医疗保健领域的AI聊天机器人：10个关键案例回顾

ByAdrienLaurent,CEOatIntuitionLabs•10/4/2025•30minread ExecutiveSummary Artificialintelligence(AI)–poweredchatbotsareemergingastransformativetoolsinhealthcare,offeringautomatedpatientsupport,symptomtriage,andmentalhealthcounseling.Sincethemid-20thcenturywithpioneerslikeELIZA—thefirstchatbotsimulatingapsychotherapist([1]pmc.ncbi.nlm.nih.gov)—healthcarechatbotshaveevolvedfromsimplerule-basedscriptstosophisticatedAIsystems.Thecurrentgenerationofchatbotsleveragesadvancednaturallanguageprocessingandmachinelearning,includinglargelanguagemodels(LLMs)suchasOpenAIʼsChatGPT.Contemporaryhealthcarechatbotsspandiverseroles:virtualnurses(e.g.,Senselyʼs“Molly”)forchroniccaremanagement,symptom-checkers/triagebots(e.g.,AdaHealth,Babylon,Buoy)toguidecare-seeking,anddigitaltherapists(e.g.,Woebot,Wysa,Youper)providingcognitive-behavioralsupport.Empiricalstudiessuggestmixedbutgenerallypositiveoutcomes.Forinstance,anRCTfoundthattwoweeksofusingtheWoebotmental-healthchatbotsignificantlyreducedanxietyanddepressionsymptomscomparedtostandardself-helpmaterials([2]formative.jmir.org),andobservationaldatashowhighuserengagementwithAImentalhealthapps.Symptom-checkeraccuracyvaries:inonebenchmarkofCOVID-19screeningbots,Symptoma(F1≈0.92)significantlyoutperformedothertools([3]www.medrxiv.org).Large-scaleevaluationsindicatethatleadingAItriagesystems(e.g.Babylon,Ada)approachhumanclinicianperformanceinsafetyforurgentcareadvice([4]pmc.ncbi.nlm.nih.gov).Adoptionisrising:chatbotscanoperate24/7atscale,andduringtheCOVID-19pandemic,agenciesliketheCDCandWHOdeployedAIchatbotstoguidepatientdecisions(e.g.,US CDCʼsCOVID-19self-checker([5]www.businessinsider.com);WHOʼschatbotinformingbillionsviaFacebook/WhatsApp(www.who.int)).Surveysshowbroadpatientwillingnesstousesuchtools:amajorityagreed“ahealthchatbotisagoodidea”andmanyreportedcomfortsharingsymptomswithachatbot([6]pmc.ncbi.nlm.nih.gov)([7]pmc.ncbi.nlm.nih.gov).However,trustandprivacyconcernsremainsignificant,andnotallbotsareequallyreliable.Thisreportreviewsthetop10AIchatbotsinhealthcare,evaluatingtheirdesign,usage,andevidence.Itprovideshistoricalcontext,examinescurrentcapabilities,presentscasestudiesandresearchfindings,anddiscusseschallengesandfuturedirectionsforAI-drivenconversationalagentsinmedicine. IntroductionandBackground HealthcareisincreasinglyleveragingAI-drivenconversationalagents(“chatbots”)toaddresschallengesofaccess,patientengagement,andproviderworkload.Achatbotisasoftwareagentthatinteractsthroughtextorvoiceinnaturallanguage.AIchatbotsaugmenthumancarebyprovidinginstantresponses,triagingpatients,answeringFAQs,ordeliveringinterventions.Advancesinnaturallanguageprocessing(NLP)andmachinelearningenablecontemporarybotstounderstanduserinputsandgeneratecontext-awarereplies.TheycanaccesslargemedicalknowledgebasesorbebuiltonneuralLLMs(e.g.GPT-4)trainedonbiomedicaldata.Unlikeearliersimpledatabasesordecisiontrees,modernAIchatbotscaninterpretcomplexsymptomdescriptionsandlearnfromdata. Thehealthcarecontexthasstrongdemanddriversforchatbots.Surveysindicatethatalargefractionofpeopleseekmedicalinformationonlinebeforeseeingaclinician(e.g.roughlytwo-thirdsofpatients“google”theirsymptoms)([8]pmc.ncbi.nlm.nih.gov).Overburdenedhealthcaresystemsandprovidershortagescreatedelays;chatbotscanofferinitialreliefbyansweringquestionsorguidingcare-seeking,potentiallyreducingunnecessaryvisits.InpublichealthemergenciessuchasCOVID-19,chatbotsprovidedscalablesymptomscreeningandup-to-dateguidance:forexample,theU.S.CentersforDiseaseControl(CDC)partneredwithMicrosofttodeployaCOVID-19self-checkeronwebandmobileplatforms([5]www.businessinsider.com),andtheWorldHealthOrganizationlaunchedaFacebookMessengerchatbotcapableofreachingbillionsforCOVID-19 advice(www.who.int).ThesetoolsillustratehowAIbotscandisseminatetimelyhealthinformationandtriagelargevolumesofpatients. AIchatbotsareusedacrossmedicaldomains.Theyserveasvirtualsymptomcheckers/triageassistantstohelppatientsdecidetheurgencyofcare.Theycanactasdigitalnurses/assistantstomanagechronicdiseases(reminders,monitoring),asexecutorofFAQsandadministrativetasks,andasmentalhealthcoaches.Inmentalhealth,appslikeWoebotandWysaprovidecognitive-behavioraltherapy(CBT)–basedsupportviaconversation.Otherbotsprovidemedicationguidanceortreatmentadherencereminders(e.g.‘Florenceʼchatbotformedicationschedules).Somerecentchatbots(e.g.GoogleʼsMed-PaLM2underdevelopment)aimtoanswergeneralclinicalquestions. Historicalcontext:Theideaofcomputerconversationalagentsdatesbackdecades.TheearliestnotableexampleisELIZA(1966),apattern-matchingbotthatsimulatedaRogerianpsychotherapistbyrephrasinguserinputs([1]pmc.ncbi.nlm.nih.gov).ThoughELIZAwassimplistic,itdemonstratedthecompellingillusionofunderstanding(“ELIZAeffect”).Soonafter,otherprototypeslikePARRY(asimulationofparanoidthinking)werebuilt.However,theseearlybotslackedtrueAIinferenceandwererule-based.Overtime,chatbotsinhealthcareremainedlargelyresearchcuriositiesuntilrecently.Thepastdecadehasseenanexplosionofinterest,propelledbyubiquitoussmartphones,dataconnectiv