda ion sys em whe e di e en expe s come oge he o ind he bes possible diagnosis o a pa ien . These expe s could be human agen s, a i icial agen s o a combina ion o human and a i icial agen s. The goal o he collabo a ion would be o in eg a e mul iple pe spec i es h ough knowledge ans e and con lic esolu ion in o de o ecommend he bes possible diagnosis. Addi ionally, he sys- em should o e explana ions o i s ecommenda ion in o de o build us be ween he use s and he sys em [44]. *Co esponding au ho : Qu a -ul-ain Shaheen, So wa e Comp ence Cen e Hagenbe g, Hagenbe g, Aus ia. E-mail: qu a - [email p o ec ed]. 1946-2166 © 2024 – The au ho s. Published by IOS P ess. This is an Open Access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion-NonComme cial License (CC BY-NC 4.0). 356 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions Table 1 Example dialogue be ween h ee medical expe s α,β,γ Id Dialogue 1α: Pa ien A is a 48 yea s old emale. He symp oms a e a igue, cons ipa ion, weigh gain, d owsiness, and d y skin. I looks like a case o dep ession. I ecommend li es yle changes such as heal hy ea ing and walking. I no e ha she has a amily his o y o au oimmune diseases. 2β: Can you jus i y you diagnosis o dep ession? 3γ: I would say ha i looks like a ypical case o hypo hy oidism. 4α: Because he symp oms a e ypical o dep ession. 5γ: I no e ha she does no ha e headache o back pain, which a e common in case o dep ession. 6β: Can you jus i y why you diagnose hypo hy oidism? 7γ: Because she has d y skin and a amily his o y o au o immune diseases, which a e ypical o hypo hy oidism. 8α: You migh be igh . I ecommend ha we es he TSH, T4 and T3 le els. 9β: I ag ee. I also ecommend doing a blood comple e coun o ule ou o he simila condi ions like anaemia. 10 γ: I ag ee o es ing TSH, T4 and blood comple e coun . I hink we can close he discussion now. 11 β: I ag ee. 12 α: I disag ee. Wha abou es ing o T3? 13 γ: Why do you wan o es o T3? 14 α: Because I wan o ule ou Hype hy oidism. 15 β: Yes, i makes sense. 16 γ: I don’ hink i is necessa y o es T3 a his s age since she is no asymp oma ic. And he symp oms a e close o Hypo hy oidism. 17 α: Okay. I hink we can close he discussion now. 18 β: I ag ee. 19 γ: I ag ee. As an example scena io, conside he dialogue gi en in Table 1be ween h ee medical expe s ep- esen ed by α,βand γ. The i s column o Table 1shows he iden i ie o he s a emen while he second column indica es he expe name ollowed by he s a emen hey a e making. The expe s a e pa icipa ing in a semi-s uc u ed o mal discussion simila o he ones ha ake place in oncology [43]. They a e al eady awa e o he objec i es and o ma o he mee ing be o e he dialogue s a s so he e is no build up on he objec i es o he discussion as is he case in oncology mee ings. We co e his in de ail in Sec ion 3. In his example, he mee ing is no abou a cance diagnosis, bu a he a mo e gene al diagnosis o a pa ien deno ed by A. This is a con i ed example, aimed a illus a ing he con e sa ion low among heal hca e expe s who a e pa icipa ing in a semi-s uc u ed discussion o collabo a i ely make a ecommenda ion o a pa ien . The goal o his example is no o p o ide a comp ehensi e medi- cal discussion on he pa ien ’s diagnosis as his is no he objec i e o his wo k. Mo eo e , no e ha he dialogue only shows he knowledge and easoning o he expe s o he ex en ha hey choose o e eal h ough hei con e sa ion. I does no show he comple e knowledge base o he easoning p ocess o he expe agen s (who can be human o a i icial). In his example, we assume ha agen αloosely ep esen s a clinical psychologis , agen βa gene al p ac i ione and agen γan endoc inologis (specialis in ea , nose and h oa ). Agen α i s p esen s he ac s o he case and o e s his own diagnosis (dep ession). This is challenged by agen β.α hen jus i ies hei s ance which is ejec ed by γ.γ hen p oposes hei own diagnosis. βasks γ o explain hei diagnosis. Subsequen ly, bo h αand βaccep γ’s explana ion. Then αp oposes some es s o which β ag ees and p opose an addi ional es . γag ees and p oposes ye ano he es . Howe e , hei sugges ion is igno ed by he o he wo as hey p opose o close he discussion. Howe e , γdoes no consen and Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 357 Fig. 1. Wo k low diag am o an explainable mul iagen ecommenda ion sys em employing EDG. asks he o he agen s o espond o hei sugges ion. They do espond and subsequen ly γag ees o end he discussion. Th oughou he discussion, he agen s y o collabo a e in a coope a i e manne . We use his dialogue as a unning example h oughou he es o he pape o illus a e ou app oach. As a i s s ep owa ds ealising such a hyb id human-a i icial mul iagen sys em capable o such a dialogue, we p opose a no el in e ac ion p o ocol be ween expe s agen s (whe he human o a i icial). We call his p o ocol as Expe s’ Dialogue Game (EDG). Figu e 1 isualises how EDG would i in o he pipeline o building such a sys em. As a i s s ep, eal-li e consul a ions among medical expe s a e o malised as a equi emen speci ica ion. Building on his speci ica ion, EDG is de ined as a dia- logue game among he pa icipan s in a hyb id human-a i icial mul iagen sys em (MAS). The ou pu o EDG is a ecommenda ion om he sys em along wi h a sequence o explana ions jus i ying he ecom- menda ion. These explana ions can hen be plugged in o a sys em-use dialogue o jus i y he sys em’s ecommenda ion o he use . The use in his case could be he pa ien himsel o a physician in cha ge o he pa ien . While mos o he cu en li e a u e [3,17,36,65,76] ocuses on explaining he wo king o a sys em o a human use h ough a sys em-use dialogue, EDG in es iga es how an explana ion dialogue can be used wi hin a mul iagen sys em as pa o agen easoning. Dialogue games a e dialec ical sys ems [22] in he adi ion o in o mal logic [77] and a e o mally de ined as e bal in e ac ions be ween wo o mo e playe s acco ding o some p e-de ined ules o he dialogue [39]. Each in e ac ion is speci ied wi h he use o locu ions which ep esen speech ac s pe mi ed in a gi en dialogue [62]. Dialogue games equi e ha each pa icipan main ains consis ency ac oss i s s a emen s, also called commi men s, a any poin in he dialogue [77]. Cha ac e is ics o hese e bal in e ac ions a e ypically de ined in mul iagen communica ion acco ding o he popula ypology in oduced by Wal on and K abbe [77]. Dialogue games ha e a long adi ion o being used o sol e o mal p oblems as well as o model na u al language communica ion in eal-li e se ings. In he i s case, hey ha e been employed o sea ch o o mal logical p oo s in he adi ion o Lo enzen [1]and leading o he ield o dialogical logic,aswellasin hep esc ip i e app oach such as Hamblin’s sys em [22], o disallow logical allacies du ing na u al language a gumen a ion. In he second case, hey a e used o s udy he communica ion dynamics in eal li e se ings, gi ing ise o he desc ip i e app oach o dialogue games [11,27,36,40,53,64]. This app oach can also be used o in o m compu a ional models o in e ac ions be ween agen s in a mul iagen sys em. 358 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions The goal o EDG is o he pa icipan s o collabo a i ely ind he bes ecommenda ion h ough ex- change o knowledge and mu ual ag eemen . In Wal on–K abbe’s ypology, his scena io i s an inqui y ype o dialogue in which an ini ial si ua ion is a need o ha e p oo (i.e. in ou case – o ind he bes ecommenda ion), he pa icipan s’ goal is o ind and e i y an e idence (i.e. o consul an obse a ion), and he goal o he dialogue is o p o e o disp o e a hypo hesis (i.e. o a gue o o agains a ecom- menda ion). This is in con as o a delibe a ion dialogue whe e he ini ial si ua ion is a need o ac ion, he pa icipan s’ goal is o in luence he ou come and he goal o he dialogue is o each a decision on he bes possible ac ion [77]. In o de o gene a e explana ions o he ecommenda ions, we de ine agen s’ communica i e beha iou in he dialogue h ough explana ion-based illocu iona y o ces which can hen be aced back and e ie ed in esponse o a que y and p esen ed o he use as an explana ion o he ecommenda ion: (1) explana ion eques s such as wh-explain(p) when he speake knows ha p is he case, bu does no unde s and why i is he case; wh-jus i y(p) when he speake does no ag ee ha pis he case and asks he hea e o he jus i ica ion o p;andwh-cla i y(p) when he speake does no unde s and a e m in pand asks o he cla i ica ion o his e m; and (2) explana ion eplies such as explain(p) when he speake p o ides an explana ion o p;jus i y(p) when hey p o ide a jus i ica ion o p;andcla i y(p) when hey p o ide a cla i ica ion o he e m in p. The con ibu ion o his wo k is h ee old. Fi s , we p esen a equi emen speci ica ion o collabo a- ion be ween expe s. While he equi emen speci ica ion is g ounded in he medical domain, i ocuses on he gene al communica ion issues ha can come up du ing expe collabo a ions. Hence, i can be ab- s ac ed o consul a ions among expe s in gene al. Nex , inspi ed by he adi ion o dialogue embedding [77], we combine explana ion-based illocu iona y o ces in an inqui y dialogue o gene a e iche aces o he inqui y p ocess han wha is possible wi h he asse locu ion, ypically used in inqui y dialogues. Mo eo e , he dialogue game is unique in ha sense ha i mee s he equi emen speci ica ion om he domain expe s. While he combina ion o di e en dialogue ypes is no g ound b eaking o mally, he p oposal makes an impo an me hodological s ep in o he applica ions o such o mal dialogue sys ems in o eal-li e domains which is empi ically g ounded in use s’ equi emen s. Fu he mo e, o he bes o ou knowledge, no o he desc ip i e dialogue game has ocused on in e ac ions among expe s. The ich game aces allow he sys em o be anspa en and can be used o explain he ecommenda ions o he sys em o a human use h ough human-machine in e aces such as e bal and isual in e ac ion. Finally, we e alua e he dialogue game agains he equi emen speci ica ion and e i y he e alua ion h ough a o ma i e use s udy. Thus, we in oduce he me hodology o use -cen ed so wa e enginee ing p ac- ice o dialogue games. While o he wo ks [11,27,36,40,53,64]indesc ip i e dialogue games ha e used insigh s om hei domain o in e es o in o m he dialogue games, none o hem ha e p o ided a equi emen speci ica ion as a as we a e awa e. The es o his wo k is s uc u ed as ollows. Sec ion 2summa ises ela ed wo k. Sec ion 3p esen s he equi emen speci ica ion o expe collabo a ions. Sec ion 4 o mally p esen s he Expe s’ Dia- logue Game (EDG) while Sec ion 5p o ides implemen a ion de ails o a web-based pla o m imple- men ing he EDG. In Sec ion 6, EDG is e alua ed agains he equi emen speci ica ion p esen ed in Sec ion 3. Sec ion 7highligh s use pe spec i es on EDG and he pla o m desc ibed in Sec ion 5.Fi- nally, Sec ion 8concludes he pape and p o ides di ec ions o u u e wo k. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 359 2. Rela ed wo k This sec ion summa ises ela ed wo k on explana o y and inqui y dialogues, dialogue games in mul i- agen sys ems o heal hca e, a gumen a ion in heal hca e and communica ion ools o mul idisciplina y collabo a ions in heal hca e. Amgoud e al. [2] p esen an a gumen a ion sys em o esol ing inconsis encies in an agen ’s knowl- edge base. Subsequen ly, hey show how dialogue game heo y can be applied on op o his o ealise he di e en dialogue ypes in Wal on and K abbe’s ypology [77]. A minimal amewo k o an explana o y dialogue sys em is p esen ed in Wal on [76]. The goal o he dialogue is o an explaine (an en i y ha explains, usually he sys em) o ill in he gaps in he knowledge base o he explainee ( he a ge o he explana ion, usually a human use ) by in o ming hem why some hing happened. This is conside ed as a ans e o unde s anding om he explaine o he explainee. Building on Wal on’s minimal amewo k, A ioua e al. [3] combine an explana o y dialogue wi h a gumen a i e illocu ions. The explana o y illo- cu iona y o ce is used by he sys em o explain he beha iou o some phenomena o he use while he a gumen a i e o ce helps o esol e inconsis ency in he knowledge bases o pa icipan s. In addi ion o commi men s o es o each pa icipan , hey in oduce an unde s anding s o e o he use which s o es he missing links in unde s anding a he han wha is cu en ly unde s ood. The discha ging o all issues in he unde s anding s o e con i ms a success ul ans e o unde s anding. Madumal e al. [36] analyse human-human and human-agen explana o y dialogues om a ious do- mains and p opose an explana o y dialogue p o ocol based on induc ion. Thei p o ocol also combines an explana o y dialogue wi h a gumen a i e acul y. The goal o he dialogue is he same as p oposed by Wal on [76]. They use double acknowledgemen o con i ming unde s anding, ha is, acknowledgemen by he explainee on being sa is ied wi h he explana ion and explaine ’s e u n acknowledgemen . Dennis e al. p esen an explana o y dialogue game o explaining he beha iou o a Belie –Desi e–In en ion Sys em [17]. Thei p o ocol, like EDG, is g ounded in dialogue game heo y om in o mal logic using a gumen a ion schemes and c i ical ques ions a he han a gumen a ion heo e ic dialogue [75]. The goal o he dialogue in his case is o unde s and sys em beha iou h ough compa ing aces o he sys- em om di e en pa icipan s. Simila o hese wo ks, we use explana ion-based illocu iona y o ces o ill in missing gaps in he knowledge bases o pa icipan s. Ilia e al. [65] ex end an in o ma ion seeking dialogue wi h explana o y illocu iona y o ces. The goal o he dialogue is o o e ac ual and coun e ac ual explana ions o a classi ie ’s ou pu o a human use . They also e alua e he p o ocol h ough a use s udy and p ocess analy ics. P akken and Ra sma [54] p opose a case-based explana ion dialogue o explain he ou come o a linea bina y classi ie . The goal o he dialogue is o p o ide a model-agnos ic local explana ion. The explana ions in his case a e no ying o explain he sys em easoning bu a he ying o come up wi h easons o jus i y he sys em esul . The dialogue game s a s wi h he p oponen o he dialogue p esen ing a simila example om he aining se wi h he same ou come as he cu en ins ance. The opponen can a gue agains his using wo s a egies. The i s is o use coun e examples om he aining se . The second is o highligh he di e ences be ween he cu en ins ance and he ins ance being p esen ed as a jus i ica ion by he p oponen . A success ul explana ion amoun s o a winning s a egy o he p oponen . Howe e , unlike hese app oaches, we inco po a e he explana o y illocu iona y o ces in he easoning p ocess i sel a he han o only explain why he sys em beha ed in a ce ain way. While hese wo ks a ge a ans e o unde s anding om he sys em o a human use , EDG in ol es explana ions amongs he easoning agen s. 360 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions Al hough no as common as pe suasion dialogues, a ew o he wo ks ha e explo ed inqui y dialogues as s and alone dialogues o in combina ion wi h o he dialogue ypes. Bex e al. [7] combine an inqui y dialogue wi h a pe suasion dialogue o discussions be ween c iminal in es iga o s. The goal o he dialogue is o come up wi h he mos obus explana ion. They assume an ad e sa ial se ing in which each agen ad oca es o i s own p e e ed explana ion. Unlike his wo k, EDG assumes a coope a i e se ing whe e he main goal is o come up no only wi h he mos obus explana ions bu decisions as well. Black and A kinson p opose a amewo k ha embeds an inqui y dialogue o e belie s wi h a pe suasion dialogue o e ac ions. The inqui y dialogue allows he pa icipan s o collabo a i ely decide wha o belie e whe eas he goal o he pe suasion dialogue is o collabo a i ely decide wha is he bes ac ion o do in o de o each he p oponen ’s goal. Once all he a gumen s ha e been gi en, i is up o o he p oponen o he dialogue o make he inal decision based his pe sonal p e e ences [8]. Black and Hun e [9,10] p opose wo ypes o inqui y dialogues, which hey call as a gumen inqui y and wa an inqui y. The goal o a gumen inqui y dialogue [9] is o wo agen s o join ly cons uc an a gumen o a claim by sha ing ele an belie s. The p o ocol has h ee mo es and allows nes ing o a gumen inqui y dialogues. In addi ion o a con en ional commi men s o e, hey also in oduce a ques ion s o e which keep acks o he p emises ha need o be p o en in o de o p o e he claim ep esen ing he dialogue opic. They p o e soundness and comple eness o hei p o ocol. The goal o awa an inqui y dialogue [10] is o wo agen s o sha e a gumen s o join ly cons uc a dialec ical ee in o de o de e mine he accep abili y o a pa icula a gumen . The main di e ence be ween he wo ypes o dialogues is ha a gumen inqui y is no conce ned abou he accep abili y o he a gumen cons uc ed while he la e is. Wa an inqui y dialogue allows embedding a gumen inqui y dialogue and also in ol es a ques ion s o e like he o me . They also p o ide a s a egy o selec ing he nex mo e o a pa icipan o bo h ypes o dialogues. Howe e , o he bes o ou knowledge, none o hese wo ks combine an explana o y dialogue wi h an inqui y dialogue o make agen easoning explainable. O he wo ks ha e inco po a ed a gumen a ion and dialogue games in mul iagen sys ems o p o ide clinical decision suppo in a dis ibu ed en i onmen such as cance diagnosis and managemen . Huang, Jennings and Fox [24,25] p esen a mul iagen a chi ec u e o medical decision suppo in an in e dis- ciplina y se ing. The a chi ec u e has ou componen s; a h ee laye ed knowledge base, a cen alised wo king memo y, a communica ions manage and a human-compu e in e ace. The a chi ec u e also co e s decision making unde unce ain y, ask managemen and agen coope a ion. The communica ion manage wo ks wi h communica ion p imi i es o locu ions and a communica ion p o ocol. Howe e , he locu ions in his case a e gea ed owa ds managing he asks in a dis ibu ed en i onmen a he han a discussion amongs di e en expe s. Fo example, he locu ions eques , accep , ejec and al e a e used in he ask alloca ion s age. The locu ion in o m is used o epo on he alloca ed ask while he locu ion p opose is used o ecommend a ea men plan in esponse o a que y. In con as , EDG is ocused on acili a ing he discussion amongs he expe s a he han a dis ibu ed managemen o esponsibili ies. Be e idge and Fox [6] use a dialogue game as an in e ace be ween he unde lying ask s uc u e and on ological knowledge and he spoken dialogue gene a ion sys em. They implemen hei app oach o p o ide clinical decision suppo o b eas cance diagnosis. They use se e al locu ions as ini ia ing locu ions. Fo example, in o m is used o p esen new in o ma ion, ins uc is used o eques an ac ion om he use , que y-yn o ask a ques ion wi h a yes o no answe and que y-w o elici a alue om he use as pa o da a en y. They ea he dialogue s a ed om each ini ia ing locu ion as a sub-game. In con as , he que y locu ions in EDG a e a ge ing owa ds inco po a ing di e en ypes o explana ions Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 361 in o he mul idisciplina y discussion amongs expe s. This is because he goal o EDG is o suppo collabo a i e discussion among expe s a he han suppo ing an indi idual use wi h decision suppo . Vasileiou e al. [73] p esen an a gumen a ion-based jus i ica ion dialogue be ween wo pa icipan s. The explainee is a human use who wan s o unde s and he explaine ’s (a i icial in elligen agen ) ea- soning. The dialogue game has ou locu ions, wo o which a e ese ed o each o he pa icipan s. The game only allows a single locu ion pe u n. They p o ide e alua ion o hei dialogue game h ough a use s udy and discuss i s p ope ies. Rago e al. [56] p esen he no ion o explana o y dialogue be- ween wo pa icipan s as an A gumen eXchange (AX). They discuss desi able p ope ies o AX o agen s equipped wi h quan i a i e bipola a gumen a ion amewo ks and g adual seman ics. [29,59,60] p opose an in e ac i e clinical decision suppo sys em, called CONSULT, o mul imo bidi y pa ien s o sel -manage hei ea men . The sys em in eg a es ou ypes o da a sou ces; he pa ien ’s elec onic heal h eco d, da a om senso s moni o ing he pa ien ’s symp oms, he clinician’s inpu and inally ea - men guidelines. I uses compu a ional a gumen a ion o agg ega e he da a and esol e inconsis encies in he da a sou ces. I also p o ides an a gumen a ion-based dialogue in e ace o sys em-pa ien in e - ac ion o in e ac i ely deli e he ecommenda ions o he use . The sys em-pa ien in e ac ion uses em- pla es o i s ex -based na u al language in e ace. I is based on h ee di e en a gumen a ion schemes and hei associa ed c i ical ques ions. These co e delibe a ion, pe suasion and explana ion dialogues. Cas agna e al. [13] p opose an explana ion dialogue be ween wo pa icipan s ha also in e aces wi h he CONSULT sys em h ough a cha bo . They p opose an a gumen scheme based on p ac ical eason- ing and use i o he explana o y dialogue. Shaheen e al. [63] p opose an explana o y dialogue be ween wo pa icipan s o explain he ecommended ea men plan o mul imo bid pa ien s. All o hese wo ks ocus on explana ion dialogues be ween wo pa icipan s, mainly a sys em as an explaine and a human as an explainee. In con as o hese wo ks, EDG aims o use di e en ypes o explana o y dialogue o ces o gene a e iche aces du ing in e -agen easoning p ocesses. Pancho e al. [41,71] p opose an a gumen a ion-based delibe a ion dialogue be ween wo agen s o discuss he iabili y o ansplan o gans. The dialogue is implemen ed as pa o Ca el+, a heal h in- o ma ion sys em o manage o gan ansplan s in Spain. The dialogue model, called P oCLAIM,[70]is based on a gumen schemes and case-based easoning. P oCLAIM employs a media o agen o guide he pa icipan agen s on hei legal mo es, decide he alidi y o submi ed a gumen s and inalise he ecommenda ion ega ding he iabili y o he p oposed ansplan o gan. The media o agen uses a - gumen schemes, exis ing guidelines, case-based easoning and a componen manage o manage he s eng hs o submi ed a gumen s. I applies abs ac a gumen a ion seman ics [18] o decide he win- ning a gumen . Xiao e al. [81] p esen a g oup decision desc ip ion language and a consensus p o ocol o a mul ia- gen sys em. Howe e , he p o ocol is no based on speech ac heo y [62], bu is an agen communica ion p o ocol ha uses unc ions like a e aging and in e sec ion o gene a e consensus alues unlike he wo k p esen ed he e. Pa ka e al. [49] de eloped a clinical decision suppo ool, called MATE (Mul idisci- plina y mee ing Assis an and T ea men sElec o ), o suppo mul idisciplina y cance con e ences. The ool is esponsible o in o ma ion managemen and p o iding ea men ecommenda ions a e p ocess- ing he da a. I does no p esen a dialogue game o suppo mul idisciplina y discussion amongs expe s as is p esen ed he e. A comp ehensi e su ey desc ibing he use compu a ional a gumen a ion o ex- plainable a i icial in elligence can be ound in Vassiliades e al. [74]. Some o he wo ks ha e p oposed compu a ional a gumen a ion sys ems o clinical decision suppo [15,20]. [] p esen a nego ia ion p o- ocol o agen s in a Belie Desi e In en ion (BDI) a chi ec u e. Howe e , he p o ocol is g ounded in agen communica ion language a he han dialogue game heo y. 362 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions The e m Heal h In o ma ion Technology (HIT) e e s o he applica ion o in o ma ion echnology o acili a e heal hca e. HIT sys ems all on a wide spec um anging om adminis a i e suppo , pa- ien in o ma ion managemen and e ie al, communica ion and decision suppo [14]. Ca ayon e al. [12] poin ou ha mos o he exis ing HIT sys ems a e ocused owa ds indi idual asks a he han eams, e en as eam-based ca e is becoming a popula pa adigm. He e we e iew ep esen a i e HIT applica ions a ge ing mul idisciplina y communica ion and collabo a ion suppo . Ca e Connec o [46,68,69] is a communica ion and collabo a ion pla o m implemen ed in a commu- ni y hospi al in Canada. I is a web applica ion ha in eg a es in o he HIT o he hospi al o e ie e and upda e elec onic eco ds. The applica ion co e s bo h ca e planning and moni o ing modules in addi- ion o a messaging module be ween mul idisciplina y ca e p o ide s. The pa ien in o ma ion is s o ed as pa o a Ca e Planne module. The messaging modules p o ides asynch onous communica ion o non-u gen messages using he in o ma ion in he Ca e Planne as a sha ed knowledge base. The mes- saging module allows linking each con e sa ion o a pa ien . I in o ms pa icipan s o pos messages ollowing he Si ua ion, Backg ound, Assessmen and Recommenda ion (SBAR) amewo k which is used in heal hca e communica ion. Howe e , i does no o ce he pa icipan s o ame hei messages acco ding o his amewo k. Ku ahashi e al. [30] p esen ano he communica ion and collabo a ion ool called Loop. I allows mul- idisciplina y collabo a ion be ween eams. The eams can include heal hca e p o essionals, ca egi e s as well as he pa ien . Each con e sa ion loop is cen ed a ound a pa ien . The applica ion includes a ca d wi h pa ien in o ma ion on he le hand side while he igh hand side has he messaging h ead. The in o ma ion exchange is secu e and sha ing in o ma ion be ween di e en subg oups is allowed. Fo example, a p o essional only message exchange o wi h all he pa icipan s in he loop. I also allows agging messages wi h use -de ined labels o acili a e sea ch la e on. The labels ep esen di e en hemes o issues desc ibed in he message. [32] implemen ed a pla o m, called one-s op pla o m, o mul idisciplina y collabo a ion among heal hca e p o essionals in a Taiwanese hospi al. The pla o m in eg a es in o exis ing HIT sys em o he hospi al. I co e s adminis a i e and planning aspec s in addi ion o a messaging module. The mes- saging modules allows anspa ency and accoun abili y o message pos ing and iewing. I suppo s exchanging ex , audio and ideo messages. Howe e , he e is no speci ica ion on he o ma o he con en ha is exchanged. Sha ed Ca e Pla o m (SCP) [38] is ye ano he collabo a ion ool implemen ed in a hospi al in Spain ha builds on social ne wo king and open sou ce ools. The ool is a ge ed owa ds acili a ing heal h- ca e p o essionals o manage mul imo bidi y pa ien s. I has wo componen s; a social ne wo king com- ponen , called he Clinical Wall, and a decision suppo componen . The Clinical Wall p o ides social ne wo king like collabo a ion and communica ion suppo amongs heal hca e p o essionals. I is in e- g a ed in o he elec onic heal h eco ds o he pa ien . The eco d has an assessmen sec ion, a discussion sec ion and a conclusion sec ion. The assessmen sec ion includes in o ma ion on pa ien his o y and as- sessmen s. The con e sa ion s a s wi h one clinicians posing a ques ion o o he s. In he discussion sec ion he clinicians exchange messages o a i e a an ag eemen wi h ega ds o he ques ion. In he conclusion sec ion all pa icipan s need o sign o on he ag eed decision. Du ing he discussion any clinician can be added o he con e sa ion o in i e hei eedback. The decision suppo componen uses he Clinical Wall and p o ides clinical guidelines in he o m o ules. O he wo ks [42,45,80,82] implemen mobile applica ions o suppo ca e and communica ion amongs heal hca e p o essionals, ca egi e s and pa ien s. These applica ions mainly suppo adminis a i e and in o ma ion managemen asks wi h simple messaging suppo o communica ion. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 363 All hese pla o ms and applica ions p o ide secu e messaging amongs pa icipan s and well in e- g a ed in e aces o he exis ing HIT sys ems in place. In con as , he p o o ype implemen a ion p o- ided o EDG does no p o ide any o hese ea u es since he goal in his wo k was o e alua e he unde lying p o ocol a he han p esen a ull- ledged web applica ion. Howe e , none o he exis ing applica ions p o ide suppo o aming he con en and ype o messages wi h an unde lying dialogue p o ocol as is p oposed in his wo k. So he pla o m p o ides a no el idea o suppo ing collabo a i e communica ions amongs heal hca e expe s based on an unde lying dialogue p o ocol. 3. Requi emen s o expe collabo a ion In his sec ion, we p opose a equi emen speci ica ion o success ul consul a ions among expe s. Consul a ions among expe s a e common in he p o essional wo ld, especially when c i ical decisions a e conce ned such as in medicine, a ia ion and enginee ing. We ocus on consul a ions among medical expe s as ou domain o choice in o de o de elop a dialogue p o ocol o consul a ions among expe agen s. This is because i is easie o abs ac away om domain speci ic e minology in his case in o de o unde s and he in e ac ion dynamics. Howe e , he equi emen speci ica ion we p esen is abs ac enough o be applied o consul a ions among expe s in gene al since i a oids domain speci ic scena ios and e minology. In o de o unde s and collabo a ion scena ios be ween expe s, we held in o mal discussion wi h some medical expe s (speci ically a gynaecologis , a adiologis , a gene al physician and a den is ). We iden i ied wo main scena ios, in o mal consul a ions such as du ing hand-o o a pa ien om an eme gency oom o a gene al o specialis wa d, and o mal consul a ions which usually ake he o m o mul idisciplina y cance con e ences o case con e ences o sho . These con e ences a e s uc u ed discussions be ween di e en specialis s o inalise diagnosis and ea men op ions o cance pa ien s [78]. The con e ence is a ended by mul iple specialis s such as su ge y, oncology and pa hology. I s a s o wi h he specialis in cha ge p esen ing each case his o y o he panel o expe s. This is ollowed by a discussion amongs expe s as o he bes possible diagnosis and ea men op ions o each pa ien [21]. Du ing he discussion, knowledge ans e be ween expe s akes place. This happens h ough he explana o y, inquisi i e and coope a i e one o he dialogue. Because o i s explana o y alue, many specialis s conside i o ha e educa ional alue o ainees [21]. We chose he case con e ence as ou main use case o in o m he p o ocol because o i s o mal and s uc u ed o ma . Mo eo e , some o he gene al communica ion issues [33,66,72] du ing in o mal hand-o also come up in case con e ences. Subsequen ly, we e iewed medical li e a u e on communica ion in case con e ences o iden i y possible issues. We also included some wo ks on gene al communica ion issues du ing in o mal collabo a ion be ween medical expe s. These wo ks we e included since hey we e gene al enough o be unde s ood by non-medical audiences. Su cli e e al. [66] iden i y wo ypes o communica ion ailu es in he medical p o ession: sys em- a ic and indi idual. While sys ema ic ailu es esul om a lack o su icien o ganisa ion, indi idual ailu es ha e complex oo s such as hie a chical and powe dynamics and excessi e wo kload. They sugges es ablishing communica ion guidelines o minimise bo h ypes o ailu es. In o de o mi iga e agains hese ailu es, we de elop an in e ac ion p o ocol g ounded in he in e ac ion dynamics du ing case con e ences [16] as well as gene al communica ion dynamics [66] ha can come up du ing in o - mal collabo a ion be ween medical expe s as a esul o o ganisa ional subcul u e [66]. Fo mally, we used Scopus, PubMed and GoogleSchola o look o pape s om he medical communi y ha iden i y 370 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions Thus, in he case o mo e han wo agen s, he p o ocol allows e e yone o pa icipa e in he explana o y dialogue since he dialogue game allows using mul iple esponses o each locu ion (a de ailed desc ip ion is p o ided in Sec ion 4.4 when he dialogue ules a e in oduced). This means ha in esponse o locu ions om subse L2, o he agen s who we e no di ec ly add essed in he p eceding wh- eques can choose o pa icipa e in he in o ma ion exchange by making an app op ia e mo e. 4.4. Rules, R The game has h ee s ages: an opening s age go e ned by Commencemen Rules, a p og ess s age go e ned by Combina ion Rules and a e mina ion s age desc ibed by Te mina ion Rules [52]. Each o hese a e desc ibed nex , ollowed by commi men , u n- aking and poli eness ules. Commencemen ules. The opic o he dialogue game can be one o mo e subse s o O.Thegame always s a s wi h he ini ia o agen p esen ing he ac s o he case (obse a ions), i s own conclusions ( e dic s), co esponding ecommenda ions (ad ice) and any c i ical poin s (conce ns) i deems impo - an . Thus, he i s u n is composed o he i s ou locu ions om he locu ion subse L1. A mo e, ep esen ed by μ, is a uple μ=τ,τwhe e τis he new locu ion being in oduced in he cu en u n while τis a p e iously in oduced locu ion. τis null o all mo es made in he commencemen phase. Fo mally, en ies L1.1 o L1.4 in he Locu ion column o Table 4 o mally p esen he ou In o ma ional locu ions ha a e pa o he i s u n. S ic o de ing is en o ced on he ou locu ions ha make up he i s mo e and is gi en by he sequence L1.1 o L1.4 in he able. Combina ion ules. The p o ocol allows pa icipan s o s a new h eads in he con e sa ion a any ime. This is achie ed by allowing one o mo e locu ions in he same u n whe e each locu ion co e- sponds o a mo e. Fo a mo e which uses he p omp locu ion, τis null. Repea ing he same mo e is no allowed, howe e , epea ing he same locu ion in he same u n wi h di e en con en is allowed. Hence, he nex pa icipan can espond o any numbe o locu ions om he p e ious u ns. In doing so, hey ha e o speci y he locu ion hey a e esponding o (τ) and pick any o he alid esponses o ha locu ion as de ined in Table 4whe e he Reply column indica es possible eac ions o each locu ion om he Locu ion column. Subsc ip s αand βiden i y he agen playing he mo e. Te mina ion ules. The dialogue e mina es when all he pa icipan s ag ee o end i . Any pa icipan can s a he p ocess o ge ing consen om o he s o end he dialogue. They can do his by using he end locu ion. This signals he s a o he e mina ion s age. Since he pa icipan s a e assumed o be asse i e and coope a i e, his means ha any hing ha he pa icipan s do no explici ly challenge is aken o g an ed as an ag eemen . Hence, when he dialogue ends, all he pa icipan s a e assumed o ha e ag eed on all he elemen s o se Ounde discussion. Howe e , each o ing o e mina ion may no always end in success ul e mina ion since any pa icipan can e use o gi e hei consen . They a e hen in i ed o highligh any ou s anding issues hey ha e by using he p omp locu ion as explained in Sec ion 4.3. I his happens, he dialogue mo es back in o he p og ess s age. O he wise, hey gi e hei consen o end he dialogue (and o accep all he s a emen s ha wen unchallenged by hem) by using he end locu ion. A mo e which uses he end locu ion also has τse o null. Commi men ules. Dialogue games gene ally equi e each pa icipan ha e hei commi men s pub- licly a ailable in he o m o a Commi men S o e. The commi men s a e c ea ed as a esul o pa icula speech ac s and hey ensu e accoun abili y o he pa icipan s. This is use ul o making he dialogue cohe en and p oduc i e. We ollow Hamblin’s no ion o commi men s o es as done by [48]whe e Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 371 an agen ’s commi men s o e, CS(i) o agen i∈X, co esponds o a publicly a ailable subse o i s o iginal knowledge base i. Speci ically he commi men s o e CS (i) a any ime in e al o agen i∈Xcon ains elemen s o Hi∪Oi∪Fi,whe eHi⊂H,Oi⊂Oand Fi⊂F. These ep esen he obse a ions, e dic s, ad ice and conce ns known by each agen a any ime du ing he game such ha CS +1(i) =CS (i) +{c|cis a new commi men }. In addi ion we in oduce a mul ila e al ag eemen s o e o all agen s, AS(MAS)={c|∃i, j ∈Xsuch ha c∈CS(i) ∩CS(j) and c∈K∪F∪O∪H}.Tha is, i con ains he obse a ions, e dic s, ad ice and conce ns ha ha e had a mul ila e al ag eemen a any ime du ing he dialogue. As in he case o [48], he union o all indi idual commi men s o es e lec he dialogue s a e any ime whe eas AS(MAS)shows he global ag eemen s a he han he in o ma ion s a e o he dialogue. AS(MAS)can be conside ed as a collec i e ag eemen s o e which p o ides a sum- ma y o he ag eemen s du ing he dialogue o all pa icipan s. This is because o he assump ion ha he agen s a e asse i e and he e o e, commi o any s a emen ha hey do no explici ly challenge. Locu ion L3.4 (ag ee) helps o inc emen ally build up his summa y, making i easie o synch onise he collec i e commi men s o all agen s. The mechanism o how his wo ks is explained in he commi men ules ha ollow. Each ule desc ibes he changes o CS(i) o agen i∈Xand AS(MAS)in eac ion o each locu ion. •C1: Fo locu ions subse L1 and L3.1 o L3.3, he con en o he locu ion is added only o he indi idual commi men s o e o he speake , CS(i). •C2: Fo locu ion L3.4, he con en o he locu ion is added o he indi idual commi men s o e o he speake agen and also o AS(MAS). I he con en o L3.4 is al eady added o AS(MAS)as a esul o a p e ious applica ion o C2, i is no added again. •C3: Fo a locu ion loc(a g)∈L2whe ea g∈L, no changes a e made o he indi idual commi - men s o e o he speake . Howe e , i a g had been added o AS(MAS) ollowing C2 ea lie in he dialogue, i is emo ed om AS(MAS)since i only con ains mul ila e al ag eemen s a he han bila e al ones. •C4: Fo all locu ions belonging o L4, no changes a e made o nei he he indi idual commi men s o e o he speake no AS(MAS). •C5: Fo L3.5, which ep esen s a e ac , he con en o he locu ion is emo ed om he commi - men s o e o he speake and om AS(MAS)i i has been added ollowing ule C2. •C6: A e he dialogue e mina es, he union o all indi idual commi men s o es minus he con lic s is added o AS(MAS) ollowing he no ion o implici ag eemen o all unchallenged s a emen s as desc ibed in Te mina ion Rules. Table 5exempli ies he commencemen , combina ion, commi men and e mina ion ules o he un- ning example. The i s column o Table 5indica es he u n iden i ie , he second column lis s he iden- i ie o he agen making he mo e, he hi d column iden i ies he locu ions mo ed, he ou h column shows he changes in he commi men s o e o he speake and he las column indica es he changes in he mul ila e al ag eemen s o e. In T1, he i s agen se s he con ex o he dialogue in acco dance wi h he commencemen ules. In T2, agen βasks α o jus i y hei diagnosis o dep ession since i s in e ence ule 3is in con lic wi h he diagnosis made by α.InT4,αp o ides his jus i ica ion. Howe e , in T3,γ p o ides i s own diagnosis and challenges α’s jus i ica ion in T5. Simila ly, β hen asks γ o jus i y hei diagnosis o hypo hy oidism in T6because i is in con lic wi h hei in e ence ule 3.γ’s jus i ica ion is accep ed by he o he wo agen s in T8and T9 espec i ely. In T13, γasks α o explain why i ec- ommends es ing T3 le els. This is because i al eady has in e ence ule 12 in i s knowledge base bu is con used wi h α’s easoning because o in e ence ules 7and 11. Figu e 2shows how he dialogue 372 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions Table 5 Example o a dialogue game be ween h ee medical expe agen s, α, β, γ ∈X1 Id Speake Sequence o locu ions CS(Speake )AS(MAS) T1αobse a ion(h1−h7) e dic (d1) ad ise( 1, 2) conce n(c1) pass h1−h7 d1 1, 2 c1 T2βwh-jus i y(d1), e dic (d1) pass T3γ e dic (d3), e dic (d1) pass d3 T4αjus i y( 1), wh-jus i y(d1) pass CST1(α) ∪ 1 T5γasse ( 2), jus i y( 1) pass CST3(γ )∪ 2 T6βwh-jus i y(d3), e dic (d3) pass T7γjus i y(h7,c 1, 10), wh-jus i y(d3) pass CST5(γ ) ∪{h7,c 1, 10} T8αag ee(h7,c 1, 10), jus i y(h7,c 1, 10) ad ise(e1,e 2,e 3), ad ise( 1, 2) pass CST4(α)∪ {h7,c 1, 10,e 1,e 2,e 3} {h7,c 1, 10} T9βag ee(e1,e 2,e 3), ad ise(e1,e 2,e 3) ad ise(e4), ad ise(e1,e 2,e 3) asse ( 3), jus i y(h7,c 1, 10) pass {e1,e 2,e 3,e 4, 3}AST8(MAS)∪{e1,e 2,e 3} T10 γag ee(e1,e 2,e 4), ad ise(e1,e 2,e 3) end pass CST7(γ ) ∪{e1,e 2,e 4}AST9(MAS)∪e4 T11 βend,end pass No change T12 αp omp ,ad ise(e1,e 2,e 3) pass No change T13 γwh-explain(e3), ad ise(e1,e 2,e 3) pass No change AST10(MAS)−e3 T14 αexplain( 12), wh-explain(e3) pass CST8(α)∪ 12 T15 βag ee( 12), explain( 12) pass CST9(β) ∪ 12 AST13(MAS)∪ 12 T16 γasse ( 4, 7, 11), explain( 12) pass CST10(γ ) ∪{ 7, 11} T17 αag ee( 4, 7, 11), asse ( 4, 7, 11) end pass CST14(α)∪{ 4, 7, 11}AST15(MAS)∪{ 4, 7, 11} T18 βend,end pass T19 γend,end pass AST17(MAS)∪ {h1,h 2,h 3,h 4,h 5,h 6}∪ { 1, 2,d 3} 1Subsc ip s o locu ions a e no men ioned o cla i y. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 373 Fig. 2. S a e ansi ions be ween commencemen , p og ess and e mina ion s a es o he example dialogue in able 5. Table 6 Al e na i e ending o he unning example. α,β,γ Id Dialogue 13γ: I don’ hink i is necessa y o es T3 a his s age since she is no asymp oma ic. 14α: Okay. I hink we can close he discussion now. 15β: I ag ee. 16γ: I ag ee. Table 7 Dialogue game be ween h ee medical expe agen s o he al e na i e ending in Table 62 Id Speake Sequence o locu ions CS(Speake )AS(MAS) T13γasse ( 7, 11), ad ise(e1,e 2,e 3) pass CST10(γ ) ∪{ 7, 11} T14αag ee( 7, 11), asse ( 7, 11) end pass CST8(α) ∪{ 7, 11}AST10(MAS)∪{ 7, 11} T15βend,end pass No change T16γend,end pass No change AST14(MAS)∪ {h1,h 2,h 3,h 4,h 5,h 6}∪ { 1, 2,d 3} 2Subsc ip s o locu ions a e no men ioned o cla i y. swi ches be ween he commencemen , p og ess and e mina ion s a es as a esul o he di e en u ns. Each node in he diag am ep esen s a s a e wi h he s a e label gi en in he cen e. Each a c ep esen s a ansi ion wi h he a c labels co esponding o he u n Ids in Table 5. EDG p omo es making jus i ica ions, explana ions and cla i ica ions explici in he discussion by no allowing asse in esponse o locu ions L1.2–L1.4. Table 6highligh s how his can a ec he dialogue. I shows how he unning example in Table 1would change a e he wel h mo e by αi asse was al- lowed in esponse o locu ion L1.3. The i s column indica es he iden i ie s o he al e na i e s a emen s made by he expe s. The iden i ie s o he al e na i e scena io a e appended wi h a symbol o indica e ha his is he al e na i e scena io. The second column shows he expe id and he s a emen hey a e making. Table 7shows he co esponding dialogue game. The columns a e o ganised in he same way as o Table 5. In his case, γdoes no ask α o an explana ion, a he i p o ides i ’s own easoning and he dialogue can close ea lie . Hence, i is clea ha EDG p omo es jus i ica ions, explana ions and cla i ica ions a he expense o sho e discussion ime. The collec i e ag eemen s o e se es as he ou pu o he mul i-agen sys em. I allows he mos ele an knowledge o he decision making o be pooled oge he in a sys ema ic way which is mo e 374 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions compu a ionally e icien han pooling all he knowledge bases o he agen s. In he p ocess, i also p e- se es he p i acy o he agen s since only publicly sha ed in o ma ion is used. This app oach allows o he building o a modula explainable mul iagen sys em in which he mul iagen decisions can be made independen ly o he human-machine in e ace. Fo example, i can be used o p o ide jus i ied deci- sions made by expe agen s o a use using ano he explana o y p o ocol o human-machine in e ac ion such as he one p oposed by Ilia e al. [65]. In his case, he collec i e ag eemen s o e can se e as he in e ace be ween he wo modules o he explainable A i icial In elligence (XAI) sys em. EDG elies on he locu ions ag ee and e ac o inc emen ally synch onise ag eemen s om he agen s’ indi idual commi men s o es o he collec i e ag eemen s o e. Howe e , i he agen s ail o use hese ma ke s su icien ly, he bu den o synch onising he ag eemen s o e will mo e o he end o he dialogue, s essing compu a ional esou ces. Hence, he agen s need o be made awa e ha explici ag eemen s will make he p o ocol mo e e ec i e. P omo ing elici a ion o jus i ica ions, explana ions and cla i ica ions allows EDG o keep ack o collec i e ag eemen s and esol e disc epancies in he ag eemen s o e. An example o his can be seen by compa ing he example in Table 5wi h he al e na i e scena io p esen ed in Table 7. In he i s case, he wh- eques in u n T13 esul s in emo ing he dispu ed ecommenda ion o e3 om he ag eemen s o e. In con as , since γne e makes hei s ance explici in he scena io in Table 7,e3 emains in he collec i e ag eemen s o e when he dialogue ends. So, he assump ion ha he agen s a e asse i e is e y impo an o ensu e he success o EDG. An unasse i e agen migh end up being commi ed o belie s ha a e no consis en wi h i s knowledge base. Tu n- aking ules. EDG iden i ies wo oles o he pa icipan s, ini ia o and pa icipan .Howe e , he ini ia o ole ends a e he i s u n, whe eby e e yone becomes a pa icipan . The ini ia o p o ides su icien con ex o he dialogue h ough he locu ions in he i s u n. The p o ocol en o ces u ns bu no pa icula u n-o de is en o ced. Each agen has o mo e a leas one locu ion in esponse o he p oponen ’s mo ed locu ions. Since mul iple locu ions a e allowed in each u n, each agen has o end his u n wi h he pass locu ion o ma k ha he is inished. Poli eness ules. S uc u ally, dialogue games can allow pa icipan s o espond only once o each mo e (single- eply) o o e se e al esponses as well (mul i- eply), o use only one locu ion in each mo e (single-mo e) o mo e han one (mul i-mo e) and o ans e he u n as soon as some objec i e condi ion is me (immedia e- eply) o la e (non-immedia e- eply) [52]. Based on hese de ini ions, we conside EDG o be mul i- eply,mul i-mo e and non-immedia e- eply. A b ie discussion jus i ying each o hese p ope ies ollows nex . Mul i- eply. The p o ocol achie es his in h ee ways. The i s wo enable his p ope y o he e- sponden while he las one enables he speake o p oac i ely demand an addi ional esponse. Fo he esponden , i allows mul iple a gumen s in one u n by no imposing any es ic ions on he numbe o a gumen s included as con en o each locu ion. Fo wo, i allows esponden s o come back o ea lie choice poin s in he dialogue since i does no impose he es ic ion on add essing he p eceding mo e. So hey can mo e se e al a gumen s e e ing o di e en p e ious mo es i desi ed. Fo he speake , i enables hem o di ec he con e sa ion back o issues ha we e no add essed o hei sa is ac ion using locu ion L4.1. Mul i-mo e. The p o ocol does no limi he numbe o locu ions ha can be mo ed in one u n by each pa icipan (see Sec ion 4.4). Hence, i is by cons uc ion mul i-mo e. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 375 Non-immedia e- eply. Since he p o ocol does no en o ce an ex e nal condi ion o shi he u n, i allows each agen o comple e i s mo e unin e up ed and p oac i ely ans e he u n, i is hen non- immedia e- eply. All hese p ope ies make EDG e y lexible and close o na u al con e sa ion. Howe e , his lexibili y can lead o dialogues ha a e incohe en o comp omise he explanabili y and coope a i e aspec s o he dialogue. Hence, i calls o in oducing he same manne isms in place in na u al con e sa ions ha ac o coun e hese complica ions in eal li e con e sa ion. So, EDG in oduces wo such manne ism in o he dialogue as poli eness ules. I iden i ies wo such ules o ensu e dialogue p og ession and con lic esolu ion. The i s is ela ed o wh- eques . Since he p o ocol does no o ce a pa icipan o espond o only he p e ious mo e, he pa icipan s can igno e explana ion, cla i ica ion and jus i ica ion eques s, de ea ing he explana o y objec i es o he dialogue. In o de o mi iga e his, he i s poli eness ule equi es ha all Wh-Reques s mus be esponded o by he add essee i s be o e hey a e allowed o espond o any o he locu ion. O he pa icipan s who we e no he di ec add essee, can espond o a wh- eques i hey choose o do so. This allows he pa icipan s o collabo a i ely build explana ions. The second ule conce ns p omp .Ap omp se es as a eminde o he pa icipan s ha his pa icula agen is awai ing a esponse o he p omp ed locu ion. The second ule gi es he ecei e s o he p omp , he lexibili y o choose o espond o i immedia ely o in a la e u n. 4.5. Seman ics We ake a p o ocol-o ien ed iew o Agen Communica ion Language (ACL) seman ics [50,52]. In his iew, he seman ics and use o u e ances should be de ined a he dialogue le el a he ha a he le el o indi idual locu ions [52]. Pi and Mamdani [50] dis inguish be ween he con en and con e sa ional s a es o he dialogue. The o me is dependen on he in o ma ion s a e o he agen . In o ma ion s a e o an agen e lec s i s knowledge base. Seman ics a his le el de ine he change in he agen ’s in o ma ion s a e. The la e is de e mined by he speech ac s exchanged ea lie in he dialogue, which a e e e ed o as he con e sa ion s a e. The con e sa ion s a e can be desc ibed by he se o possible esponses o each speech ac . Consequen ly, he commi men ules desc ibed in Sec ion 4.4 o m he con en le el seman ics o he p o ocol while he combina ion ules gi en in Table 4de ine he con e sa ional seman ics o he dialogue. Since he p o ocol ea s he commi men s o e as a subse o he agen ’s knowledge base, he commi men ules exp ess pos -condi ions abou he agen ’s in o ma ion s a e as a esul o he speech ac . Nex we desc ibe he p e-condi ions o making he mo e. P e-condi ions o managing in o ma ion s a e. •P1. Fo locu ions subse s L1 and L3.1 o L3.3, he e a e no cons ain s on he con en excep ha i should be ele an o he dialogue opic. •P2. Fo locu ion subse s L2, L3.4 and L4.1, he con en o he locu ions should al eady be pa o he commi men s o es o one o he agen s. •P3. Fo L3.5, he con en o he locu ion should belong o he commi men s o e o he speake . •P4. Fo L4.2 and L4.3 no condi ions apply as he e is no con en . P e-condi ions o managing con e sa ion s a e. •P5. Fo locu ion subse s L1, L2, L3 and L4.1, hose imposed by Table 4. •P6. Fo L4.2, he agen inds no con lic s o objec ions in he in o ma ion s a e o he dialogue as ep esen ed by i s own commi men s o e. 376 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions •P7. Fo L4.3, he agen making his mo e mus ha e used a leas one o he alid locu ion be o e his one. While p e-condi ions can ela e o bo h cons ain s on he agen ’s in o ma ion s a e [2,48] o o con- s ain s on he con e sa ional s a e o he dialogue [76], he e we speci y p e-condi ions o manage he in o ma ion and con e sa ional s a es o he dialogue i sel . The limi s in oduced on he con en o locu- ions in Table 4also o m pa o he p e-condi ions o managing he in o ma ion s a e o he dialogue. We equi e ha he agen s main ain dialogue his o y and do no epea a locu ion wi h he same con en . 5. A pla o m o expe collabo a ion A p o o ype o EDG was implemen ed as a web applica ion in o de o e alua e i wi h human expe s h ough a use s udy. The web applica ion allows he pa icipan s o ‘cha ’ while en o cing EDG p o ocol. Howe e , he pa icipan s do no need o emembe he p o ocol, he web applica ion en o ces i o hem. Fo each ‘message’ in he cha , i shows he possible locu ions om Table 4 ha can be used in esponse as a d op down menu. The pa icipan can selec a locu ion o ame hei esponse and ype hei ex in he co esponding ex ield. This sec ion p o ides de ails on he implemen a ion and he use s udy design. 5.1. Implemen a ion EDG was implemen ed as a p o o ype ull-s ack web applica ion ha allows human pa icipan s o engage in discussion ega ding he bes diagnosis and ea men op ions o a pa ien . Figu e 3shows a sc eensho o g aphical use in e ace o he web applica ion wi h a hypo he ical example. The ap- plica ion was implemen ed using Ja aSc ip amewo ks o clien and se e . The dialogue his o y is eco ded in an SQLi e da abase on he se e . The se e keeps ack o he numbe o pa icipan s in a game and o a es he u ns in a cyclic manne in he o de in which he pa icipan s join he game ses- sion. The applica ion en o ces poli eness ules by using highligh s. I ale s he use who was he a ge o a wh- eques by highligh ing he eques in ed and no allowing his use o play any o he locu ion un il all he wh- eques s o hem ha e been discha ged. Simila ly, i a use plays he p omp locu ion, he a ge locu ion is highligh ed in blue on all pa icipan s’ use in e ace o ale hem on he eques o esponse. Howe e , hey a e no o ced o espond o his ale . The web applica ion was used o e alua e he usabili y o EDG h ough a use s udy. 5.2. Use s udy We conduc ed a use s udy o e alua e he usabili y o he pla o m and i s unde lying p o ocol. Ac- co ding o he In e na ional O ganiza ion o S anda diza ion (9241–11:2018), Usabili y measu es how e ec i ely,e icien ly and sa is ac o ily a sys em, p oduc o se ice can be used by he speci ied use s o achie ing hei speci ied goals [26]. We ca ied ou o ma i e usabili y es ing wi h a o al o six pa icipan s. Fo ma i e usabili y es ing is done du ing he de elopmen p ocess wi h a ela i ely small numbe o pa icipan s o iden i y po en ial issues [4]. While adi ional usabili y es ing equi es 30 o 50 es subjec s, a ype o o ma i e usabili y es ing app oach known as discoun usabili y es ing has been shown o unco e 85% o he issues o he ask a hand wi h 5 es subjec s, wi h no signi ican subsequen inc ease in he a io o es ing cos o bene i s gained as he numbe o pa icipan s is in- c eased [47]. Due o he highly specialised p o ile (i.e. medical expe s) equi ed o he es subjec s in Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 377 Fig. 3. Sc eensho o he web applica ion implemen ing EDG. 378 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions his use s udy, i was di icul o ec ui es subjec s. So discoun es ing was done wi h he minimum numbe o pa icipan s o he o ma i e s udy. The goal o he s udy was o elici use ’s pe spec i es on e ec i eness, e iciency and sa is ac ion o he pla o m and he unde lying p o ocol in mee ing hei p o essional communica ion needs. Pa icipan s. The pa icipan s we e inal yea medical s uden s om a medical uni e si y in Ba celona, Spain. They we e olun ee s who esponded o a call o pa icipa ion a e eading he ad e ised in o - ma ion shee h ough hei Uni e si y’s human esou ce depa men . Design. Fo each session o he usabili y es ing, pa icipan s we e di ided in o g oups o h ee. They we e hen asked wi h collabo a i ely deciding on he bes possible diagnosis and ad ice o an anony- mous pa ien wi h a hy oid diso de . The pa ien da a was aken om he publicly a ailable hy oid da ase om he UCI ML eposi o y (h ps://a chi e.ics.uci.edu/ml/da ase s/Thy oid+Disease). Each pa icipan was gi en an iden ical ask desc ip ion bu he pa ien da a was unequally dis ibu ed be ween he pa icipan s o see whe he ans e o in o ma ion would ake place. All sessions we e conduc ed in a compu e lab a a uni e si y whe e he pa icipan s had access o compu e s. All he pa icipan s in one session we e in he same oom a he same ime. This was done o acili a e he adminis a ion o he s udy. Howe e , pa icipan s we e no allowed e bal o non- e bal (ges u es such as eye con ac ) com- munica ion wi h each o he du ing he s udy in o de o ensu e ha all communica ion ook place h ough he web applica ion. A e he pa icipan s inished he ask, semi-s uc u ed in e iews we e conduc ed o ge hei quali a i e eedback on he applica ion and he unde lying p o ocol. The pa icipan s we e gi en 30 minu es o he ask and 11–20 minu es o he pos session in e iews. All he pa icipan s who we e in he same session we e in e iewed oge he as a g oup. Figu e 4shows he sequence o s eps pa - icipan s pe o med du ing he use s udy. Fi s , all pa icipan s egis e ed wi h he web applica ion. The Fig. 4. Wo k low diag am o he use s udy design. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 379 web applica ion en o ced he sign up o de as he u n aking o de in acco dance wi h he u n aking ules. The i s pa icipan hen made he opening mo e using he G aphical Use In e ace (GUI) o he web applica ion. The u n was hen passed on o he nex pa icipan in line. Each pa icipan was no i ied by he web applica ion when i was hei u n. Each pa icipan was ee o choose which locu ions hey wan ed o espond o and also which locu ion o use as hei esponse. The online dialogue p oceeded like his un il one pa icipan in oked he e mina ion p o ocol by using he end locu ion. A e wa ds, he e mina ion p o ocol was ollowed as desc ibed in e mina ion ules in Sec ion 4.4. Goals. The goal o he use s udy was o elici pe spec i es o medical expe s on aspec s ela ed o e ec i eness, e iciency, engagemen and ease o lea ning o he discussion pla o m and he unde lying p o ocol. These aspec s also ie in o he equi emen speci ica ion om Sec ion 3.Howe e , heyallow some addi ional usabili y conside a ions o be explici ly aken in o accoun . Speci ically, i aimed o answe he ollowing ques ions o each o hese aspec s: (1) E ec i e (a) Does he pla o m add alue o p o essional discussions o medical expe s? (b) Did knowledge ans e ake place as a esul o he discussion? (c) A e pa icipan s sa is ied wi h he explana ions p o ided? (d) A e pa icipan s sa is ied wi h he inal decision and i s jus i ica ion? (2) E icien (a) Do pa icipan s ha e he mo es hey need a each s ep o exp ess hemsel es? (b) Does he applica ion impede he pa icipan s in some way du ing he dialogue? (3) Engaging Do pa icipan s a e he expe ience as enjoyable? (4) Easy o Lea n (a) Does he applica ion and he p o ocol p omo e discussion? (b) Do pa icipan s ind he classi ica ion o explana ion eques s use ul o con using? The nex sec ion p o ides de ails on he esul o he use s udy. Addi ionally i p o ides a de ailed e alua ion o EDG acco ding o he equi emen speci ica ion o Sec ion 3. 6. E alua ion agains equi emen s o expe collabo a ion In his sec ion we discuss how he p o ocol and i s implemen a ion measu e up agains he di e en ypes o equi emen s iden i ied in Sec ion 3. We in oduce h ee e alua ion c i e ia, e e ed o as le els, o his: dialogue,sys em design and use s udy. Each o hese is in oduced nex . Dialogue. This le el de e mines how well he dialogue ules p esen ed in Sec ion 4con ibu e owa ds sa is ying each equi emen . Sys em design. This le el e alua es each equi emen agains he p o ocol implemen a ion since some equi emen s can only be me a he implemen a ion le el. Use s udy. This le el alida es he sa is ac ion o each equi emen h ough he use s udy desc ibed in Sec ion 5.2. Table 8p esen s a summa y o he esul s. The i s column shows he id o each equi emen , columns second o ou h show whe he he co esponding equi emen is e i ied h ough he dialogue, sys em 386 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions mul iagen sys ems. This wo k also p esen ed an e alua ion o he dialogue game agains he equi emen speci ica ion h ough a use s udy. The use s udy was also signi ican in ha i highligh ed he eal li e u ili y o such dialogue pla o ms in he medical domain. Such pla o ms can enable clea e and sys em- a ic communica ion ac oss mul iple heal hca e disciplines and sec o s. This wo k p oposes o impo he me hodology o so wa e enginee ing in o he a ea o o mal dialogues. This me hodology consis s o he ollowing s eps: he collec ion o equi emen s o a dialogue game in a selec ed domain applica ion; he design and implemen a ion o a p o ocol acco ding o hese equi emen s; and he e alua ion o a dialogue sys em agains he equi emen s. The nex s ep would be o implemen and e alua e EDG o a mul iagen sys em. This would in ol e e alua ing o mal p ope ies o he sys em such as deadlocks, li elocks and e mina ion gua an ees. One possible app oach o in es iga ing un ime e mina ion gua an ees o EDG is o explo e mul iagen amewo ks ha can p o ide his kind o gua an ee h ough an app op ia e mode a o ole. Fo example, he go e no ole in Elec onic Ins i u ions [19] can be ex ended o close e mina ion p ope y a un ime. Ano he esea ch di ec ion could be o close he p o ocol agains dis up ion by non-coope a i e and malicious agen s. I migh also be in e es ing o in es iga e whe he he p o ocol can do away wi h u n- aking and synch onous communica ion since i may no scale well o a sys em wi h many pa icipan s. While EDG p esen s which locu ions can be used in esponse o o he s, i does no in es iga e how an agen can selec he bes locu ion in esponse o ano he based on i s knowledge base. This is a e y impo an aspec o implemen he p o ocol in a mul iagen sys em. Hence, an impo an u u e di ec ion is o de elop easoning s a egies o agen s o pa icipa ing in he EDG. One possible app oach o doing his is o explo e a gumen a ion-based easoning o EDG on he lines o [10]. Subsequen ly, he nex s ep would be o adap and implemen he p o ocol in a human-a i icial agen hyb id sys em. O he possible in e es ing di ec ions o in es iga e include in es iga ing how well he equi emen speci ica ion p esen ed in his wo k gene alises o o he domains such as enginee ing o a ia ion. On he lip side, i can also be in e es ing o e alua e how well exis ing dialogue game p o ocols [10,51,58] con o m o he equi emen speci ica ion h ough a use s udy. Finally, he p o o ype o he pla o m de eloped as pa o his wo k can be e ined and eleased as open sou ce so wa e o acili a ing communica ion be ween expe s in he heal hca e domain. The poli eness ules in oduced he e show ha in o de o make dialogue games mo e human-cen ed, we need o in oduce he same socie al machine y being employed in eal li e con e sa ion in o de o sa egua d he in eg i y o he dialogue in a compu a ional con ex . Acknowledgemen The wo k epo ed in his pape has been suppo ed in pa by he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme unde he Ma ie Skłodowska-Cu ie G an Ag eemen No. 860621, in pa by he p ojec 2021 SGR 00754 o he Ca alan Go e nmen , in pa by Chis -E a unde g an 2022/04/Y/ST6/00001, in pa by POB Cybe DS o Wa saw Uni e si y o Technology wi hin he Ex- cellence Ini ia i e: Resea ch Uni e si y (IDUB) p og amme unde g an 1820/1/Z01/POB3/2021, and in pa by VW ounda ion (VolkswagenS i ung) unde g an 98 542. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 387 Re e ences [1] J. Alama, A. Knoks and S.L. Uckelman, Dialogue games in classical logic, in: M. Giese, R. Kuzne s, eds, TABLEAUX 2011: Wo kshops, Tu o ials, and Sho Pape s, Technical Repo IAM-11-002, Uni e si ä Be n, Be n, 2011, pp. 82–86, a ailable om: h ps://doi.o g/10.5167/uzh-190012. [2] L. Amgoud, N. Maude and S. Pa sons, Modelling dialogues using a gumen a ion, in: P oceedings – 4 h In e na ional Con e ence on Mul iAgen Sys ems, ICMAS 2000, Ins i u e o Elec ical and Elec onics Enginee s Inc., 2000, pp. 31–38. [3] A. A ioua, P. Buche and M. C oi o u, Explana o y dialogues wi h a gumen a i e acul ies o e inconsis en knowledge bases, Expe Sys ems wi h Applica ions 80 (2017), 244–262. doi:10.1016/j.eswa.2017.03.009. [4] C.M. Ba num, Es ablishing he essen ials, in: Usabili y Tes ing Essen ials, C.M. Ba num, ed., 2nd edn, Mo gan Kau mann, 2021, pp. 9–33, a ailable om: h ps://www.sciencedi ec .com/science/a icle/pii/B9780128169421000010. doi:10.1016/ B978-0-12-816942-1.00001-0. [5] C.M. Ba num, 8, in: Analyzing he Findings, S. Me ken, ed., Else ie , 2021, pp. 287–319. [6] M. Be e idge and J. Fox, Au oma ic gene a ion o spoken dialogue om medical plans and on ologies, Jou nal o Biomed- ical In o ma ic 39(5) (2006), 482–499. doi:10.1016/j.jbi.2005.12.008. [7] F. Bex and H. P akken, In es iga ing s o ies in a o mal dialogue game, in: F on ie s in A i icial In elligence and Appli- ca ions., Vol. 172, IOS P ess, 2008, pp. 73–84. [8] E. Black and K. A kinson, Dialogues ha accoun o di e en pe spec i es in collabo a i e a gumen a ion, in: P oceed- ings o he 8 h In e na ional Con e ence on Au onomous Agen s and Mul iagen Sys ems, Vol. 2, 2009, pp. 867–874, a ailable om: pape s2://publica ion/uuid/851E4870-2CCD-4C21-BBAD-819066F85D11. [9] E. Black and A. Hun e , A gene a i e inqui y dialogue sys em, in: P oceedings o he In e na ional Con e ence on Au- onomous Agen s, ACM P ess, New Yo k, USA, 2007, pp. 1014–1021, a ailable om: h p://po al.acm.o g/ci a ion.c m? doid=1329125.1329417. [10] E. Black and A. Hun e , An inqui y dialogue sys em, Au onomous Agen s and Mul i-Agen Sys ems 19(2) (2009), 173–209, a ailable om: www.cossac.o g. [11] K. Budzynska, A. Rocci and O. Yasko ska, Financial dialogue games: A p o ocol o ea nings con e ence calls, in: F on- ie s in A i icial In elligence and Applica ions., Vol. 266, IOS P ess, 2014, pp. 19–30, a ailable om: h ps://ebooks. iosp ess.nl/doi/10.3233/978-1-61499-436-7-19. [12] P. Ca ayon and H.P. Human, Fac o s and usabili y o heal h in o ma ion echnology: Old and new challenges, Yea book o medical in o ma ics 28(1) (2019), 71–77, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /31419818/. doi:10.1055/s- 0039-1677907. [13] F. Cas agna, A. Ga on, P. McBu ney, S. Pa sons, I. Sassoon and E.I. Skla , EQRbo : A cha bo deli e ing EQR a gumen - based explana ions, F on ie s in A i icial In elligence 6(2023), a ailable om: h ps://www.ncbi.nlm.nih.go /pmc/ a icles/PMC10076765/. [14] B. Chaudh y, J. Wang, S. Wu, M. Maglione, W. Mojica, E. Ro h e al., Sys ema ic e iew: Impac o heal h in o ma ion echnology on quali y, e iciency, and cos s o medical ca e, Ame ican College o Physicians (2006). [15] R. C a en, F. Toni, C. Cada , A. Hadad and M. Williams, E icien a gumen a ion o medical decision-making, in: P o- ceedings o he Thi een h In e na ional Con e ence on P inciples o Knowledge Rep esen a ion and Reasoning. KR12, AAAI P ess, 2012, pp. 598–602. [16] G. Delaney, S. Jacob, R. Iedema, M. Win e s and M. Ba on, Compa ison o ace- o- ace and ideocon e enced mul i- disciplina y clinical mee ings, Aus alasian Radiology 48 (2004), 487–492, Wiley P ess, a ailable om: h ps://pubmed. ncbi.nlm.nih.go /15601329/. [17] L.A. Dennis and N. O en, Explaining BDI agen beha iou h ough dialogue, in: P oceedings o he In e na ional Join Con e ence on Au onomous Agen s and Mul iagen Sys ems, AAMAS, Vol. 1, In e na ional Founda ion o Au onomous Agen s and Mul iagen Sys ems (IFAAMAS), 2021, pp. 429–437. [18] P.M. Dung, On he accep abili y o a gumen s and i s undamen al ole in nonmono onic easoning, logic p og amming and n-pe son games, A i icial In elligence 77(2) (1995), 321–357, a ailable om: h p://linkinghub.else ie .com/ e ie e/ pii/000437029400041X. [19] M. Es e a, B. Rosell, J.A. Rod íguez-Aguila and J.L. A cos, AMELI: An agen -based middlewa e o elec onic in- s i u ions, in: P oceedings o he Thi d In e na ional Join Con e ence on Au onomous Agen s and Mul iagen Sys ems, AAMAS 2004, Vol. 1, 2004, pp. 236–243. [20] X. Fan, R. C a en, R. Singe , F. Toni and M. Williams, Assump ion-based a gumen a ion o decision-making wi h p e - e ences: A medical case s udy, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 8143, Sp inge , Be lin, Heidelbe g, 2013, pp. 374–390, a ailable om: h ps://link.sp inge .com/chap e /10.1007/978-3-642-40624-9_23. [21] G.E. G oss, The ole o he umo boa d in a communi y hospi al, CA: A Cance Jou nal o Clinicians 37(2) (1987), 88–92. [22] C.L. Hamblin, Fallacies, 1s edn, Me huen, London, UK, 1970. 388 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions [23] C.H. Hennessy and A. Walke , P omo ing mul i-disciplina y and in e -disciplina y ageing esea ch in he Uni ed Kingdom, Ageing and Socie y 31(1) (2011), 52–69, a ailable om: h ps://www.camb idge.o g/co e/jou nals/ageing- and-socie y/a icle/abs/p omo ing-mul idisciplina y-and-in e disciplina y-ageing- esea ch-in- he-uni ed-kingdom/ 044A8FFB174A3BF742EBA7F7DB3A4BD1. doi:10.1017/S0144686X1000067X. [24] J. Huang, N.R. Jennings and J. Fox, Coope a ion in dis ibu ed medical ca e, in: 2nd In . Con . on Coope a i e In o ma ion Sys ems (CoopIS-94) (01/01/94), 1994, pp. 255–263, a ailable om: h ps://ep in s.so on.ac.uk/252139/. [25] J. Huang, N.R. Jennings and J. Fox, An agen a chi ec u e o dis ibu ed medical ca e, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), Vol. 890, Sp inge Ve lag, 1995, pp. 219–232, a ailable om: h ps://link.sp inge .com/chap e /10.1007/3-540-58855-8_14. [26] ISO, ISO 9241-11:2018 – e gonomics o human–sys em in e ac ion – pa 11: Usabili y: De ini ions and concep s, 2018, a ailable om: h ps://www.iso.o g/s anda d/63500.h ml. [27] M. Janie , M. Snai h, K. Budzynska, J. Law ence and C. Reed, A sys em o dispu e media ion: The media ion dialogue game, F on ie s in A i icial In elligence and Applica ion 287 (2016), 351–358, a ailable om: h ps://doi.o g/doi:10. 3233/978-1-61499-686-6-351. [28] A.R. Kagan, The mul idisciplina y clinic, In e na ional Jou nal o Radia ion Oncology Biology Physics 61(4) (2005), 967–968. doi:10.1016/j.ij obp.2004.10.040. [29] N. Kokciyan, I. Sassoon, E. Skla , S. Modgil and S. Pa sons, Applying me ale el a gumen a ion amewo ks o suppo medical decision making, IEEE In elligen Sys ems 36(2) (2021), 64–71. doi:10.1109/MIS.2021.3051420. [30] A.M. Ku ahashi, J.N. S inson, M. an Wyk, S. Luca, T. Jamieson, P. Weins ein e al., The pe cei ed ease o use and use ulness o loop: E alua ion and con en analysis o a web-based clinical collabo a ion sys em, JMIR Human Fac o s 5(1) (2018), a ailable om: h ps://pubmed.ncbi.nlm.nih.go /29317386/. [31] B.W. Lamb, N. Se dalis, S. A o a, A. Pin o, C. Vincen and J.S.A. G een, Teamwo k and eam decision-making a mul i- disciplina y cance con e ences: Ba ie s, acili a o s, and oppo uni ies o imp o emen , Wo ld Jou nal o Su ge y 35(9) (2011), 1970–1976, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /21604049/. doi:10.1007/s00268-011-1152-1. [32] H.J. Lin, Y.L. Ko, C.F. Liu, C.J. Chen and J.J. Lin, De eloping and e alua ing a one-s op pa ien -cen e ed in e p o essional collabo a ion pla o m in Taiwan, Heal hca e (Swi ze land) 8(3) (2020), a ailable om: h ps://pubmed.ncbi.nlm.nih.go / 32751264/. [33] P. Liu, A. Lyndon, J.L. Holl, J. Johnson, K.Y. Bilimo ia and A.M. S ey, Ba ie s and acili a o s o in e disciplina y communica ion du ing consul a ions: A quali a i e s udy, BMJ Open 11(9) (2021), a ailable om: h ps://pubmed.ncbi. nlm.nih.go /34475150/. [34] N.J. Look Hong, A.R. Gaglia di, S.E. B onskill, L.F. Pasza and F.C. W igh , Mul idisciplina y cance con e ences: Explo ing obs acles and acili a o s o hei implemen a ion, Jou nal o oncology p ac ice 6(2) (2010), 61–68, a ail- able om: h p://www.ncbi.nlm.nih.go /pubmed/20592777 h p://www.pubmedcen al.nih.go /a icle ende . cgi?a id= PMC2835483. [35] E.J. Macaskill, S. Th ush, E.M. Walke and J.M. Dixon, Su geons’ iews on mul i-disciplina y b eas mee ings, Eu opean Jou nal o Cance 42(7) (2006), 905–908. doi:10.1016/j.ejca.2005.12.014. [36] P. Madumal, T. Mille , L. Sonenbe g and F. Ve e e, A g ounded in e ac ion p o ocol o explainable a i icial in elligence, in: P oceedings o he In e na ional Join Con e ence on Au onomous Agen s and Mul iagen Sys ems, AAMAS,Vol.2, 2019, pp. 1033–1041, a ailable om: h p://a xi .o g/abs/1903.02409. [37] M. Magui e and B. Delahun , Doing a hema ic analysis: A p ac ical, s ep-by-s ep guide o lea ning and eaching schola s, All I eland Jou nal o Highe Educa ion 9(3) (2017), a ailable om: h ps://ojs.aishe.o g/index.php/aishe-j/a icle/ iew/ 335. [38] A. Ma ínez-Ga cía, A. Mo eno-Conde, F. Jóda -Sánchez, S. Leal and C. Pa a, Sha ing clinical decisions o mul imo - bidi y case managemen using social ne wo k and open-sou ce ools, Jou nal o Biomedical In o ma ics 46(6) (2013), 977–984. doi:10.1016/j.jbi.2013.06.007. [39] N. Maude , S. Pa sons and I. Rahwan, A gumen a ion in mul i-agen sys ems: Con ex and ecen de elopmen s, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 4766, Sp inge Ve lag, 2007, pp. 1–16. [40] P. McBu ney, R.M. Van Eijk, S. Pa sons and L. Amgoud, A dialogue game p o ocol o agen pu chase nego ia ions, Au onomous Agen s and Mul i-Agen Sys ems 7(3) (2003), 235–273. doi:10.1023/A:1024787301515. [41] S. Modgil, P. Tolchinsky and U. Co és, Towa ds o malising agen a gumen a ion o e he iabili y o human o gans o ansplan a ion, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 3789, Sp inge , Be lin, Heidelbe g, 2005, pp. 928–938, a ailable om: h ps://link.sp inge .com/chap e /10.1007/11579427_95. [42] R.S. Mo se, K. Lambden, E. Quinn, T. Ngoma, B. Mushi, Y.X. Ho e al., A mobile app o imp o e symp om con ol and in o ma ion exchange among specialis s and local heal h wo ke s ea ing Tanzanian cance pa ien s: Human-cen e ed design app oach, JMIR Cance 7(1) (2021), a ailable om: h ps://www.ncbi.nlm.nih.go /pmc/a icles/PMC8088847/. Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions 389 [43] A.J. Mun o and S. Swa zman, Wha is a i ual mul idisciplina y eam ( MDT)?, B i ish Jou nal o Cance 108(12) (2013), 2433–2441, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /23756866/. doi:10.1038/bjc.2013.231. [44] S. Na eed, T. Donke s and J. Ziegle , A gumen a ion-based explana ions in ecommende sys ems: Concep ual amewo k and empi ical esul s, in: UMAP 2018 – Adjunc Publica ion o he 26 h Con e ence on Use Modeling, Adap a ion and Pe sonaliza ion, Associa ion o Compu ing Machine y, Inc., New Yo k, NY, USA, 2018, pp. 293–298, a ailable om: h ps://dl.acm.o g/doi/10.1145/3213586.3225240. doi:10.1145/3213586.3225240. [45] V. Ngo, C.G. Ma sumo o, J.G. Joseph, J.F. Bell, R.J. Bold, A. Da is e al., The pe sonal heal h ne wo k mobile app o chemo he apy ca e coo dina ion: Quali a i e e alua ion o a andomized clinical ial, JMIR mHeal h and uHeal h 8(5) (2020), e16527, a ailable om: h ps://mheal h.jmi .o g/2020/5/e16527. [46] J.X. Nie, C. Heideb ech , A. Ze le , J. Pea ce, R. Cunha, S. Quan e al., The pe cei ed ease o use and pe cei ed use ulness o a web-based in e p o essional communica ion and collabo a ion pla o m in he hospi al se ing: In e iew s udy wi h heal h ca e p o ide s, JMIR Human Fac o s 10(1) (2023), e39051, a ailable om: h ps://human ac o s.jmi .o g/2023/1/ e39051. [47] J. Nielsen, Why you only need o es wi h 5 use s, 2000, a ailable om: h ps://www.nng oup.com/a icles/why-you- only-need- o- es -wi h-5-use s/. [48] S. Pa sons, M. Wooldkidge and L. Amgoud, P ope ies and complexi y o some o mal in e -agen dialogues, Jou nal o Logic and Compu a ion 13(3) (2003), 347–376. doi:10.1093/logcom/13.3.347. [49] V. Pa ka , D. Acos a, T. Da idson, A. Jones, J. Fox and M. Kesh ga , Using compu e ised decision suppo o imp o e compliance o cance mul idisciplina y mee ings wi h e idence-based guidance, BMJ Open (2012), e000439, a ailable om: h p://dx.doi.o g/10.1136/bmjopen-2011-000439. [50] J. Pi and A. Mamdani, Communica ion p o ocols in mul i-agen sys ems: A de elopmen me hod and e e ence a chi- ec u e, in: Issues in Agen Communica ion, Sp inge , Be lin, Heidelbe g, 2000, pp. 160–177, a ailable om: h ps://link. sp inge .com/chap e /10.1007/10722777_11. doi:10.1007/10722777_11. [51] H. P akken, Rela ing p o ocols o dynamic dispu e wi h logics o de easible a gumen a ion, Syn hese 127(1–2) (2001), 187–219. doi:10.1023/A:1010322504453. [52] H. P akken, Cohe ence and lexibili y in dialogue games o a gumen a ion, Jou nal o Logic and Compu a ion 15(6) (2005), 1009–1040, a ailable om: h p://academic.oup.com/logcom/a icle/15/6/1009/1086845/Cohe ence-and- Flexibili y-in-Dialogue-Games- o . doi:10.1093/logcom/exi046. [53] H. P akken, A o mal model o adjudica ion dialogues, A i icial In elligence and Law 16(3) (2008), 305–328, a ailable om: h ps://link.sp inge .com/a icle/10.1007/s10506-008-9066-4. doi:10.1007/s10506-008-9066-4. [54] H. P akken and R. Ra sma, A op-le el model o case-based a gumen a ion o explana ion: Fo malisa ion and expe i- men s, A gumen & Compu a ion 13 (2022), 159–194. [55] M. Quinn, J. Fo man, M. Ha od, S. Win e , K.E. Fowle , S.L. K ein e al., Elec onic heal h eco ds, communica ion, and da a sha ing: Challenges and oppo uni ies o imp o ing he diagnos ic p ocess, Diagnosis 6(3) (2019), 241–248, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /30485175/. doi:10.1515/dx-2018-0036. [56] A. Rago, H. Li and F. Toni, In e ac i e explana ions by con lic esolu ion ia a gumen a i e exchanges, 2023, a Xi :2303. 15022. [57] R.B. Rajaseka an, D. Whi well, T.D.A. Coske , C.L.M.H. Gibbons and A. Ca , Will i ual mul idisciplina y eam mee - ings become he no m o musculoskele al oncology ca e ollowing he COVID-19 pandemic? – Expe ience om a e - ia y sa coma cen e, BMC Musculoskele al Diso de s 22(1) (2021), 18, a ailable om: h ps://pubmed.ncbi.nlm.nih.go / 33402136/. [58] L. Riley, K. A kinson, T. Payne and E. Black, An implemen ed dialogue sys em o inqui y and pe suasion, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma - ics), LNAI, Vol. 7132, Sp inge , Be lin, Heidelbe g, 2012, pp. 67–84, a ailable om: h ps://link.sp inge .com/chap e / 10.1007/978-3-642-29184-5_5. [59] I. Sassoon, N. Kökciyan, S. Modgil and S. Pa sons, A gumen a ion schemes o clinical decision suppo , A gumen and Compu a ion 12(3) (2021), 329–355. doi:10.3233/AAC-200550. [60] I. Sassoon, N. Kökciyan, E. Skla and S. Pa sons, Explainable a gumen a ion o wellness consul a ion, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 11763, Sp inge Ve lag, 2019, pp. 186–202, a ailable om: h ps://doi.o g/10.1007/978-3-030-30391-4_11. [61] B. Schellenbe ge , A. Diekmann, C. Heuse , N. Gambashidze, N. E ns mann and L. Ansmann, Decision-making in mul idisciplina y umo boa ds in b eas cance ca e – an obse a ional s udy, Jou nal o Mul idisciplina y Heal hca e 14 (2021), 1275–1284, a ailable om: /pmc/a icles/PMC8179814/ /pmc/a icles/PMC8179814/? epo =abs ac h ps:// www.ncbi.nlm.nih.go /pmc/a icles/PMC8179814/. doi:10.2147/JMDH.S300061. [62] J.R. Sea le, Speech Ac s, Camb idge Uni e si y P ess, 1969, a ailable om: h ps://www.camb idge.o g/co e/p oduc / iden i ie /9781139173438/ ype/book. 390 Q. Shaheen e al. / Dialogue game o expe collabo a i e ecommenda ions [63] Q. Shaheen, A. Toniolo and J.K.F. Bowles, A gumen a ion-based explana ions o mul imo bidi y ea men plans, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 12568, Sp inge Science and Business Media Deu schland GmbH, 2021, pp. 394–402, a ail- able om: h ps://doi.o g/10.1007/978-3-030-69322-0_29. [64] M. Snai h, D. De F anco, T. Beinema, H. Op Den Akke and A. Pease, A dialogue game o mul i-pa y goal-se ing in heal h coaching, in: F on ie s in A i icial In elligence and Applica ions., Vol. 305, IOS P ess, 2018, pp. 337–344, a ailable om: h ps://ebooks.iosp ess.nl/doi/10.3233/978-1-61499-906-5-337. [65] I. S epin, K. Budzynska, A. Ca ala, M. Pe ei a-Fa iña and J.M. Alonso-Mo al, In o ma ion-seeking dialogue o explain- able a i icial in elligence: Modelling and analy ics, A gumen & Compu a ion (2023), 1–59, p ep in . [66] K.M. Su cli e, E. Lew on and M.M. Rosen hal, Communica ion ailu es: An insidious con ibu o o medical mishaps, Academic Medicine 79(2) (2004), 186–194, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /14744724/. doi:10.1097/ 00001888-200402000-00019. [67] M. Tabe na, F.G. Moncayo, E. Jané-Salas, M. An onio, L. A ibas, E. Vilajosana e al., The Mul idisciplina y Team (MDT) App oach and Quali y o Ca e, F on ie s Media S.A., 2020, a ailable om: /pmc/a icles/PMC7100151/ /pmc/a - icles/PMC7100151/? epo =abs ac h ps://www.ncbi.nlm.nih.go /pmc/a icles/PMC7100151/. [68] T. Tang, C. Heideb ech , A. Cobu n, E. Mans ield, E. Robe o, E. Lucez e al., Using an elec onic ool o imp o e eamwo k and in e p o essional communica ion o mee he needs o complex hospi alized pa ien s: A mixed me hods s udy, In e na ional Jou nal o Medical In o ma ics 127 (2019), 35–42. doi:10.1016/j.ijmedin .2019.04.010. [69] T. Tang, M.E. Lim, E. Mans ield, A. McLachlan and S.D. Quan, Clinician use in ol emen in he eal wo ld: Designing an elec onic ool o imp o e in e p o essional communica ion and collabo a ion in a hospi al se ing, In e na ional Jou nal o Medical In o ma ics 110 (2018), 90–97. doi:10.1016/j.ijmedin .2017.11.011. [70] P. Tolchinsky, K. A kinson, P. Mcbu ney, S. Modgil and U. Co és, Agen s delibe a ing o e ac ion p oposals using he P oCLAIM model, in: Lec u e No es in Compu e Science (Including Subse ies Lec u e No es in A i icial In elligence and Lec u e No es in Bioin o ma ics), LNAI, Vol. 4696, Sp inge Ve lag, 2007, pp. 32–41, a ailable om: h ps://link. sp inge .com/chap e /10.1007/978-3-540-75254-7_4. [71] P. Tolchinsky, U. Co és, S. Modgil, F. Caballe o and A. López-Na idad, Inc easing human-o gan ansplan a ailabili y: A gumen a ion-based agen delibe a ion, Ins i u e o Elec ical and Elec onics Enginee s Inc. (2006). [72] I. Va gas, I. Ga cia-Subi a s, A.S. Mogollón-Pé ez, M. Fe ei a-De-Medei os-Mendes, P. Eguigu en, A.I. Cisne os e al., Unde s anding communica ion b eakdown in he ou pa ien e e al p ocess in La in Ame ica: A c oss-sec ional s udy on he use o clinical co espondence in public heal hca e ne wo ks o six coun ies, Heal h Policy and Planning 33(4) (2018), 494–504, a ailable om: h ps://pubmed.ncbi.nlm.nih.go /29452401/. doi:10.1093/heapol/czy016. [73] S.L. Vasileiou, A. Kuma , W. Yeoh, T.C. Son and F. Toni, DR-HAI: A gumen a ion-based dialec ical econcilia ion in human–AI in e ac ions, 2023, a Xi :2306.14694. [74] A. Vassiliades, N. Bassiliades and T. Pa kos, A gumen a ion and Explainable A i icial In elligence: A Su ey, Camb idge Uni e si y P ess, 2021, a ailable om: h ps://doi.o g/10.1017/S0269888921000011. [75] D. Wal on, A gumen a ion heo y: A e y sho in oduc ion, in: A gumen a ion in A i icial In elligence, Sp inge US, Bos on, MA, 2009, pp. 1–22, a ailable om: h p://link.sp inge .com/10.1007/978-0-387-98197-0_1. [76] D. Wal on, A dialogue sys em speci ica ion o explana ion, Syn hese 182(3) (2011), 349–374, a ailable om: h ps://link. sp inge .com/a icle/10.1007/s11229-010-9745-z. [77] D. Wal on and E.C.W. K abbe, Commi men in Dialogue: Basic Concep s o In e pe sonal Reasoning, S a e Uni e si y o New Yo k P ess, Albany NY, 1995, a ailable om: h ps://books.google.com.pk/books?id=6nU8TpVmW08C. [78] F.C. W igh , C. De Vi o, B. Lange and A. Hun e , Mul idisciplina y cance con e ences: A sys ema ic e iew and de el- opmen o p ac ice s anda ds, Eu opean Jou nal o Cance 43(6) (2007), 1002–1010. doi:10.1016/j.ejca.2007.01.025. [79] F.C. W igh , N. Look Hong, D. U bach, D. Da is, R.S. McLeod and A.R. Gaglia di, Mul idisciplina y cance con e ences: Iden i ying oppo uni ies o p omo e implemen a ion, Annals o Su gical Oncology 16(10) (2009), 2731–2737, a ailable om: h ps://link.sp inge .com/a icle/10.1245/s10434-009-0639-6. [80] R. Wu, P. Rossos, S. Quan, S. Ree es, V. Lo, B. Wong e al., An e alua ion o he use o sma phones o communica e be ween clinicians: A mixed-me hods s udy, Jou nal o Medical In e ne Resea ch 13(3) (2011), a ailable om: h ps:// pubmed.ncbi.nlm.nih.go /21875849/. [81] L. Xiao, K. Hu and J. Fox, A g oup decision desc ip ion language and i s clinical applica ion, in: ACM In e na ional Con e ence P oceeding Se ies, Associa ion o Compu ing Machine y, 2021, pp. 160–167, a ailable om: h ps://dl.acm. o g/doi/10.1145/3488838.3488866. [82] C. Zheng, X. Chen, L. Weng, L. Guo, H. Xu, M. Lin e al., Bene i s o Mobile Apps o Cance Pain Managemen : Sys ema ic Re iew, JMIR Publica ions Inc., 2020, a ailable om: h ps://mheal h.jmi .o g/2020/1/e17055.