/Users/andrea/_magisterarbeit/korpus/clean/testkorpus/1/file6.html NN ----------------------------------------- : Is VBZ it PP an DT Agent NP , , or CC just RB a DT Program NN . SENT . SENT A DT Taxonomy NN for IN Autonomous JJ Agents NPS Stan NP Franklin NP and CC Art NP Graesser NP Institute NP for IN Intelligent JJ Systems NPS University NP of IN Memphis NP Proceedings NNS of IN the DT Third NP International NP Workshop NN on IN Agent NP Theories NNS , , Architectures NNS , , and CC Languages NNS , , Springer NP Verlag NP , , 1996 CD . SENT Abstract JJ The DT advent NN of IN software NN agents NNS gave VVD rise NN to TO much JJ discussion NN of IN just RB what WP such PDT an DT agent NN is VBZ , , and CC of IN how WRB they PP differ VVP from IN programs NNS in IN general NN . SENT Here RB we PP propose VVP a DT formal JJ definition NN of IN an DT autonomous JJ agent NN which WDT clearly RB distinguishes VVZ a DT software NN agent NN from IN just RB any DT program NN . SENT We PP also RB offer VVP the DT beginnings NNS of IN a DT natural JJ kinds NNS taxonomy NN of IN autonomous JJ agents NNS , , and CC discuss VV possibilities NNS for IN further JJR classification NN . SENT Finally RB , , we PP discuss VVP subagents NNS and CC multiagent JJ systems NNS . SENT Introduction NN On IN meeting VVG a DT friend NN or CC colleague NN that IN we PP haven't VVD seen VVN for IN a DT while NN , , or CC a DT new JJ acquaintance NN , , some DT version NN of IN the DT following VVG conversation NN often RB ensues VVZ . SENT What WP are VBP you PP working VVG on IN these DT days NNS . SENT Control NN structures NNS for IN autonomous JJ agents NNS . SENT Autonomous JJ agents NNS . SENT What WP do VVP you PP mean VVP by IN that DT . SENT A DT brief JJ explanation NN is VBZ then RB followed VVN by IN . SENT But CC agents NNS sound VVP just RB like VV computer NN programs NNS . SENT How WRB are VBP they PP different JJ . SENT This DT elicits VVZ a DT more RBR satisfying JJ explanation NN that WDT distinguishes VVZ between IN agent NN and CC program NN . SENT The DT nature NN of IN this DT more RBR satisfying JJ explanation NN motivates VVZ this DT essay NN . SENT After IN a DT review NN of IN some DT of IN the DT many JJ ways NNS the DT term NN agent NN has VHZ been VBN used VVN within IN the DT context NN of IN autonomous JJ agents NNS , , we'll JJ propose VVP and CC defend VVP a DT notion NN of IN agent NN that WDT is VBZ clearly RB distinct JJ from IN a DT program NN . SENT This DT discussion NN will MD lead VV us PP to TO a DT discussion NN of IN possible JJ classifications NNS for IN autonomous JJ agents NNS . SENT What WP is VBZ an DT agent NN . SENT Workers NNS involved VVN in IN agent NN research NN have VHP offered VVN a DT variety NN of IN definitions NNS , , each DT hoping VVG to TO explicate VV his PP$ or CC her PP$ use NN of IN the DT word NN agent NN . SENT These DT definitions NNS range VVP from IN the DT simple NN to TO the DT lengthy JJ and CC demanding JJ . SENT We PP suspect VVP that IN each DT of IN them PP grew VVD directly RB out RB of IN the DT set NN of IN examples NNS of IN agents NNS that IN the DT definer NN had VHD in IN mind NN . SENT This DT is VBZ certainly RB the DT case NN for IN the DT version NN we'll NN propose VVP below RB . SENT Let's NNS orient VV ourselves PP by IN examining VVG and CC comparing VVG some DT of IN these DT definitions NNS . SENT The DT MuBot NP Agent NP http NN . SENT www JJ . SENT crystaliz NP . SENT com NN logicware NN mubot NN . SENT html NN The DT term NN agent NN is VBZ used VVN to TO represent VV two CD orthogonal JJ concepts NNS . SENT The DT first JJ is VBZ the DT agent's JJ ability NN for IN autonomous JJ execution NN . SENT The DT second NN is VBZ the DT agent's JJ ability NN to TO perform VV domain NN oriented JJ reasoning NN . SENT P NN This DT pointer NN at IN definitions NNS come VVN from IN an DT online JJ white JJ paper NN by IN Sankar NP Virdhagriswaran NP of IN Crystaliz NP , , Inc NP . SENT , , defining VVG mobile JJ agent NN technology NN . SENT Autonomous JJ execution NN is VBZ clearly RB central JJ to TO agency NN . SENT The DT AIMA NP Agent NP Russell NP and CC Norvig NP 1995 CD , , page NN 33 CD An DT agent NN is VBZ anything NN that WDT can MD be VB viewed VVN as IN perceiving VVG its PP$ environment NN through IN sensors NNS and CC acting VVG upon IN that DT environment NN through IN effectors NNS . SENT AIMA NP is VBZ an DT acronym NN for IN Artificial JJ Intelligence NP . SENT a DT Modern NP Approach NN , , a DT remarkably RB successful JJ new JJ AI VVZ text NN that WDT was VBD used VVN in IN 200 CD colleges NNS and CC universities NNS in IN 1995 CD . SENT The DT authors NNS were VBD interested VVN in IN software NN agents NNS embodying VVG AI VVP techniques NNS . SENT Clearly RB , , the DT AIMA NP definition NN depends VVZ heavily RB on IN what WP we PP take VVP as IN the DT environment NN , , and CC on IN what WDT sensing VVG and CC acting VVG mean NN . SENT If IN we PP define VV the DT environment NN as IN whatever WDT provides VVZ input NN and CC receives VVZ output NN , , and CC take VV receiving VVG input NN to TO be VB sensing VVG and CC producing VVG output NN to TO be VB acting VVG , , every DT program NN is VBZ an DT agent NN . SENT Thus RB , , if IN we PP want VVP to TO arrive VV at IN a DT useful JJ contrast NN between IN agent NN and CC program NN , , we PP must MD restrict VV at IN least JJS some DT of IN the DT notions NNS of IN environment NN , , sensing VVG and CC acting VVG . SENT The DT Maes NP Agent NP Maes NP 1995 CD , , page NN 108 CD Autonomous JJ agents NNS are VBP computational JJ systems NNS that WDT inhabit VV some DT complex JJ dynamic JJ environment NN , , sense NN and CC act VV autonomously RB in IN this DT environment NN , , and CC by IN doing VVG so RB realize VV a DT set NN of IN goals NNS or CC tasks NNS for IN which WDT they PP are VBP designed VVN . SENT Pattie NP Maes NP , , of IN MIT's NP Media NP Lab NP , , is VBZ one CD of IN the DT pioneers NNS of IN agent NN research NN . SENT She PP adds VVZ a DT crucial JJ element NN to TO her PP$ definition NN of IN an DT agent NN . SENT agents NNS must MD act VV autonomously RB so RB as IN to TO realize VV a DT set NN of IN goals NNS . SENT Also RB environments NNS are VBP restricted JJ to TO being VBG complex JJ and CC dynamic JJ . SENT It's NP not RB clear JJ whether IN this DT rules NNS out IN a DT payroll NN program NN without IN further JJR restrictions NNS . SENT The DT KidSim NP Agent NP Smith NP , , Cypher NP and CC Spohrer NP 1994 CD Let VV us PP define VV an DT agent NN as IN a DT persistent JJ software NN entity NN dedicated VVN to TO a DT specific JJ purpose NN . SENT Persistent JJ distinguishes VVZ agents NNS from IN subroutines NNS . SENT agents NNS have VHP their PP$ own JJ ideas NNS about IN how WRB to TO accomplish VV tasks NNS , , their PP$ own JJ agendas NNS . SENT Special JJ purpose NN distinguishes VVZ them PP from IN entire JJ multifunction NN applications NNS . SENT agents NNS are VBP typically RB much RB smaller JJR . SENT The DT authors NNS are VBP with IN Apple NP . SENT The DT explicit JJ requirement NN of IN persistence NN is VBZ a DT new JJ and CC important JJ addition NN here RB . SENT Though IN many JJ agents NNS are VBP special JJ purpose NN we PP suspect VVP this DT is VBZ not RB an DT essential JJ feature NN of IN agency NN . SENT The DT Hayes NP Roth NP Agent NP Hayes NP Roth NP 1995 CD Intelligent JJ agents NNS continuously RB perform VVP three CD functions NNS . SENT perception NN of IN dynamic JJ conditions NNS in IN the DT environment NN . SENT action NN to TO affect VV conditions NNS in IN the DT environment NN . SENT and CC reasoning NN to TO interpret VV perceptions NNS , , solve VV problems NNS , , draw VV inferences NNS , , and CC determine VV actions NNS . SENT Barbara NP Hayes NP Roth NP of IN Stanford's NP Knowledge NN Systems NP Laboratory NP insists VVZ that DT agents NNS reason NN during IN the DT process NN of IN action NN selection NN . SENT If IN reasoning NN is VBZ interpreted VVN broadly RB , , her PP$ agent NN architecture NN does VVZ allow VV for IN reflex JJ actions NNS as RB well RB as IN planned JJ actions NNS . SENT The DT IBM NP Agent NP http NN . SENT activist NN . SENT gpl NN . SENT ibm NN . SENT com NN . SENT 81 CD WhitePaper NP ptc NN 2 CD . SENT htm JJ Intelligent JJ agents NNS are VBP software NN entities NNS that WDT carry VVP out RP some DT set NN of IN operations NNS on IN behalf NN of IN a DT user NN or CC another DT program NN with IN some DT degree NN of IN independence NN or CC autonomy NN , , and CC in IN so RB doing VVG , , employ VV some DT knowledge NN or CC representation NN of IN the DT user's JJ goals NNS or CC desires NNS . SENT This DT definition NN , , from IN IBM's NP Intelligent JJ Agent NP Strategy NP white JJ paper NN , , views VVZ an DT intelligent JJ agent NN as IN acting VVG for IN another DT , , with IN authority NN granted VVN by IN the DT other JJ . SENT A DT typical JJ example NN might MD be VB an DT information NN gathering NN agent NN , , though IN the DT white JJ paper NN talks NNS of IN eight CD possible JJ applications NNS . SENT Would MD you PP stretch VV some DT degree NN of IN independence NN to TO include VV a DT payroll NN program NN . SENT What WP if IN it PP called VVD itself PP on IN a DT certain JJ day NN of IN the DT month NN . SENT The DT Wooldridge NP shyp NN . SENT Jennings NP Agent NP Wooldridge NP and CC Jennings NP 1995 CD , , page NN 2 CD . SENT . SENT . SENT a DT hardware NN or CC more RBR usually RB software NN based VVN computer NN system NN that WDT enjoys VVZ the DT following VVG properties NNS . SENT autonomy NN . SENT agents NNS operate VVP without IN the DT direct JJ intervention NN of IN humans NNS or CC others NNS , , and CC have VH some DT kind NN of IN control NN over IN their PP$ actions NNS and CC internal JJ state NN . SENT social JJ ability NN . SENT agents NNS interact VV with IN other JJ agents NNS and CC possibly RB humans NNS via IN some DT kind NN of IN agent NN communication NN language NN . SENT reactivity NN . SENT agents NNS perceive VVP their PP$ environment NN , , which WDT may MD be VB the DT physical JJ world NN , , a DT user NN via IN a DT graphical JJ user NN interface NN , , a DT collection NN of IN other JJ agents NNS , , the DT INTERNET NN , , or CC perhaps RB all RB of IN these DT combined JJ , , and CC respond VV in IN a DT timely JJ fashion NN to TO changes NNS that WDT occur VVP in IN it PP . SENT pro JJ activeness NN . SENT agents NNS do VVP not RB simply RB act VV in IN response NN to TO their PP$ environment NN , , they PP are VBP able JJ to TO exhibit NN goal NN directed VVD behavour NN by IN taking VVG the DT initiative NN . SENT The DT Wooldridge NP and CC Jennings NP definition NN , , in IN addition NN to TO spelling VVG out RP autonomy NN , , sensing VVG and CC acting VVG , , allows VVZ for IN a DT broad JJ , , but CC finite JJ , , range NN of IN environments NNS . SENT They PP further RBR add VV a DT communications NNS requirement NN . SENT What WP would MD be VB the DT status NN of IN a DT payroll NN program NN with IN a DT graphical JJ interface NN and CC a DT decidedly RB primitive JJ communication NN language NN . SENT The DT SodaBot NP Agent NP Michael NP Coen NP http NN . SENT www JJ . SENT ai VVP . SENT mit NP . SENT edu NN people NNS sodabot NN slideshow NN total JJ P NN 001 CD . SENT html NP Software NP agents NNS are VBP programs NNS that WDT engage VVP in IN dialogs NNS and CC negotiate VV and CC coordinate VV transfer NN of IN information NN . SENT SodaBot NP is VBZ a DT development NN environment NN for IN software NN agent NN being VBG constructed VVN at IN the DT MIT NP AI VVP Lab NP by IN Michael NP Coen NP . SENT Note VV the DT apparently RB almost RB empty JJ intersection NN between IN this DT definition NN and CC the DT preceding JJ seven CD . SENT we PP say VVP apparently RB since IN negotiating VVG , , for IN example NN , , requires VVZ both DT sensing VVG and CC acting VVG . SENT And CC dialoging VVG requires VVZ communication NN . SENT Still RB the DT feeling NN of IN this DT definition NN is VBZ vastly RB different JJ from IN the DT first JJ few JJ , , and CC would MD seem VV to TO rule VV out RP almost RB all DT standard JJ programs NNS . SENT The DT Foner NP Agent NP Lenny NP Foner NP Download NP from IN ftp NN . SENT media NNS . SENT mit NP . SENT edu NN pub NN Foner NP Papers NP Julia NP Agents NPS Julia NP . SENT ps NNS or CC online JJ at IN http NN . SENT foner NP . SENT www JJ . SENT media NNS . SENT mit NP . SENT edu NN people NNS foner NP Julia NP click NN on IN What's NP an DT agent NN . SENT Crucial JJ notions NNS Foner NP requires VVZ much RB more JJR of IN an DT agent NN . SENT His PP$ agents NNS collaborate VVP with IN their PP$ users NNS to TO improve VV the DT accomplishment NN of IN the DT users NNS tasks NNS . SENT This DT requires VVZ , , in IN addition NN to TO autonomy NN , , that IN the DT agent NN dialog NN with IN the DT user NN , , be VB trustworthy JJ , , and CC degrade VV gracefully RB in IN the DT face NN of IN a DT communications NNS mismatch NN . SENT However RB , , this DT quick JJ paraphrase NN doesn't NN do VVP justice NN to TO Foner's NP analysis NN . SENT The DT Brustoloni NP Agent NP Brustoloni NP 1991 CD , , Franklin NP 1995 CD , , p NN . SENT 265 CD Autonomous JJ agents NNS are VBP systems NNS capable JJ of IN autonomous JJ , , purposeful JJ action NN in IN the DT real JJ world NN . SENT The DT Brustoloni NP agent NN , , unlike IN the DT prior JJ agents NNS , , must MD live VV and CC act VV in IN the DT real JJ world NN . SENT This DT definition NN excludes VVZ software NN agents NNS and CC programs NNS in IN general NN . SENT Brustoloni NP also RB insists VVZ that IN his PP$ agents NNS be VB reactive JJ shyp NN . SENT that DT is VBZ , , be VB able JJ to TO respond VV to TO external JJ , , asynchronous JJ stimuli NNS in IN a DT timely JJ fashion NN . SENT As IN these DT definitions NNS make VVP clear JJ , , there's RB no DT general JJ agreement NN as IN to TO what WP constitutes VVZ an DT agent NN , , or CC as IN to TO how WRB agents NNS differ VVP from IN programs NNS . SENT The DT Software NP Agents NPS Mailing NN List NP on IN the DT Internet NN provides VVZ a DT FAQ NP frequently RB asked VVD questions NNS that WDT says VVZ , , The DT FAQ NP Agent NP http NN . SENT www JJ . SENT ee NNS . SENT mcgill NP . SENT ca MD . SENT 80 CD belmarc NN agent NN faq NP . SENT html NN This DT FAQ NP will MD not RB attempt VV to TO provide VV an DT authoritative JJ definition NN . SENT . SENT . SENT It PP does VVZ provide VV a DT list NN of IN attributes NNS often RB found VVD in IN agents NNS . SENT Autonomous JJ , , goal NN oriented JJ , , collaborative JJ , , flexible JJ , , self NN starting NN , , temporal JJ continuity NN , , character NN , , communicative JJ , , adaptive JJ , , mobile JJ , , Etzioni NP and CC Weld NP . SENT Several JJ of IN these DT would MD seem VV to TO rule VV out RP our PP$ payroll NN program NN . SENT The DT Essence NN of IN Agency NP We PP normally RB avoid VV prescriptive JJ arguments NNS about IN how WRB a DT word NN should MD be VB used VVN . SENT Russell NP and CC Norvig NP put VVD it PP this DT way NN . SENT The DT notion NN of IN an DT agent NN is VBZ meant VVN to TO be VB a DT tool NN for IN analyzing VVG systems NNS , , not RB an DT absolute JJ characterization NN that WDT divides VVZ the DT world NN into IN agents NNS and CC non JJ agents NNS . SENT 1995 CD , , page NN 33 CD The DT only JJ concepts NNS that WDT yield VVP sharp JJ edge NN categories NNS are VBP mathematical JJ concepts NNS , , and CC they PP succeed VVP only RB because IN they PP are VBP content NN free JJ . SENT Agents NNS live VVP in IN the DT real JJ world NN or CC some DT world NN , , and CC real JJ world NN concepts NNS yield VVP fuzzy JJ categories NNS . SENT Nevertheless RB , , we PP will MD propose VV a DT mathematical JJ style NN definition NN of IN an DT autonomous JJ agent NN , , knowing VVG full JJ well RB that IN it PP must MD fail VV around RP the DT edges NNS . SENT Our PP$ definition NN attempts VVZ to TO capture VV the DT essence NN of IN being VBG an DT agent NN , , and CC to TO define VV the DT broadest JJS class NN of IN agents NNS . SENT Further JJR restrictions NNS can MD then RB be VB added VVN to TO define VV more JJR particular JJ classes NNS of IN agents NNS . SENT Ideally RB , , such PDT an DT endeavor NN would MD produce VV a DT nomenclature NN of IN agents NNS that WDT could MD be VB used VVN relatively RB unambiguously RB by IN researchers NNS in IN the DT field NN , , resulting VVG in IN clearer JJR communications NNS . SENT The DT definitions NNS of IN the DT previous JJ section NN seem VVP to TO derive VV from IN one CD or CC both DT of IN two CD common JJ uses NNS of IN the DT word NN agent NN . SENT 1 CD one NN who WP acts VVZ , , or CC who WP can MD act VV , , and CC 2 CD one NN who WP acts VVZ in IN place NN of IN another DT with IN permission NN . SENT Since IN one CD who WP acts VVZ in IN place NN of IN acts NNS , , the DT second JJ usage NN requires VVZ the DT first JJ . SENT Hence RB , , let's NNS go VVP for IN a DT definition NN of IN the DT first JJ notion NN . SENT What WP are VBP examples NNS of IN agents NNS in IN this DT first JJ sense NN upon IN which WDT we PP can MD build VV our PP$ mathematical JJ style NN definition NN . SENT Well RB , , humans NNS act VVP , , as RB do VVP most RBS other JJ animals NNS . SENT I PP say VVP most JJS since IN some DT animals NNS act VVP during IN a DT portion NN of IN their PP$ lives NNS and CC not RB during IN others NNS , , for IN example NN the DT sea NN squirt NN Dethie NP 1986 CD . SENT Also RB , , some DT autonomous JJ mobile JJ robots NNS act NN , , for IN example NN Brooks NP Herbert NP Brooks NP 1990 CD , , p NN . SENT 8 CD . SENT Franklin NP 1995 CD , , p NN 263 CD 5 CD . SENT All DT of IN these DT are VBP real JJ world NN agents NNS . SENT Software NN agents NNS live VVP in IN computer NN operating NN systems NNS , , databases NNS , , networks NNS , , MUDs NNS , , etc FW . SENT Almost RB all PDT the DT definitions NNS in IN the DT previous JJ section NN refer VVP to TO software NN agents NNS . SENT Finally RB , , artificial JJ life NN agents NNS live VVP in IN artificial JJ environments NNS on IN a DT computer NN screen NN or CC in IN its PP$ memory NN Langton NP 1989 CD , , Franklin NP 1995 CD , , pp NP . SENT 185 CD 208 CD . SENT What WP do VVP these DT agents NNS share NN that WDT constitutes VVZ the DT essence NN of IN being VBG an DT agent NN . SENT Each DT is VBZ situated VVN in IN , , and CC is VBZ a DT part NN on IN some DT environment NN . SENT Each DT senses NNS its PP$ environment NN and CC act VV autonomously RB upon IN it PP . SENT No DT other JJ entity NN is VBZ required VVN to TO feed VV it PP input NN , , or CC to TO interpret VV and CC use VV its PP$ output NN . SENT Each DT acts NNS in IN pursuit NN of IN it's JJ own JJ agenda NN , , whether IN satisfying VVG evolved VVN drives NNS as IN in IN humans NNS and CC animals NNS , , or CC pursuing VVG goals NNS designed VVN in IN by IN some DT other JJ agent NN , , as RB in IN software NN agents NNS . SENT Artificial JJ life NN agents NNS may MD be VB of IN either DT variety NN . SENT Each DT acts NNS so RB that IN its PP$ current JJ actions NNS may MD effect VV its PP$ later JJR sensing VVG , , that WDT is VBZ its PP$ actions NNS effect VV its PP$ environment NN . SENT Finally RB , , each DT acts NNS continually RB over IN some DT period NN of IN time NN . SENT A DT software NN agent NN , , once RB invoked VVN , , typically RB runs VVZ until IN it PP decides VVZ not RB to TO . SENT An DT artificial JJ life NN agent NN often RB runs VVZ until IN it's NNS eaten VVN or CC otherwise RB dies VVZ . SENT Of IN course NN , , some DT human NN can MD pull VV the DT plug NN , , but CC not RB always RB . SENT Mobile JJ agents NNS on IN the DT Internet NN may MD be VB beyond IN calling VVG back RB by IN the DT user NN . SENT To TO us PP , , these DT requirements NNS constitute VVP the DT essence NN of IN being VBG an DT agent NN . SENT Let's NNS formalize VV them PP into IN a DT definition NN . SENT An DT autonomous JJ agent NN is VBZ a DT system NN situated VVN within IN and CC a DT part NN of IN an DT environment NN that IN senses NNS that IN environment NN and CC acts VVZ on IN it PP , , over IN time NN , , in IN pursuit NN of IN its PP$ own JJ agenda NN and CC so RB as RB to TO effect VV what WP it PP senses NNS in IN the DT future NN . SENT One CD way NN of IN clarifying VVG the DT boundaries NNS of IN this DT definition NN is VBZ by IN looking VVG at IN extreme JJ cases NNS . SENT Humans NNS and CC some DT animals NNS are VBP at IN the DT high JJ end NN of IN being VBG an DT agent NN , , with IN multiple JJ , , conflicting JJ drives NNS , , multiples NNS senses NNS , , multiple JJ possible JJ actions NNS , , and CC complex JJ sophisticated JJ control NN structures NNS minds NNS Franklin NP 1995 CD . SENT At IN the DT low JJ end NN , , with IN one CD or CC two CD senses NNS , , a DT single JJ action NN , , and CC an DT absurdly RB simple JJ control NN structure NN mind NN . SENT we PP find VVP a DT thermostat NN . SENT A DT thermostat NN . SENT Yes UH , , a DT thermostat NN satisfies VVZ all PDT the DT requirements NNS of IN the DT definition NN , , as RB does VVZ a DT bacterium NN . SENT Strange JJ things NNS sometimes RB happen VVP at IN the DT extremes NNS . SENT Espousing VVG a DT definition NN entails VVZ these DT risks NNS . SENT Our PP$ definition NN yields VVZ a DT large JJ and CC varied JJ class NN of IN agents NNS as IN was VBD to TO be VB expected VVN of IN one CD requiring VVG only RB the DT essence NN . SENT No DT doubt NN it's NNS too RB large JJ to TO be VB useful JJ as RB is VBZ . SENT Adding VVG additional JJ requirements NNS for IN different JJ purposes NNS will MD produce VV useful JJ subclasses NNS of IN agents NNS . SENT We'll NN discuss VV some DT of IN these DT in IN the DT next JJ section NN . SENT But CC first RB , , there EX are VBP a DT couple NN of IN basic JJ points NNS to TO clarify VV . SENT Autonomous JJ agents NNS are VBP situated VVN in IN some DT environment NN . SENT Change VV the DT environment NN and CC we PP may MD no RB longer RB have VH an DT agent NN . SENT A DT robot NN with IN only JJ visual JJ sensors NNS in IN an DT environment NN without IN light NN is VBZ not RB an DT agent NN . SENT Systems NNS are VBP agents NNS or CC not RB with IN respect NN to TO some DT environment NN . SENT The DT AIMA NP agent NN discussed VVD above RB requires VVZ that IN an DT agent NN can MD be VB viewed VVN as IN sensing VVG and CC acting VVG in IN an DT environment NN , , that WDT is VBZ , , there EX must MD exist VV an DT environment NN in IN which WDT it PP is VBZ an DT agent NN . SENT What WP about IN ordinary JJ programs NNS . SENT A DT payroll NN program NN in IN a DT real JJ world NN environment NN could MD be VB said VVN to TO sense VV the DT world NN via IN it's JJ input NN and CC act VV on IN it PP via IN its PP$ output NN , , but CC is VBZ not RB an DT agent NN because IN its PP$ output NN would MD not RB normally RB effect VV what WP it PP senses NNS later RBR . SENT A DT payroll NN program NN also RB fails VVZ the DT over IN time NN test NN of IN temporal JJ continuity NN . SENT It PP runs VVZ once RB and CC then RB goes VVZ into IN a DT coma NN , , waiting VVG to TO be VB called VVN again RB . SENT Most RBS ordinary JJ programs NNS are VBP ruled VVN out RP by IN one CD or CC both DT of IN these DT conditions NNS , , regardless RB of IN how WRB we PP stretch VVP to TO define VV a DT suitable JJ environment NN . SENT All DT software NN agents NNS are VBP programs NNS , , but CC not RB all DT programs NNS are VBP agents NNS . SENT Nor CC are VBP software NN agents NNS defined VVN by IN their PP$ tasks NNS . SENT A DT spell NN checker NN adjunct JJ to TO a DT wordprocessor NN is VBZ typically RB not RB an DT agent NN for IN the DT reasons NNS given VVN in IN the DT preceding JJ paragraph NN . SENT However RB , , a DT spell NN checker NN that WDT watched VVD as IN I PP typed VVN and CC corrected VVN on IN the DT fly NN might MD well RB be VB an DT agent NN . SENT Tasks NNS can MD be VB specified VVN so RB as IN to TO require VV agents NNS to TO fulfill VV them PP . SENT Subroutines NNS of IN agents NNS need MD not RB be VB agents NNS for IN the DT same JJ reasons NNS that IN programs NNS need MD not RB be VB . SENT However RB agents NNS can MD have VH subagents NNS . SENT Herbert NP , , the DT robot NN mentioned VVN above IN , , is VBZ built VVN using VVG a DT subsumption NN architecture NN Brooks NP 1990 CD , , a DT layered VVN architecture NN in IN which WDT each DT layer NN senses NNS and CC acts NNS in IN order NN to TO perform VV its PP$ task NN . SENT Each DT layer NN satisfies VVZ all PDT the DT requirements NNS of IN an DT autonomous JJ agent NN . SENT Thus RB the DT layers NNS constitute VVP a DT multiagent JJ system NN that WDT controls VVZ Herbert NP . SENT Sumpy NP Song NP , , Franklin NP and CC Negatu NP , , 1996 CD is VBZ a DT software NN agent NN living VVG in IN a DT unix NP file NN system NN . SENT Sumpy NP , , also RB built VVD using VVG subsumption NN architecture NN , , consists VVZ of IN subagents NNS that WDT wander VVP , , that IN compress NN files NNS , , that IN backup JJ files NNS , , and CC that WDT put VVD Sumpy NP to TO sleep VV when WRB the DT system NN is VBZ busy JJ . SENT Thus RB , , Sumpy NP is VBZ both CC an DT agent NN and CC a DT multiagent JJ system NN . SENT Our PP$ definition NN of IN an DT autonomous JJ agents NNS has VHZ succeeded VVN in IN distinguishing VVG between IN agents NNS and CC programs NNS . SENT An DT agent NN need MD not RB be VB a DT program NN at IN all DT . SENT it PP may MD be VB a DT robot NN or CC a DT school NN teacher NN . SENT Software NN agents NNS are VBP , , by IN definition NN , , programs NNS , , but CC a DT program NN must MD measure VV up RP to TO several JJ marks NNS to TO be VB an DT agent NN . SENT But CC our PP$ definition NN of IN autonomous JJ agents NNS yield VVP a DT class NN of IN agents NNS so RB large JJ as IN not RB to TO promise VV great JJ utility NN . SENT Let's NNS look VVP at IN subclasses NNS of IN agents NNS with IN more JJR promise NN . SENT Agent NN Classifications NNS The DT various JJ definitions NNS discussed VVD above RB involve VV a DT host NN of IN properties NNS of IN an DT agent NN . SENT Having VHG settled VVN on IN a DT much RB less RBR restrictive JJ definition NN of IN an DT autonomous JJ agent NN , , these DT properties NNS may MD help VV us PP further RBR classify VV agents NNS in IN useful JJ ways NNS . SENT The DT table NN that WDT follows VVZ lists NNS several JJ of IN the DT properties NNS mentioned VVN above IN . SENT PropertyOther NP NamesMeaning NP reactive JJ sensing VVG and CC acting VVG responds NNS in IN a DT timely JJ fashion NN to TO changes NNS in IN the DT environment NN autonomousexercises VVZ control NN over IN its PP$ own JJ actions NNS goal NN orientedpro NN active JJ purposeful JJ does VVZ not RB simply RB act VV in IN response NN to TO the DT environment NN temporally RB continuousis NN a DT continuously RB running VVG process NN communicativesocially RB able JJ communicates VVZ with IN other JJ agents NNS , , perhaps RB including VVG people NNS learningadaptivechanges VVZ its PP$ behavior NN based VVN on IN its PP$ previous JJ experience NN mobileable JJ to TO transport VV itself PP from IN one CD machine NN to TO another DT flexibleactions NNS are VBP not RB scripted VVN characterbelievable JJ personality NN and CC emotional JJ state NN . SENT Agents NNS may MD be VB usefully RB classified VVN according VVG to TO the DT subset NN of IN these DT properties NNS that IN they PP enjoy VVP . SENT Every DT agent NN , , by IN our PP$ definition NN , , satisfies VVZ the DT first JJ four CD properties NNS . SENT Adding VVG other JJ properties NNS produces VVZ potentially RB useful JJ classes NNS of IN agents NNS , , for IN example NN , , mobile JJ , , learning VVG agents NNS . SENT Thus RB a DT hierarchical JJ classification NN based VVN on IN set NN inclusion NN occurs VVZ naturally RB . SENT Mobile NP , , learning VVG agents NNS are VBP then RB a DT subclass NN of IN mobile JJ agents NNS . SENT There EX are VBP , , of IN course NN , , other JJ possible JJ classifying VVG schemes NNS . SENT For IN example NN , , we PP might MD classify VV software NN agents NNS according VVG to TO the DT tasks NNS they PP perform VVP , , for IN example NN , , information NN gathering NN agents NNS or CC email NP filtering VVG agents NNS . SENT Or CC , , we PP might MD classify VV them PP according VVG to TO their PP$ control NN architecture NN . SENT Sumpy NP , , then RB , , would MD be VB a DT fuzzy JJ subsumption NN agent NN , , while IN Etzioni NP and CC Weld's NP Softbot NP would MD be VB a DT planning NN agent NN 1994 CD . SENT Agents NNS may MD also RB be VB classified VVN by IN the DT range NN and CC sensitivity NN of IN their PP$ senses NNS , , or CC by IN the DT range NN and CC effectiveness NN of IN their PP$ actions NNS , , or CC by IN how WRB much JJ internal JJ state NN they PP possess VVP . SENT Brustoloni's NP taxonomy NN of IN software NN agents NNS 1991 CD begins VVZ with IN a DT three CD way NN classification NN into IN regulation NN agents NNS , , planning VVG agents NNS , , or CC adaptive JJ agents NNS . SENT A DT regulation NN agent NN , , probably RB named VVN with IN regulation NN of IN temperature NN by IN a DT thermostat NN or CC similar JJ regulation NN of IN bodily JJ homeostasis NN , , reacts VVZ to TO each DT sensory JJ input NN as IN it PP comes VVZ in IN , , and CC always RB knows VVZ what WP to TO do VV . SENT It PP neither RB plans VVZ nor CC learns VVZ . SENT Planning NN agents NNS plan VVP , , either CC in IN the DT usual JJ AI VVP sense NN problem NN solving VVG agent NN , , or CC using VVG the DT case NN based VVN paradigm NN case NN based VVN agents NNS , , or CC using VVG operations NNS research NN based VVN methods NNS OR CC agents NNS , , or CC using VVG various JJ randomizing VVG algorithms NNS randomizing VVG agent NN . SENT Brustoloni's NP adaptive JJ agents NNS not RB only RB plan NN , , but CC learn VV . SENT Thus RB there EX are VBP adaptive JJ problem NN solving VVG agents NNS , , and CC so RB on IN , , yielding VVG a DT two CD layer NN taxonomy NN . SENT Yet RB another DT possible JJ classification NN scheme NN might MD involve VV the DT environment NN in IN which WDT the DT agent NN finds VVZ itself PP , , for IN example NN software NN agents NNS as RB opposed VVN to TO artificial JJ life NN agents NNS . SENT And CC , , there EX must MD be VB many JJ , , many JJ more JJR such JJ possibilities NNS . SENT Which WDT one NN , , or CC ones NNS , , shall MD we PP choose VV . SENT A DT Natural NP Kinds NNS Taxonomy NN of IN Agents NNS In IN thinking VVG about IN a DT taxonomy NN of IN agents NNS two CD possible JJ models NNS come VVP to TO mind VV , , the DT biological JJ model NN and CC the DT mathematical JJ model NN . SENT The DT biological JJ taxonomy NN takes VVZ the DT form NN of IN a DT tree NN with IN living VVG creatures NNS at IN the DT root NN and CC individual JJ species NNS at IN the DT leaves NNS . SENT For IN example NN , , we PP humans NNS are VBP classified VVN as IN kingdom NN animal NN phylum NN chordata NN class NN mammalia NN order NN primate NN family NN pongidae NNS subfamily NN hominidae NNS genus NN homo NN species NNS sapiens JJ where WRB each DT line NN represents VVZ a DT branching VVG point NN of IN the DT tree NN . SENT Might MD it PP be VB possible JJ to TO create VV such PDT a DT taxonomy NN of IN autonomous JJ agents NNS . SENT Let's NNS start VVP and CC see VVP where WRB we PP get VVP . SENT At IN the DT kingdom NN level NN let's NNS classify VV our PP$ agents NNS as IN either DT biological JJ , , robotic JJ , , or CC computational JJ , , as IN these DT seem VVP to TO be VB natural JJ kinds NNS Keil NP , , 1989 CD . SENT Every DT culture NN and CC even RB very RB young JJ children NNS readily RB distinguish VV between IN animate JJ organisms NNS , , artifacts NNS and CC abstract JJ concepts NNS . SENT At IN the DT phylum NN level NN we PP can MD reasonably RB subclassify VV computational JJ into IN software NN agents NNS and CC artificial JJ life NN agents NNS . SENT At IN the DT class NN level NN we PP might MD subclassify VV software NN agents NNS into IN task NN specific JJ agents NNS like IN Sumpy NP , , entertainment NN agents NNS like IN Julia NP , , and CC computer NN viruses NNS . SENT At IN this DT point NN we've NN succeeded VVN in IN categorizing VVG our PP$ major JJ classes NNS of IN autonomous JJ agents NNS , , that WDT is VBZ the DT known VVN families NNS of IN examples NNS . SENT Further JJR Classification NN Suppose VVP we PP wished VVD to TO classify VV software NN agents NNS further RBR . SENT How WRB might MD we PP go VV about IN it PP . SENT The DT major JJ subclassification NN schemes NNS that WDT come VVP to TO mind VV are VBP via IN control NN structures NNS , , via IN environments NNS database NN , , file NN system NN , , network NN , , Internet NP , , via IN language NN in IN which WDT written JJ or CC via IN applications NNS . SENT Each DT might MD be VB useful JJ . SENT Let's NNS try VVP the DT first JJ . SENT Let's NNS list VVP some DT of IN the DT possible JJ initial JJ classification NN schemes NNS for IN software NN agents NNS via IN their PP$ control NN structures NNS . SENT Brustoloni NP offers VVZ regulation NN , , planning NN and CC adaptive JJ . SENT Another DT strategy NN would MD be VB to TO classify VV by IN type NN of IN control NN mechanism NN , , algorithmic JJ , , rule VV based VVN , , planner NN , , fuzzy JJ , , neural JJ net NN , , machine NN learning NN , , etc FW . SENT Or CC we PP might MD distinguish VV agents NNS with IN a DT central JJ executive NN from IN those DT enjoying VVG distributed VVN control NN . SENT Other JJ binary JJ classifications NNS might MD be VB planning VVG vs NP . SENT non JJ planning NN , , learning VVG vs NP . SENT non JJ learning NN , , mobile JJ vs NP . SENT non JJ mobile JJ , , communicative JJ vs NP . SENT non JJ communicative JJ , , etc FW . SENT Suppose VVP we PP used VVD the DT binary JJ classification NN above IN , , including VVG central JJ vs NP . SENT distributed VVN , , in IN the DT order NN mentioned VVD , , to TO create VV a DT binary JJ classification NN tree NN . SENT The DT first JJ branching VVG would MD be VB according VVG to TO the DT first JJ pair NN . SENT On IN each DT of IN these DT branches NNS we PP then RB branch VV according VVG to TO the DT second JJ pair NN , , and CC on IN each DT of IN these DT four CD we PP branch VV again RB via IN the DT third JJ pair NN , , and CC so RB on IN . SENT We've NP essentially RB listed VVD a DT pool NN of IN features NNS and CC classified JJ according VVG to TO subsets NNS of IN these DT features NNS . SENT Viewing VVG our PP$ taxonomic JJ tree NN from IN this DT perspective NN calls VVZ to TO mind VV a DT mathematical JJ taxonomy NN which WDT also RB employs VVZ collections NNS of IN properties NNS . SENT A DT mathematician NN might MD define VV a DT topological JJ space NN please VV don't NN bother VV yourself PP about IN the DT meanings NNS of IN this DT mathematical JJ term NN or CC others NNS . SENT This DT essential JJ definition NN defines VVZ the DT class NN of IN spaces NNS to TO be VB studied VVN . SENT Then RB the DT notion NN of IN a DT Hausdorff NP space NN might MD be VB defined VVN by IN an DT explicit JJ property NN of IN some DT spaces NNS . SENT Thus RB the DT subclass NN of IN Hausdorff NP spaces NNS is VBZ specified VVN . SENT Next IN the DT notion NN of IN a DT compact JJ space NN may MD be VB defined VVN , , yielding VVG the DT subclass NN of IN compact JJ spaces NNS . SENT The DT intersection NN of IN these DT two CD is VBZ the DT subclass NN of IN compact JJ Hausdorff NP spaces NNS , , about IN which WDT theorems NNS are VBP often RB proved VVN . SENT The DT topological JJ classification NN continues VVZ in IN this DT way NN with IN defining VVG properties NNS giving VVG rise NN to TO subclasses NNS of IN spaces NNS which WDT are VBP then RB studied VVN . SENT This DT type NN of IN classification NN scheme NN is VBZ known VVN as IN a DT matrix NN organization NN among IN psychologists NNS . SENT Each DT feature NN defines VVZ a DT dimension NN . SENT With IN n NN features VVZ an DT n NN dimensional JJ matrix NN is VBZ created VVN , , so RB that IN each DT cell NN of IN the DT matrix NN corresponds VVZ to TO a DT collection NN of IN features NNS , , and CC provides VVZ one CD possible JJ category NN for IN the DT classification NN . SENT Having VHG given VVN the DT essential JJ definition NN of IN an DT autonomous JJ agent NN above IN , , the DT class NN of IN agent NN is VBZ specified VVN . SENT We PP may MD then RB speak VV of IN planning VVG agents NNS , , or CC of IN mobile JJ agents NNS , , or CC even RB of IN mobile JJ , , communicative JJ , , planning VVG agents NNS , , each DT specifying VVG a DT subclass NN of IN agents NNS . SENT Of IN course NN , , we PP must MD have VH given VVN definitions NNS of IN these DT three CD properties NNS . SENT Having VHG the DT basic JJ definition NN of IN an DT autonomous JJ agent NN to TO build VV on RP , , and CC using VVG features NNS for IN further JJR classification NN , , we PP may MD rephrase VV some DT of IN the DT definitions NNS given VVN earlier RBR in IN a DT more RBR convenient JJ manner NN . SENT A DT KidSim NP Agent NP is VBZ dedicated VVN to TO a DT specific JJ purpose NN , , i NP . SENT e SYM . SENT , , is VBZ a DT task NN specific JJ agent NN . SENT A DT Hayes NP Roth NP Agent NP reasons VVZ to TO interpret VV perceptions NNS , , solve VV problems NNS , , draw VV inferences NNS , , and CC determine VV actions NNS , , i NP . SENT e SYM . SENT , , is VBZ a DT reasoning NN agent NN . SENT An DT IBM NP Agent NP carries VVZ out RP some DT set NN of IN operations NNS on IN behalf NN of IN a DT user NN or CC another DT program NN , , i NP . SENT e SYM . SENT , , is VBZ a DT task NN specific JJ agent NN . SENT A DT Wooldridge NP shyp NN . SENT Jennings NP Agent NP interacts VVZ with IN other JJ agents NNS and CC possibly RB humans NNS via IN some DT kind NN of IN agent NN communication NN language NN , , i NP . SENT e SYM . SENT , , is VBZ a DT communicative JJ agent NN . SENT A DT SodaBot NP Agent NP engages VVZ in IN dialog NN , , and CC negotiates VVZ and CC coordinates VVZ transfer NN of IN information NN , , i NP . SENT e SYM . SENT , , is VBZ a DT negotiating NN , , information NN agent NN . SENT Subagents NNS and CC Societies NNS of IN Agents NPS Sumpy NP , , the DT file NN system NN maintenance NN agent NN mentioned VVN above IN , , can MD be VB thought VVN of IN as IN a DT single JJ agent NN , , or CC as IN a DT multiagent JJ system NN consisting VVG of IN Wanderer NN , , Compressor NN , , Back RB Up IN and CC Sleepy JJ . SENT Each DT of IN these DT have VHP independent JJ access NN to TO sensors NNS certain JJ unix NP commands NNS such JJ as IN ls NNS and CC to TO actions NNS other JJ unix NP commands NNS such JJ as IN cd NN , , and CC each DT has VHZ its PP$ own JJ simple JJ agenda NN . SENT Also RB , , each DT runs NNS continuously RB , , and CC acts VVZ so RB as RB to TO effect VV its PP$ next JJ sensing VVG . SENT Thus RB , , each DT may MD be VB considered VVN an DT agent NN in IN it's JJ own JJ right NN , , and CC hence RB a DT subagent NN . SENT Sumpy NP is VBZ thus RB a DT multiagent JJ system NN . SENT Some DT agents NNS with IN a DT layered VVN architecture NN are VBP not RB multiagent JJ systems NNS . SENT M NP ller NP , , Pischel NP , , and CC Thiel NP 1995 CD classify VV such JJ architectures NNS into IN vertically RB and CC horizontally RB layered VVN . SENT In IN horizontally RB layered VVN systems NNS each DT layer NN has VHZ access NN to TO sensing VVG and CC acting VVG , , making VVG a DT decomposition NN into IN subagents NNS likely JJ . SENT In IN vertically RB layered VVN system NN , , only RB the DT lowest JJS layer NN senses NNS , , and CC only RB the DT highest JJS acts NNS , , making VVG a DT multiagent JJ decomposition NN unlikely JJ . SENT As IN a DT multiagent JJ system NN , , Sumpy NP is VBZ particularly RB simple JJ in IN that WDT there EX is VBZ almost RB no DT communication NN between IN the DT subagents NNS . SENT Each DT is VBZ , , of IN course NN , , privy JJ to TO sensing VVG initiated VVN by IN the DT others NNS , , and CC Sleepy's NP action NN effects VVZ the DT others NNS . SENT Also RB , , each DT subagent NN sometimes RB suppresses VVZ the DT actions NNS of IN the DT lower JJR layers NNS . SENT One PP might MD ask VV if IN Wanderer NN is VBZ truly RB autonomous JJ if IN Compressor NN can MD suppress VV its PP$ actions NNS . SENT A DT person NN in IN jail NN , , or CC in IN an DT elevator NN , , has VHZ lost VVN some DT freedom NN of IN movement NN , , but CC is VBZ still RB autonomous JJ . SENT Environment NN may MD be VB expected VVN to IN imposes VVZ limits NNS on IN an DT agent's JJ actions NNS . SENT Going VVG back RB to TO our PP$ topological JJ analogy NN , , we PP might MD call VV a DT system NN with IN no DT communication NN between IN its PP$ subagents NNS a DT discrete JJ multiagent JJ system NN . SENT A DT multiagent JJ system NN in IN which WDT each DT agent NN communicates VVZ with IN every DT other JJ might MD be VB called VVN fully RB connected JJ , , . SENT Thus RB multiagents NNS systems NNS can MD be VB classified VVN according VVG to TO the DT possible JJ communications NN paths NNS through IN the DT system NN . SENT We PP might MD also RB classify VV such JJ systems NNS by IN their PP$ communications NNS bandwidth NN . SENT In IN addition NN to TO multiagent JJ systems NNS that WDT can MD reasonably RB be VB viewed VVN as IN constituting VVG a DT single JJ agent NN , , other JJ multiagent JJ system NN are VBP better RBR classified VVN as IN societies NNS of IN agents NNS . SENT For IN example NN , , when WRB a DT collection NN of IN scheduling NN agents NNS gather VVP to TO schedule VV a DT meeting NN between IN their PP$ users NNS , , they PP pursue VVP a DT common JJ goal NN and CC intelligent JJ group NN behavior NN emerges VVZ see VV Kautz NP , , Selman NP , , and CC Coen NP 1994 CD for IN a DT similar JJ situation NN . SENT Yet RB , , as IN a DT group NN , , our PP$ definition NN of IN agent NN is VBZ not RB met VVN in IN that DT persistence NN is VBZ missing VVG . SENT When WRB scheduling NN is VBZ complete JJ , , our PP$ agents NNS disperse VV , , perhaps RB never RB to TO gather VV again RB in IN this DT same JJ grouping NN . SENT One PP could MD argue VV that IN the DT collection NN of IN all DT such JJ scheduling NN agents NNS at IN a DT given VVN site NN constitute VVP a DT single JJ agent NN . SENT To TO do VV so RB , , the DT notions NNS of IN sensing VVG , , acting VVG , , and CC having VHG its PP$ own JJ agenda NN would MD have VH to TO be VB considerably RB stretched VVN . SENT As IN Russell NP and CC Norvig NP have VHP reminded VVN us PP , , the DT issue NN here RB is VBZ not RB truth NN or CC falsity NN , , but CC what's NNS useful JJ in IN communicating VVG about IN agents NNS . SENT The DT notion NN of IN a DT society NN of IN agents NNS leads VVZ to TO a DT caution NN . SENT The DT term NN agent NN as IN used VVN by IN Minsky NP 1985 CD does VVZ not RB necessarily RB refer VV to TO an DT autonomous JJ agent NN as IN the DT term NN is VBZ used VVN here RB . SENT In IN the DT context NN of IN trying VVG to TO explain VV intelligence NN , , Minsky NP speaks VVZ of IN mental JJ agents NNS , , saying VVG Each DT mental JJ agent NN by IN itself PP can MD only RB do VV some DT simple JJ thing NN that WDT needs VVZ no DT mind NN or CC thought VVD at IN all DT . SENT I PP suspect VVP that IN some DT , , if IN not RB many JJ , , of IN his PP$ agents NNS don't VVD meet VV all PDT our PP$ criteria NNS for IN autonomous JJ agents NNS . SENT Conclusions NNS An DT attempt NN has VHZ been VBN made VVN to TO capture VV the DT essence NN of IN agency NN in IN a DT formal JJ definition NN , , which WDT allows VVZ a DT clear JJ distinction NN between IN a DT software NN agent NN and CC an DT arbitrary JJ program NN . SENT The DT beginnings NNS of IN a DT natural JJ kinds NNS taxonomy NN for IN autonomous JJ agents NNS is VBZ proposed VVN , , as RB is VBZ further JJR classification NN via IN collections NNS of IN features NNS . SENT References NNS Brooks NP , , Rodney NP A NP . SENT 1990 CD , , Elephants NNS Don't NP Play NP Chess NP , , In IN Pattie NP Maes NP , , ed NP . SENT , , Designing NP Autonomous JJ Agents NPS , , Cambridge NP , , MA NN . SENT MIT NP Press NP Brustoloni NP , , Jose NP C NP . SENT 1991 CD , , Autonomous JJ Agents NNS . SENT Characterization NN and CC Requirements NNS , , Carnegie NP Mellon NP Technical NP Report NP CMU NP CS NP 91 CD 204 CD , , Pittsburgh NP . SENT Carnegie NP Mellon NP University NP Dethie NP , , Vincent NP G NP . SENT 1986 CD , , The DT Magic NP of IN Metamorphosis NN . SENT Nature's NP Own NP Sleight NP of IN Hand NP , , Smithsonian NP , , v NN . SENT 17 CD , , p NN . SENT 122 CD ff NNS Etzioni NP , , Oren NP , , and CC Daniel NP Weld NP 1994 CD , , A DT Softbot NN Based VVN Interface NP to TO the DT Internet NN . SENT Communications NNS of IN the DT ACM NP , , 37 CD , , 7 CD , , 72 CD shyp NN . SENT 79 CD . SENT Franklin NP , , Stan NP 1995 CD , , Artificial JJ Minds NNS , , Cambridge NP , , MA NN . SENT MIT NP Press NP Hayes NP Roth NP , , B NP . SENT 1995 CD . SENT An DT Architecture NP for IN Adaptive JJ Intelligent JJ Systems NNS , , Artificial JJ Intelligence NP . SENT Special JJ Issue NN on IN Agents NNS and CC Interactivity NN , , 72 CD , , 329 CD 365 CD , , . SENT Kautz NP , , H NN . SENT , , B NP . SENT Selman NP , , and CC M NP . SENT Coen NP 1994 CD , , Bottom NP up IN Design NP of IN Software NP Agents NPS . SENT Communications NNS of IN the DT ACM NP , , 37 CD , , 7 CD , , 143 CD 146 CD Keil NP , , F NP . SENT C SYM . SENT 1989 CD . SENT Concepts NNS , , Kinds NNS , , and CC Cognitive JJ Development NP . SENT Cambridge NP , , MA NN . SENT MIT NP Press NP . SENT Langton NP , , Christopher NP , , ed NP . SENT 1989 CD , , Artificial JJ Life NP , , Redwood NP City NP , , CA MD . SENT Addison NP Wesley NP Maes NP , , Pattie NP 1990 CD ed NP . SENT , , Designing NP Autonomous JJ Agents NPS , , Cambridge NP , , MA NN . SENT MIT NP Press NP Maes NP , , Pattie NP 1995 CD , , Artificial JJ Life NP Meets VVZ Entertainment NP . SENT Life NN like IN Autonomous JJ Agents NPS , , Communications NPS of IN the DT ACM NP , , 38 CD , , 11 CD , , 108 CD 114 CD Minsky NP , , Marvin NP 1985 CD , , The DT Society NP of IN Mind NN , , New NP York NP . SENT Simon NP and CC Schuster NP M NP ller NP , , J NP . SENT P NN . SENT , , M NP . SENT Pischel NP , , and CC M NP . SENT Thiel NP 1995 CD , , Modeling VVG Reactive JJ Behaviour NN in IN Vertically RB Layered VVN Agent NP Architectures NNS , , in IN Wooldridge NP and CC Jennings NP Eds NPS . SENT , , Intelligent JJ Agents NNS , , Berlin NP . SENT Springer NP Verlag NP , , 261 CD 276 CD Russell NP , , Stuart NP J NP . SENT and CC Peter NP Norvig NP 1995 CD , , Artificial JJ Intelligence NP . SENT A DT Modern NP Approach NN , , Englewood NP Cliffs NPS , , NJ NP . SENT Prentice NP Hall NP Smith NP , , D NP . SENT C SYM . SENT , , A NP . SENT Cypher NP and CC J NP . SENT Spohrer NP 1994 CD , , KidSim NP . SENT Programming NN Agents NNS Without IN a DT Programming NN Language NP , , Communications NPS of IN the DT ACM NP , , 37 CD , , 7 CD , , 55 CD 67 CD Song NN , , Hongjun NP , , Stan NP Franklin NP and CC Aregahegn NP Negatu NP 1996 CD , , A DT Fuzzy JJ Subsumption NN Softbot NN , , Proceedings NNS of IN the DT ISCA NP Int NP Conf NP on IN Intelligent JJ Systems NPS , , Reno NP Nevada NP Wooldridge NP , , Michael NP and CC Nicholas NP R NP . SENT Jennings NP 1995 CD , , Agent NP Theories NNS , , Architectures NNS , , and CC Languages NNS . SENT a DT Survey NN , , in IN Wooldridge NP and CC Jennings NP Eds NPS . SENT , , Intelligent JJ Agents NNS , , Berlin NP . SENT Springer NP Verlag NP , , 1 CD 22 CD