Artificial intelligence (AI) est wiswa theorias ed techniques enovert kay tyices machines qua ghehde simule intelligence. Idghi correspondt uno menghia om concepts ed technologias meis quem un wardhus swobod ilm. Sems, kaund id pau precis definition os AI, introducent tod topic ka id megil mythos nosters tid.
Ops classifiet in id grupp iom cognitive sciencen, refert ibs nervenets, ad mathematic logique (part om mathematique ed philosophia) ed ad informatique. Paursct methodes kay solve problemes samt akster logic au algorithmic complexitat. Extension-ye kyeust, in id current bahmen, ia ijadhs imitend we taynd el mensc in certain enovers elsen cognitive bungsen.
Ids finalitats ed antplehcen sesuweer aiwois multa tolks, fantasmes au sweurghs wehkwnda tant in Science fiction narns au films quem in philosophic essais.
In informatique, AI, yando kalen "machine intelligence", est intelligent manifesten ab machines, in contraste dia natural intelligence prehpiht ab menscens. Autoritativer AI science buks defineihnt id camp ka id studye om "intelligent agents" : quodkwe apparat quod kwaht sien ambient ed emt actions qua maximisent idso chance os kamyabe nake idsa ziels. Colloquial-ye, id lexis "artificial intelligence" est ops nudt kay describe machines (au computers) qua imite "cognitive" functions qua mensci assocyent con id menscment, kam "eucen" ed "problemen solven".
Wawod machines bihnt meis kabil, taska ayt ka naudhend "intelligence" bihnt ops oistract ex id definition os AI, un phenomen gnoht ka id "AI effect". Un bonmot ios Tesla Theoreme sayct AI est quanto ne hat est bad kwohrt. Mathalan, optic personenrecognition biht frequent-ye exclus ex jects ayt AI, bihvs un routine technologia. Moderne machinen imkans daydey classifien ka AI include kamyabe prete menscwekwos, spehrde ye un hog nivell in strategic leiken systemes (kam schakh ed go), swobod-ye wehghend woghs, routend intelligent-ye in mathmounenlever netwehrgs, ed militar simulations.
Aw automats[]
Oino iom awsten tracen ios theme al "mensc in id machine" tarikhet ex -800 pre JC, in Misr. Id statue os div Amon liv sien brakhs kay kyuses iom novo pharaoh bayna iens prohgs quoy ghieng pri id, dind "swohr" un consecrationslogos. Misris conscier shayad id presence os un sakerdot actionend uno mecanisme ed karugend ia iser werds apter id statue, yed to kwecto ne eet contradictoro pro i binisbat id incarnation os divtat. Circum id sam zaman, Homer, in id Iliad (XVIII, 370-421), descript ia automats tyict ab iom kovach-div Hephaystum : tripods samt golden kwekwls, qua ghehde behre objects tiel id Olymp ed rikes saul do id baytel ios div; auti, dwo slougas kovachen ex gold quas assisteer iom in eys opos. Sam-ye, is aysen giganto Talos, palek iom ripas Cretes, eet yando ayn un wehrg ios div.
Vitruvius, roman architecte, descript id existence inter ia IIIt ed Isto secules pre JC uns ingenieurenscol sult ab Ktesibium in Alexandria, concepend mecanismes bestohmt pro glewos kam kanend crauvs. Heron os Alexandria descript in sien tractat Automata un carrousell moven ab steum ed ayn un anticipation iom steummachines. Automats disprehpent poskwo tiel id fin os Mediev. Buit anadehn ad Roger Bacon id conception om bahnd automats, in fact shayad mecanismes simulend id swehrmen om certain simple werds.
Leonardo da Vinci construxit in 1515 in lionskeip automat kay zalisse iom Roy os France, Francisco iom 1st. Gio Battista Aleotti ed Salomon de Caus men struxeer artificial kanend avs, mecanic pfeifers, nymphes, dragons ed animet satyrs kay udkwitres aristocratic festa, gardens ed grottes. René Descartes de habiet concept in 1649 un automat is kiel ma fille Francine. Duct eti un reflexion os un stauneihndo modernisme de ia differences inter id buhsa iom automats, ed tom hois im animals (neid difference) ed ciois tod im menscen (neid assimilation). Ta analyses iom beuwnt iom misgnoht precursor oinios iom magnen themen os science fiction : id indistinction inter gwiv ed artificial, inter menscens ed robots, androids au artificial intelligences.
In id Ars magna generalis (circum 1274), Ramon Llull inventit un formal lurhat keihnd ep combinator logique kay ghehde tolke de quanto dayir philosophia ed religion aun id barrier os bahsa. Id skip un important part os eys objective de converte i Muslims ad christianisme. Buit id prest peiten os cree un universal lurhat ed influencit Leibniz.
Id formalisation os automatic men[]
Ramon Llull, catalan schwenkjiauschi, philosoph ed theolog ios XIIIt secule, hat ka presto tentet gehne idees med uno mecanic systeme. Combinit random-ye concepts med un genos os hissabschiber ep-yod wohrt concentric diska quer eent grohft buksteivs ed philosophic concepts. Is niem sieno methode Ars Magna, keihnd ep id identification om basic concepts, dind ir mecanic combination, auter inter ia, we con connex idees. Ramon Llull applicit sieno methode ad metaphysique, dind ad moralitat, medicin, ed astrologia. Yed is nud tik deductive logique, quo ne permiss eysi systeme hassile un talimat, ni eti rencause sien inkapsdasturs : tik inductive logique permitti to.
Gottfried Wilhelm Leibniz, ye id XVIIt secule, hat exmohnt uno mehnhissab (calculus rationator), assignend-ye eino numer ielgi concept. Id manipulation tom numern habiet permitten solve ia difficilst questions, eti adduce uni universal bahmen. Leibniz hat lakin demonstren od oino iom magnen difficultaten tos methode, kathalika trohft in moderne orbats de artificial intelligence, est id interconnexion vasyen concepts quo ne permitt isole oin idee ud wiswa kay simplifie ia problemes bohndt ad men.
George Boole hat inventen id mathematic formulation iom suldasturs os aumen, gnohn sub id nam Booles algebra. Conscicit ia diemens om sien orbats con ia mecanismes os intelligence, kam diken ab id titule sienios magnios wehrg paruen in 1852 : The laws of thought.
Gottlob Frege tawrlienit id systeme os Boole formalisend-ye id concepto predicat, quod est un logic entitat quod est auter ver au fals ("cada dom hat un suter") yed quod mathmounti non-logic variables, qua hant per se neid vertatsdegree ("dom", "suter"). Tod formalisation hieb uno megil ahammiyat, chunke id permiss demonstre general theoremes, just appliend-ye typographic reuls ad symbolen menghias. Eti, el neuder act un important rol chunke el gnoht id maynen iom symbols el hat invent ed tod maynen ne esti semper formaliset, quo bringhti tsay do id probleme os maynen in artificial intelligence ed ios subjectivitat im neuders.
Bertrand Russell ed Alfred North Whitehead publier in-kap id XXt secule un wehrgo kalt Principia Mathematica, in-yod resolve contradictions eni id theoria os Gottlob Frege. Ta orbats siseer spehtum adduce uni complete formulation os mathematique.
Allen Newell (San Francisco - 1927 / Pittsburgh - 1992)
Inkapems os artificial intelligence[]
Id xeimen os AI paursken gnahsit in un atelier in Dartmouth College in 1956. Prispehcers Allen Newell (CMU), Herbert Simon (CMU), John McCarthy (MIT) ed Arthur Samuel (IBM) bihr ies dehtors ed liders os AI paursken. Ies ed ir studyents produxeer programmes qua id presse descripsit ka "stauneihnd" : computers uc sadrenc strategias (circum 1954, ed yant 1959 likwert gohder quem el mutawassit mensc), solveer werdenproblemes med algebra, pruv logic theoremes (Logic Theorist, wehrgihn ye id prest ker circum 1956) ed bahr Englisch. Medsu ia 1960tias, paursken in ia Uniet Stats os America buit akster-ye financet ab id Department os Defense ed laboratoria buir establieta tienxia. Ies dehtors os AI eent optimiste de id future; Herbert Simon presieyg : "machines ghehdsient, pre dwogim yars, kwehre quanto wehrg unis menscis ghehdt kwehre." Marvin Minsky samstohm, scribend : "unte oin generation... id probleme os cree artificial intelligence sessiet substance-ye solwt".
Quan access ibs digital computers bihsit possible medsu ia 1950tias, AI paursken inkiep explore id possibilitat od menscenintelligence ghohd bihe reducen do symbolmanipulation. Id paursken eet centret in tri institutions : Carnegie Mellon University, Stanford ed MIT, ed ielg antplohc sien wi style os paursken. John Haugeland niem ta symbolic proschgumts kye AI good old-fashioned AI wa "GOFAI". Unte ia 1960tias, symbolic proschgumts hieb nact megil success kay simule hognivell bren in smulk demonstration programmes. Proschgumts baset ep cybernetique, au artificial neural netwehrgs buir abandont we puscta do id aptergrund. Paurskers in ia 1960tias ed 1970tias eent convict od symbolic proschgumts vikamyabient cree uno machine samt artificial general intelligence ed iey to id ziel irs ilm.
Is economiste Herbert Simon ed Allen Newell studyeer menscen probleme-solutions ilms ed pit formalise ia, ed ir wehrg dehsit ia suls os artificial intelligence, tem quem os cognitive science, operational paursken ed management science. Ir paurskenswoino nud ia resultats om psychologic experiments kay antplehce programmes qua simuleer ia techniques folossien ab leuds kay solve problemes. Tod tradition, centren in id Carnegie Mellon University viculminit in id antplehcen ios Soar architecture medsu ia 1980tias.
Herbert A. Simon (1916-2001) eet un US economist, politique-muallim ed cognitive psychology, quos primar paursken interesse eet decisionemsa eni organisations. Is ielgv id Economia Nobelpris in 1978 ed id Turing Award in 1975. Eys paursken eet skaun ob sien interilm buhsa ed strohc unte ia camps om cognitive science, informatique, public administration, management, ed politic science. Id major part os eys carriere jinkdousit in id Carnegir Mellon University ex 1949 do 2001.
Functionment os artificial intelligence[]
Un typic AI analyset sien ambient ed cheust actions qua maximise idso chance os success. Id iskwto ziel uns AI ghehdt ses simple ("1" sei id AI vinct in id leik os Go, "0" sonst) au complexe (Kwehr actions mathematic-ye samlik tibs qua kamyieb previ). Ziels maghe bihe explicit-ye definiht, we induct. Sei id AI est programmet pro "kardwntalim'', ziels maghe ses implicit-ye induct meizdend-ye sem types os sulouk we puneihnd-ye alya. Alternative-ye, un evolutionar systeme ghehde inductum ziels neudend-ye fitnessfunctions kay mute ed preferential-ye replie hogscore AI systemes, samlika ad kam animals hant evolwt kay ingnaht-ye wehne certain ziels kam trehve pitu. Sem AI systemes, kam id "prokwst nieber", vice aume per analogia, ne hant daydey sem dahn ziels, ploisko od ta ziels sont implicit in ira training data.
Aus paurskers antplohc algorithmes qua imiteer stieup-pos-stieup raisonen quod mensci neude quando solvent enigmas we kwehre logic deductions. Unte ia cert 80tias ed 90tias, AI paursken hieb antplohct methodes kay diles con rhayr-amin au incomplete information, employend concepts ex probabilitat ed economia. Ta algorithmes obkwohk rhayr-kafi pro solve large problemes os raisonen, ob subihr un "combinator explosion" : bihr exponential-ye lenter ye meis ia problemes crohsc. In fact, hatta mensci rar-ye neude id stieup-pos-stieup deduction quod aus AI paursken ghohd modelise. Solvent ir pleist problemes med oku, intuitive kehnsens.
AI ops wehrct med algorithmes. Un algorithme esr uno menghia om non-ambigu bereulens quod uno mecanic computer ghehdti dostringes. Un complexe algorithme est ops construgen atopp alya, simpler, algorithmes. Un simple exempel algorithmes est id sehkwnd recept kay likes tic-tac-toe :
- Sei semquel hat uno "menace" (yani dwo in oin roig), streich id etileikwnd pold, sonst;
- Sei uno mov "sa-ghabelt" kay cree dwo menaces sammel, streich tod pold, sonst;
- Streich id centre pold sei id est leur, sonst;
- Sei tien duschman hat oistricht un wangwlpold, streich id witer pold, sonst;
- Streich un tuich wangwl pold sei est oin, sonst;
- Streich quodlibt tuich pold
Alan Turing (1912-1954) buit un Englisch mathematiquer plaut-ye ayt is pater om theoric informatique ed artificial intelligence. Unte Mund Weir Dwo, Turing ieg un pivotal rol skweitend-ye intercept samdihs qua permiss im Allieits marloubihes i Nazis; kad eys wehrgo bragv id weir unte dwo yars ed sielv upper 14 millions gwiten. Turing buit sudyen in 1952 ob homosexual agsa, qua eent un crime in id Uniet Roydem. Is acceptit chemic quapt, ka alternative ad prison. Is buit trohft mortu in 1954, shayad ob cyanacid suicide. Pos un Internet campagne, Rayn Elizabeth II ijieb postmortem aman ad Turing in 2013 ed, sekwos un petition starten ab Marnanel Thurman, id Bank os Englando mohld od Alan Turings kip esiet ep 50£ banknotes ab 2021.
Id Turing Test[]
In-kap ia 1950tias, inter id gnahsa os cybernetique ed id emergence oika yars serter os artificial intelligence, menxu i gohdst ments tos zamans daume de id possibilitat os construge mehnend machines, Alan Turing propont, yant id inkapem uns article remanus maschour, un test kay determine an uno machine maght bihe definiht ka "conscient".
Definihes intelligence est un challenge ed ne est yakin an ghehdsiemos to sem dien ye un satisfacend weidos. Nu-pussirit tod remarke iom britisch mathematiquer Alan Turing, in 1950, ad propones id "imitation leik" quod pag un precis objective ei tunkye gnahnd computerscience. Tod "imitation leik" suggestit od sem fictice kehnser magha dialehge hois con uno machine ed ciois con un mensc med un terminal aun ghehde enderkrihes i.
Tuntro, neid software hat bad alnos kamyapt tod test, nu-sulouke mae bihe enderkrihn ud un mensc, nespekent multa tentatives.
Nundiens, uno machine ghehdt albatta revide ed evolvihes objectives qua ei buir tribuen. Semo machine ghehdt hatta bihe programmen kay ghehde restructure sien startgnohsa ab daken we kwahn informations. Yed nundiensmachine ne wisayct-ye mehnt, neghi consciet seswo (ni besonters sien limites), id khact ultim-ye decide sien ziels ni exmehne nova fikerformes ios mund.
Un bit (wa "Shannon" in information sciences) est un basic informationsunitat yani binar, "0" au "1". Un recent linguistic studye hat nuper pruven od vasya bahsas transmittent id samo mikdar os information ye id sam syrat, yani 39 bits pro secunde.
Mumkinia kehnsen[]
Compare id beorbatensimkan os information unios mensckers tei uns computer maght hehlpe ad ghabe ia megenaurdhens kay kehnse id practic au ne possibilitat uns akster artificial intelligence, tem quem un simple hissab om kW-stieure permitt grosso modo sayge od onwogh-ono maght spehe wehghtum gadab-ye filan onos we an to ei sessiet impossible. En oik exempels om megenaurdhens in informationsbeorbaten :
- tienbing : oin bit pro secunde;
- typic mainframe iom 1970tias : oino million operationen pro secunde ep 32 bits;
- Intel Parangon XP/S, 4000 processors i860@50mHZ (1992) : 160 milliards operationen pro secunde.
Tod stieure ne est awietu literal-ye. Precidt bilhassa ia megenaurdhens in presence ed ir nisbatan oku evolution.
AI hieb dahto tik mihn resultats ep ia typic 1970tia computers tyeicend 107 logic operations pro secunde. Id mensckers, formet ex 1011 neurons qua ielg khake commute meis quem 100ens keren pro secunde bi-sabab iro relaxationswakt permiss meg meisa logic beorbatens pro waktunitat (1013 logic operations pro secunde). "Tod-on" technic kyehmses neti exist ep ia actual computers, wehrgend ye 64 bits samt horloges cadencet ye takriban 4 gHZ.
Id material est yinjier nundiens present, lakin AI kyunggiawt id difficultato de bfwendwane vasya gnohsas util ei resolution uns complexe probleme. Sema gnohsas sayct implicit sont acquisihn per experience ed dusformalisable. Mathalan, quod enderkweit un familiar lige ud dwocent alya ? To ne bad sagvmos expremes clar-ye.
Intelligent agent paradigme[]
Un intelligent agent est un systeme quod kwaht sien ambient ed emt actions maximisent-ya idsa chances os success. Ia simplest intelligent agents sont programmes qua solve specific problemes. Complicater agents include ensans ed organisations ensanen (kam firmas). Id paradigme permitt im paurskers direct-ye combine we compare different proschgumts kye isolen problemes, suwalend-ye qualg agent est gohdst kay maximise un "ziel-function". Un agent quod solvt un specific probleme maght nudes quodkwe proschgumt quod wehrct - sem agents sont symbolic ed logic, sems sont sub-symbolic artificial neural netwehrgs ed alya maghe nudes nov proschgumts. Id paradigme daht yaschi im paurskers un commun lurhat kay communie con alya camps - kam decisionstheoria ed economia - quaschi neude concepts abstracten agents. Struge un complete agent naudht ke paurskers betolke realiste problemes os integration; mathalan, ob sensor systemes dahnt rhayr-amin information de ambient, planificationsystemes dehlgent ghehde wehrge in presence os rhayr-yakinia. Id intelligent agent paradigme bihsit plaut-ye acceptet unte ia 1990tias.
Intelligent agents dehlgent ghehde sib panges ziels ed nake ia. Naudhent un zariya kay visualise id future - un fiker ios stand ios mund ed ghehde presayge kam ir actions sient change id - ed ghehde kwehre cheusa qua maximise id nude (au "wirt") iom behandet cheusen.
In classic planensproblemes, id agent maght fiker id est id saul systeme quod act in id mund, permittend ei agent ses yakin de ia consequences idsen actions. Lakin, sei id agent ne est id saul action, tun id agent tehrpt raisone in rhayr-yakinia. Zaruri est un agent quod ghehdt ne tik vehrte sien ambient ed presayge, sontern yaschi kehnse sien presaygens ed adapte ia sekwent sien kehnsen.
Multiagent planen naudht id cooperation ed id competition multen agents kay nake un certain ziel. Un talg emergent sulouk est nudt ab evolutionar algorithmes ed swarm intelligence.
Akster artificial intelligence[]
Id concept os akster artificial intelligence refert uni machine quod ne tik-ye ghehdti dughes un intelligent sulouk, sontern spruves un pondos uns real conscience de se, sem "druv khisses" (quakwe khisses), ed "un ghaben siens wi raisonen".
Akster artificial intelligence hat daughen ka motor ei ilm, ed hat kathalika suwt multa debats. Se seulend ef hypothese, quod ia neurosciences trende confirme ed quod paurskers ne keungent de confirme, od conscience hat un biologic edghi material sprijin, i mualims daydey ne vide sem prabhilobstacle face cree sem dien un conscient intelligence ep un sprijin alyo quem biologic. Sekwent i tarafdars os akster artificial intelligence, sei taiper ne sont computers ni robots tem intelligent quem el ensan, to ne est un probleme os alat sontern os conception. Esiet neid functional limite (un computer est un wiswank Turing machine samt ka saula limites ia limites os calculabilitat), esient tik limites bohnden ei menscaptitude ad concepe ia ciautaung softwares (programme, datenbank...). Permitt besonters modelise sem abstract idees.
SMARTHUS Det enkle er ofte det beste REMA 1000-0
Deubeucen[]
Deubeucen (Deep learning) wa "profundo talimat" est quodgvonc neural netwehrgo quod ghehdt uces un long zangir om causal diemens. Mathalan, un predrehb netwehrg samt six kohlt schikhts ghehdt uces septdiemen causal zangir (six kohlt schikht + expiut schikht) ed hat un "credit assignement paund" (CAP) dube om sept. Maung deubeucen systemes tehrbe ghehde uces zangirs om dec au meis causal diemens in longe. Deubeucen hat transformen maung important subcamps os artificial intelligence, pati computer vision, voc recognition, natural bahsa beorbaten ed alya.
Idwert wekwos "Deubeucen" buit introducto do id machine talimat communitat ab Rina Dechter in 1986 ed gaiss elan pos Igor Aizenberg ed colleghs introduxeer id do artificial neural netwehrgs in 2000. Ia prest functional deubeucen netwehrgs buir publiet ab Alexey Grigorevich Ivakhnenko ed V.G.Lapa in 1965. Ta netwehrgs bihnti trainiret ye ein schikht seni. Ivakhnenkos 1971 article descript id talimat uns deubios predrebh multschikht perception samt oct schikhts, ja maung deuber quem baygha serter netwehrgs.
Economic impact os AI ep economia[]
Ia longmeiwrnt effects os AI sont rhayr-yakin. Un ijmalstudye ab economistens dik ikhtilaf de kweter id crehscend neud om robots ed AI siet cause un substantial augen os longmeiwrnt beghsorbatia, yed i daydey samstehme od maghiet ses uno nett benefice, sei ia productivitatsgains bihnt renohmen.
Id sibia inter automation ed employ est complicat. Menxu automation eliminet veut jobs, id yaschi creet nov jobs per microeconomic ed macroeconomic effects. Nekam prever volns os automation, maung jobs medyios classe maghsient bihe eliminet ab artificial intelligence; The Economist script od "id sweurgh an AI ghehdiet inflige ibs albhcolnier jobs quo steummagho kwohr ibs bloucolnier jobs unte id Industrial Revolution" est "emtu serieus-ye". Subjective vehrtens ios risk varyent plaut-ye; mathalan Michael Osborne ed Carl Benedikt Frey vehrte od 47% iom US jobs sont sub "hog risk" os potential automation, menxu un OECA report classifiet tik 9% om US jobs sub "hog risk". Jobs sub extreme risko strehce ex paralegals do okupitu cokwens, menxu job demande shayad maghti crehsce pro kaurrelat professions ex personal ilaja do cler. Martin Ford ed alyi leitent dalger ed arguent od un plaut numer joben sont routinic, repetitive ed (pro un AI) presayget; Ford warnt od ta jobs maghe bihe automaten unte id niebst pair om decennia, ed od maungs iom nov jobs maghe ne ses "accessible ad leuds samt mutawassit imkans", hatta med retraining. Economistes pfehrste od in id prev id technologia hat trenden crisces petis quem reduce total orbatia, bet recogneihnt od smos in "un aunmap territorium" con AI.
Joseph Weizenbaum scripsit od AI applications khake, per definition, kamyabe simule genuino menscempathia ed od id neud os AI technologia in camps kam clientenservice au psychotherapia eet profund-ye misniht. Weizenbaum yaschi swurgh od AI paurskers (ed sem philosophs) ieyskweer id menscment ka neid meis quem un computer programme (un position nunkye gnohn ka computationalisme). Sekwent Weizenbaum suggestent ta points od AI paursken niweirt menscgwit.