斯坦福大学2024AI指数报告（HAI_AI-Index-Report2024）（英）.pdf

ArtificialIntelligenceIndexReport2024ArtificialIntelligenceIndexReport20242IntroductiontotheAIIndexReport2024WelcometotheseventheditionoftheAIIndexreport.The2024IndexisourmostcomprehensivetodateandarrivesatanimportantmomentwhenAI’sinfluenceonsocietyhasneverbeenmorepronounced.Thisyear,wehavebroadenedourscopetomoreextensivelycoveressentialtrendssuchastechnicaladvancementsinAI,publicperceptionsofthetechnology,andthegeopoliticaldynamicssurroundingitsdevelopment.Featuringmoreoriginaldatathaneverbefore,thiseditionintroducesnewestimatesonAItrainingcosts,detailedanalysesoftheresponsibleAIlandscape,andanentirelynewchapterdedicatedtoAI’simpactonscienceandmedicine.TheAIIndexreporttracks,collates,distills,andvisualizesdatarelatedtoartificialintelligence(AI).Ourmissionistoprovideunbiased,rigorouslyvetted,broadlysourceddatainorderforpolicymakers,researchers,executives,journalists,andthegeneralpublictodevelopamorethoroughandnuancedunderstandingofthecomplexfieldofAI.TheAIIndexisrecognizedgloballyasoneofthemostcredibleandauthoritativesourcesfordataandinsightsonartificialintelligence.Previouseditionshavebeencitedinmajornewspapers,includingtheTheNewYorkTimes,Bloomberg,andTheGuardian,haveamassedhundredsofacademiccitations,andbeenreferencedbyhigh-levelpolicymakersintheUnitedStates,theUnitedKingdom,andtheEuropeanUnion,amongotherplaces.Thisyear’seditionsurpassesallpreviousonesinsize,scale,andscope,reflectingthegrowingsignificancethatAIiscomingtoholdinallofourlives.ArtificialIntelligenceIndexReport20243MessageFromtheCo-directorsAdecadeago,thebestAIsystemsintheworldwereunabletoclassifyobjectsinimagesatahumanlevel.AIstruggledwithlanguagecomprehensionandcouldnotsolvemathproblems.Today,AIsystemsroutinelyexceedhumanperformanceonstandardbenchmarks.Progressacceleratedin2023.Newstate-of-the-artsystemslikeGPT-4,Gemini,andClaude3areimpressivelymultimodal:Theycangeneratefluenttextindozensoflanguages,processaudio,andevenexplainmemes.AsAIhasimproved,ithasincreasinglyforceditswayintoourlives.CompaniesareracingtobuildAI-basedproducts,andAIisincreasinglybeingusedbythegeneralpublic.ButcurrentAItechnologystillhassignificantproblems.Itcannotreliablydealwithfacts,performcomplexreasoning,orexplainitsconclusions.AIfacestwointerrelatedfutures.First,technologycontinuestoimproveandisincreasinglyused,havingmajorconsequencesforproductivityandemployment.Itcanbeputtobothgoodandbaduses.Inthesecondfuture,theadoptionofAIisconstrainedbythelimitationsofthetechnology.Regardlessofwhichfutureunfolds,governmentsareincreasinglyconcerned.Theyaresteppingintoencouragetheupside,suchasfundinguniversityR&Dandincentivizingprivateinvestment.Governmentsarealsoaimingtomanagethepotentialdownsides,suchasimpactsonemployment,privacyconcerns,misinformation,andintellectualpropertyrights.AsAIrapidlyevolves,theAIIndexaimstohelptheAIcommunity,policymakers,businessleaders,journalists,andthegeneralpublicnavigatethiscomplexlandscape.Itprovidesongoing,objectivesnapshotstrackingseveralkeyareas:technicalprogressinAIcapabilities,thec