Share
Embed
Transcript
42 slides
-
ਓ "*ͷҙࢥܾఆࣗಈతʹྑ͘ͳΒͳ͍ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ ˔ จɾ࣮ݧΛରʹͨ͠ϝλੳɿҙࢥܾఆਫ਼ɺ ˙ ਓ "*ɺਓؒ୯ಠΑΓྑ͍͜ͱ͕ଟ͍ ˙ ਓ
"*͕ɺਓؒ୯ಠɾ"*୯ಠͷྑ͍ํΛ͑ΔͱݶΒͳ͍ 5 Vaccaro et al. When combinations of humans and AI are useful. Nature Human Behaviour 2024. "*͕ਓؒΑΓߴਫ਼ͳ໘Ͱɺ "*ͷஅΛ࠾༻Ͱ͖͍ͯͳ͍ -
ਓ "*Ͱਫ਼্͕͕ͬͯదʹཔΕ͍ͯΔͱݶΒͳ͍ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ ˔ ࠶൜༧ଌͱϩʔϯ৹ࠪΛࡐʹɺ"*அͷఏࣔํ๏Λม࣮͑ͯݧ ˙ ਓ "*ਓؒ୯ಠΑΓߴਫ਼ʢ"*୯ಠΑΓਫ਼ʣ ˙ ͕ͩʮࣗ"*͕ͲΕ͘Β͍͍ͨͬͯΔ͔ʯ
ʮ͍ͭ"*ʹཔΔ͖͔ʯΛࢀՃऀѲͰ͖͍ͯͳ͔ͬͨ 6 Green and Chen. The principles and limits of algorithm-in-the-loop decision making. CSCW 2019. ਓؒ୯ಠ "*୯ಠ "* ਓؒ "* ਓؒʢ"*༧ଌΛॳظೖྗʹઃఆʣ "* ਓؒ ʢਓؒͷ༧ଌͷޙʹ"*ఏࣔʣ "* ਓؒʢઆ໌͖ʣ "* ਓؒʢճޙʹਖ਼ղఏࣔʣ ༗ҙʹਖ਼ʢෛʣͷؔɺ༗ҙͳؔͳ͠ ࣗݾஅͷ ֬৴ͱ ࣮ࡍͷਖ਼ͷؔ "*ੑೳͷ ओ؍తධՁͱ "*ਖ਼ͷؔ "*͕ਖ਼͍࣌͠ʹ "*அΛ࠾༻Ͱ͖Δ͔ -
จ຺ใʹجͮ͘அ"*ͷա৴ΛऑΊ͏Δ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ 7 De-Arteaga et al. A case for humans-in-the-loop:
Decisions in the presence of erroneous algorithmic scores. CHI 2020. "*͕༧ଌͨ͠ϦεΫείΞΛ ୲ऀʹఏࣔ ௐࠪෆཁPS ཁௐࠪΛஅ "*είΞ͕Ҏ্ͷ߹ ʮௐࠪෆཁʯஅʹ ্࢘ͷڐՄ͕ඞཁ ˔ ࣇಐٮɾωάϨΫτ௨ใͷεΫϦʔχϯά"*ͷޡ࡞ಈΛར༻͠ɺ "*ʹର͢ΔઐՈʢࣇಐࢱ୲ऀʣͷৼΔ͍Λੳ ˔ ߴϦεΫͱޡදࣔɿ"*ʹैΘͣʮௐࠪෆཁʯͱஅ͢Δɺ ϦεΫͱޡදࣔɿඞཁͳέʔεͰʮཁௐࠪʯͱਖ਼͘͠அ ˙ ઐՈ"*அҎ֎ͷจ຺ใ ʢ௨ใ༰ߦهʣΛ ༻͍ͯஅ͍ͯͨ͠ -
ࣦഊ݅Λֶͼ͍͢"*ͷํ͕ਓ "*ʹద͍ͯ͠Δ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ ˔ ਓؒʹ"*ͷࣦഊ݅ΛֶͤΔήʔϜΛઃܭ ˔ ෆྑఆ͕ࡐɻͷಛ͕༩͑ΒΕɺ ࢀՃऀҎԼͷ͍ͣΕ͔Λબ ˙ "*ʹͤΔɿਖ਼ͳΒใुɾޡͳΒଛࣦ
˙ ࣗͰஅɿใुଛࣦͳ͠ ˔ ϥϯυ͕ਐΉͱਓؒ"*ͷࣦഊ݅Λֶͼɺ "*͕ਖ਼ͷ࣌ʹͤɺޡͷ࣌ʹୀ͚ΔΑ͏ʹͳͬͨ ˔ ಉਫ਼ͳΒɺ୯७ɾ࣍ݩɾҰ؏ͨ͠அͷ"*΄Ͳ ਓ "*ͷྦྷੵใु্͕ 8 Bansal et al. Beyond accuracy: The role of mental models in human-AI team performance. HCOMP 2019. -
આ໌͕͋ΕదͳཔΓํΛֶΔͷ͔ʁ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ ˔ ༿ը૾ྨλεΫͰɺਓ͕"*அΛ࠾༻͢Δ͔Λෳͷ"*આ໌Ͱൺֱ ˔ આ໌͕͋Δͱɺ"*͕ਖ਼ͷ࣌ʹ࠾༻͠ɺޡͷ࣌ʹڋ൱͢Δ૿ ˙ ը૾આ໌͕ޮՌେɻղऍ͍ۙ͢͠ࣄྫ͕ɺݸผࣄྫͰ"*࠾༻ ͷஅΛॿ͚ͨՄೳੑ͕͋Δ 9
Yang et al. How does visual explanation foster appropriate trust in machine learning? IUI 2020. ۙࣄྫΛը૾Ͱ දݱ͢Δઆ໌ ۙࣄྫΛಛϕΫτϧͰ දݱ͢Δઆ໌ "*அ ܇࿅σʔλதͷۙࣄྫ -
0WFSSFMJBODFͱVOEFSSFMJBODF ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ 10 "*ͷஅ ਓͷ࠷ऴஅ ˓ ˓ ˓ ʷ ʷ
˓ ʷ ʷ ˓ਖ਼ɺʷޡ 6OEFSSFMJBODF "*͕ਖ਼ͷ࣌ɺ"*அΛڋ൱͢Δׂ߹ 0WFSSFMJBODF "*͕ޡͷ࣌ɺ"*அΛ࠾༻͢Δׂ߹ -
"QQSPQSJBUFSFMJBODF 3"*3 343 ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ 11 Schemmer et al. Appropriate reliance
on AI advice: Conceptualization and the effect of explanations. IUI 2023. ਓͷॳظஅ "*ͷஅ ਓͷ࠷ऴஅ ʷ ˓ ˓ ʷ ˓ ʷ ʷ ʷ ˓ ʷ ʷ ʷ ਓͷॳظஅ "*ͷஅ ਓͷ࠷ऴஅ ˓ ˓ ˓ ˓ ˓ ʷ ˓ ʷ ˓ ˓ ʷ ʷ 3"*3 3FMBUJWF"*SFMJBODF ਓؒͷॳظஅ͕ޡͰɺ"*͕ਖ਼ͷͱ͖ɺ"*அΛ࠾༻͢Δׂ߹ -
"QQSPQSJBUFSFMJBODF 3"*3 343 ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ 12 Schemmer et al. Appropriate reliance
on AI advice: Conceptualization and the effect of explanations. IUI 2023. ਓͷॳظஅ "*ͷஅ ਓͷ࠷ऴஅ ʷ ˓ ˓ ʷ ˓ ʷ ʷ ʷ ˓ ʷ ʷ ʷ ਓͷॳظஅ "*ͷஅ ਓͷ࠷ऴஅ ˓ ˓ ˓ ˓ ˓ ʷ ˓ ʷ ˓ ˓ ʷ ʷ 343 3FMBUJWFTFMGSFMJBODF ਓؒͷॳظஅ͕ਖ਼Ͱɺ"*͕ޡͷͱ͖ɺ"*அΛڋ൱͢Δׂ߹ -
આ໌ɺਖ਼͍͠"*ʹཔΔߦಈ͚ͩΛଅ͢ ͍ͭ"*ʹཔΓɺ͍ͭୀ͚Δ͔ ˔ ِϗςϧϨϏϡʔͷݕग़λεΫͰઆ໌ͷӨڹΛݕূ ˙ -*.&ܕͷ୯ޠϋΠϥΠτઆ໌ ˔ આ໌3"*3Λ༗ҙʹߴΊΔ͕343มԽ͠ͳ͍ ˙ ʮઆ໌͕ɺޡͬͨ"*Λڋ൱͢ΔͷΛॿ͚Δʯͱݴ͑ͳ͍
˔ "*ͷओ؍త৴པ͕ߴ͍ͱɺ3"*3͕ߴ·Δ͕343Լ͕Δ ˙ ओ؍త৴པਖ਼͍͠"*ͷ ࠾༻Λॿ͚Δ͕ ޡͬͨ"*ͷա৴ ট͖͏Δ 13 Schemmer et al. Appropriate reliance on AI advice: Conceptualization and the effect of explanations. IUI 2023. -
ϞσϧͷղऍੑɺదͳґଘΛଅ͢ͱݶΒͳ͍ આ໌ͱ֬৴ɺཔΓํΛͲ͏ม͑Δ͔ ˔ ΞύʔτՁ֨༧ଌͰͷ࣮ݧɻઆ໌ͱͯ͠ઢܗճؼϞσϧͷ ΛݟͤΔ߹ $MFBS ͱݟͤͳ͍߹ #MBDLCPY Λൺֱ ˔
ม$MFBS݅ ˙ Ϟσϧ༧ଌΛ༧͢ΔλεΫߴਫ਼Ͱղ͚ͨ ˙ ਓ "*ͷਫ਼վળ͠ͳ͔ͬͨɻ"*͕ଥͳ࣌ʹ࠾༻͢Δߦಈʹ ܨ͕Βͳ͔ͬͨ ˙ ֎ΕͰ ޡͬͨϞσϧ༧ଌΛ ୀ͚ʹ͔ͬͨ͘ 15 Poursabzi-Sangdeh et al. Manipulating and measuring model interpretability. CHI 2021. $MFBS݅ɿ͕ݟ͑Δ #MBDLCPY݅ɿ͕ݟ͑ͳ͍ -
આ໌ݕূࡐྉ͔ʁઆಘ͔ʁ આ໌ͱ֬৴ɺཔΓํΛͲ͏ม͑Δ͔ ˔ "*ͱਓ͕ؒಉͷਖ਼ͷλεΫʢײྨɺཧਪʣͰ࣮ݧ ˔ "*ͷ֬৴ͱઆ໌Λఏࣔ͢Δͱ ਓ "*ͷਖ਼ਓؒ୯ಠɾ"*୯ಠΛ্ճͬͨ ˙ "*ਖ਼࣌ਓ
"*ͷਖ਼্͕͕͕ͬͨ"*ޡ࣌Լ͕ͬͨ ˙ આ໌͕ɺݕূͰͳ͘తैΛଅͯ͠͠·ͬͨڪΕ 16 Bansal et al. Does the whole exceed its parts? The effect of AI explanations on complementary team performance. CHI 2021. ༧ଌࠜڌΛϋΠϥΠτ͢Δઆ໌ -
આ໌͕ޮ͘ͷɺਓ͕ؒ"*ͷޡΓΛ࿑ྗͰݕূͰ͖Δ࣌ આ໌ͱ֬৴ɺཔΓํΛͲ͏ม͑Δ͔ ˔ ໎࿏ͷग़ޱΛͰ͑ΔλεΫɻܦ࿏આ໌Λछྨ༻ҙ ˔ қ͕ߴ͍໎࿏Ͱ )JHIUMJHIUઆ໌ɿઆ໌ແ͠ΑΓPWFSSFMJBODF͕վળ 8SJUUFOઆ໌ɿɹઆ໌ແ͠ʹରͯ͠༗ҙͳվળͳ͠ ˙ ݕূ͍͢͠આ໌PWFSSFMJBODFͷվળʹཱͭ
˔ ਖ਼࣌ͷϘʔφεֹΛ্͛ΔͱɺPWFSSFMJBODF͕͞Βʹվળ 17 Vasconcelos et al. Explanations can reduce overreliance on AI systems during decision-making. CSCW 2023. )JHIMJHIUઆ໌ 8SJUUFOઆ໌ "*࣌ʑนΛ͢Γൈ͚Δɻ IJHIMJHIUઆ໌ͩͱؾ͖͍ͮ͢ -
ࣗݾա৴"*ΛΘͳ͍ޡΓΛੜΉ આ໌ͱ֬৴ɺཔΓํΛͲ͏ม͑Δ͔ ˔ ཧਪλεΫΛࡐʹࣗݾೳྗͷࣗ৴͕ "*அͷ࠾༻ʹ༩͑ΔӨڹΛௐࠪ ˔ ࣗݾೳྗΛաେධՁͨ͠ਓɺ"*ͷਖ਼࣌Ͱ "*Λ࠾༻ͤͣɺ࠷ऴஅͷਖ਼͕͔ͬͨ ˔ "*ޡΛݟͤΔνϡʔτϦΞϧΛಋೖ
˙ ࣗݾೳྗ͕աେධՁͷਓɿBQQSPQSJBUF SFMJBODF͕վળ͢Δ ˙ ࣗݾೳྗ͕աখධՁͷਓɿBQQSPQSJBUF SFMJBODF͕༗ҙʹѱԽɻ"*ͷʢޡͬͨʣ ༏ӽײͱ"*ճආ͕ੜͨ͡ 18 He et al. Knowing about knowing: An illusion of human competence can hinder appropriate reliance on AI systems. CHI 2023. -
ࣗͱ"*ͷͲͪΒΛٙ͏͖͔ આ໌ͱ֬৴ɺཔΓํΛͲ͏ม͑Δ͔ ˔ "*ͷ֬৴ʹՃ͑ͯࣗͷਖ਼ՄೳੑΛ༻͍པΓํͷௐΛࢧԉ ˙ ࢀՃऀ͝ͱͷॳظஅσʔλΛ༻͍ͯਖ਼ՄೳੑΛਪఆ ˔ "*ͷ֬৴ߴ͍͕"*ޡͷέʔεͰ ਓ "*ͷਖ਼͕༗ҙʹ্
˔ "*֬৴ࣗͷਖ਼Մೳੑͷ࣌ʹ "*அΛӅ͢ํ๏ 0WFSSFMJBODFվળ͕͋ͬͨ ˙ "*ͷ֬৴͍͕"*ਖ਼ͷέʔεͰ ਓ "*ͷਖ਼͕༗ҙʹѱԽ 19 Ma et al. Who should I trust: AI or myself? Leveraging human and AI correctness likelihood to promote appropriate trust in AI-assisted decision-making. CHI 2023. ࣗͷ ਖ਼Մೳੑ "*ͷ ֬৴ -
"*Λ͙͢ݟͤͳ͍ຎࡲɺա৴ΛऑΊ͏Δ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ৯ࣄը૾͔ΒਫԽݮʹޮ͘৯ࡐΛબͿλεΫ͕ࡐ ˔ "*அઆ໌Λ࠷ॳ͔ΒఏࣔͤͣʮຎࡲʯΛ༩͑Δํ๏Λ࣮ɿ ϘλϯΛԡͨ͠Βఏࣔɺॳظஅͷޙʹఏࣔɺඵޙʹఏࣔ ˔ ຎࡲ͋Γ"*ͷPWFSSFMJBODF Λ͕͑༗ҙʹվળ
˔ Ұํɺຎࡲ͋Γɺ ओ؍త৴པɾ·͠͞ΛߴΊΔ ઃܭͱݶΒͣɺೝతෛ୲Λ ͏ 21 Bucinca et al. To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI-assisted decision- making. CSCW 2021. -
࠶ߟհೖɺݱͷޮੑͱিಥ͢Δ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ පບͷεΫϦʔχϯά"*͕ҟৗͳ͠ͱ༧ଌͨ͠ ߹ɺ؟Պҩͷ֬ೝ͕࣌ؒ͘ͳΔόΠΞε͕໌ ˔ όΠΞεରࡦʹਓͷॳظஅΛಋೖ ˙ ॳظஅˠ"*ˠ࠷ऴஅɿ "*ʹΑΔޮԽΛ્͢Δͱෆධ
˙ ॳظஅ͕ҟৗͳ͠ˠ"*͕ൃ͢Δ߹ʹਓʹ࠶֬ೝ Λґཔɿධ͕ͩʮ"*ʹಜ͞Ε͍ͯΔʯͱෆշʹײ ͡Δਓ͍ͨ ˔ ߴ͍࡞ۀྔ͕ٻΊΒΕΔຊࡐͰόΠΞεͷةݥੑ ೝࣝ͞Ε͍͕ͯͨɺͦΕΑΓޮੑ͕ॏࢹ͞Ε͍ͯͨ 22 Bach et al. “If I had all the time in the world ” : Ophthalmologists' perceptions of anchoring bias mitigation in clinical AI support. CHI 2023. ɿ ҟৗͳ͠ ԫɿ গྔͷ පมݕग़ ɿ ΑΓଟ͘ͷ පมݕग़ -
આ໌Λ͋͑ͯΔͱɺ"*ͷैԼ͕Δ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ਓʹߟ͑ͤ͞ΔͨΊʹʮ"*આ໌Λ෦తʹݟͤΔʯख๏ΛධՁ ˙ ࡐɿ࠷ܦ࿏ίετͷճɺจষதͷޡΓͷճ ˔ ෦తͳઆ໌શͳઆ໌ͱൺֱͯ͠ɺ"*ਖ਼࣌ޡ࣌ "*ͷ࠾༻Λ༗ҙʹԼ͛ͨ ˙
PWFSSFMJBODF͑ΒΕ͕ͨVOEFSSFMJBODF͕ੜͨ͡ 23 DeJong et al. Cognitive forcing for better decision-making: Reducing overreliance on AI systems through partial explanations. CSCW 2025. ෦తͳઆ໌ શͳઆ໌ ෦తͳઆ໌ શͳઆ໌ -
આ໌Λ͍ʹม͑ΔͱɺओுΛݕূ͘͢͠ͳΔ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ʮ"*આ໌จΛ͍ʹม͑Δʯํ๏ʢ࣭ܕʣΛఏҊ ˙ ࡐɿ૪తτϐοΫͷจষΛಡΈɺཧతଥੑΛஅ ˔ ࣭ܕʹΑΓɺਖ਼͕༗ҙʹ্ͨ͠ ˙ આ໌ܕʮ͜ͷใͰेʯͱײͤ͘͢͡͞ɺ
࣭ܕ΄Ͳ࠶ݕ౼Λଅ͞ͳ͔ͬͨՄೳੑ͕͋Δ 24 Danry et al. Don ’ t just tell me, ask me: AI systems that intelligently frame explanations as questions improve human logical discernment accuracy over causal AI explanations. CHI 2023. ࣭ܕ આ໌ܕ -
ࣗͱ"*ͷࠜڌࠩΛৼΓฦΒͤΔ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ॳظஅͷޙʹ"*આ໌Λݟ ͤΔʮߏతͳৼΓฦΓʯ ˙ ࣗͱ"*ͷҧ͍Λ֬ೝ ˠ"*அɾઆ໌Ͱଥͳ ɾෆࣗવͳΛߟ͑Δ ˠॳظஅΛৼΓฦΓɺ
ҡ࣋͢Δ͔"*࠾༻͢Δ͔ ΛܾΊͯཧ༝Λॻ͘ ˔ ਖ਼͕༗ҙʹ্ɺ PWFSSFMJBODF͕͑ΒΕͨ 25 Li et al. Guided reflection in AI-assisted decision-making: Effects on AI overreliance and decision accuracy. CHI 2026. ᶃ"*அͱઆ໌ΛݟΔ ᶄ"*அΛϨϏϡʔ ᶅॳظஅͷৼΓฦΓ ᶆཧ༝Λॻ͍ͯ࠷ऴஅ -
ਓͱ"*ͷෆҰகΛٞ͠ɺ"*ͷஅج४Λௐ͢Δ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ਓͱ"*͕֤ʑͷஅͱஅج४Λݟͯٞ͢Δʢ--.Λ׆༻ʣˠਓͷ ओுͷڧ͞ʹج͖ͮ"*͕அج४Λௐˠਓʹఏࣔ͠ਓ͕࠷ऴஅ ˔ ӃࢼͷεΫϦʔχϯάΛࡐʹ࣮ݧɻ"*આ໌͚ͩΛݟͤΔ߹ͱൺֱ ͯ͠ਖ਼͕༗ҙʹ্ɻ0WFSSFMJBODFΛݮΒ͢͜ͱ͕Ͱ͖ͨ ˙ ҰํͰ"*ʹର͢Δओ؍త৴པײɺຬԼͨ͠
26 Ma et al. Towards human-AI deliberation: Design and evaluation of LLM-empowered deliberative AI for AI-assisted decision-making. CHI 2025. ᶃਓͱ"*͕அͱઆ໌Λఏࣔ ᶄਓͱ"*͕ٞ ᶅ"*͕அج४Λௐ͠அΛߋ৽ ᶆਓ͕࠷ऴஅ -
"*ਓؒͷஅج४Λݟ͢ڭࡐʹͳΔ "*ॿݴΛ࠶ߟͷ͖͔͚ͬʹ͢Δ ˔ ਓؒ୯ಠͰެฏͳஅ͕Ͱ͖ΔΑ͏ʹɺެฏϞσϧͱࣗΛ฿ͨ͠ ʢෆެฏͳʣϞσϧͷઆ໌Λڭࡐͱͯ͠ఏࣔ ˔ ଟ͘ͷࢀՃऀͰஅͷެฏੑ ্͕ɺʮಛఆͷใ ͚ͩͰஅ͍ͯͨ͠ʯͷ ࠶ߟؾ͖ͮΛ༩͑ɺ
அج४ͷௐΛଅͨ͠ 27 Yang et al. Fair machine guidance to enhance fair decision making in biased people. CHI 2024. Example of an appropriate response. You predicted that the person below would have a LOW INCOME. To be fair, you should have predicted a HIGH INCOME. Age: 50, Gender: Male Race: Asian Workclass: Self-employed Education: Professional school #years of education: 15 Marital status: Married Relationship: Husband Occupation: Professional specialty Working time: 50h/week Native country: Philippines Your criteria vs. fair criteria. Age: 50, Gender: Male Race: Asian Workclass: Self-employed Education: Professional school #years of education: 15 Marital status: Married Relationship: Husband Occupation: Professional specialty Working time: 50h/week Native country: Philippines Age: 50, Gender: Male Race: Asian Workclass: Self-employed Education: Professional school #years of education: 15 Marital status: Married Relationship: Husband Occupation: Professional specialty Working time: 50h/week Native country: Philippines Your criteria Fair criteria HIGH INCOME LOW INCOME The left column of the figure shows your decision criteria, as estimated from your answers using AI. You tend to predict a high income when the information is blue (or when the value of blue information is high). You tend to predict low income when the information is red (or when the value of red information is high). The right column of the figure shows fair decision criteria, as estimated by Fair AI. Your decision will be fairer if you follow these criteria. To be fair, you should predict a high income when the information is blue (or when the value of blue information is high). To be fair, you should predict a low income when the information is red (or when the value of red information is high). B "Fair AI," which simulates what your judgment would look like if it were fair. ࢀՃऀΛ฿ͨ͠ ෆެฏϞσϧͷઆ໌ ࢀՃऀ͕ެฏͳ߹Λ฿ͨ͠ ެฏϞσϧͷઆ໌ -
--.ݕࡧ͍͕ɺޡʹؾ͖ͮʹ͍͘ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ ैདྷܕݕࡧPS--.࣭ԠΛ׆༻ͯ͠ 467ͷੵࡌޮΛൺֱ͢ΔλεΫ ˔ --.࣭Ԡޮత͕ͩ--.͕ؒҧ͑Δ λεΫͰਖ਼͕Լ ˙ --.ޡใΛग़͢໘͕͕͋ͬͨɺ
ࢀՃऀͷओ؍త৴པैདྷܕͱಉ ˙ ମݧશମैདྷܕݕࡧΑΓߴධՁ ˔ --.ͷτʔΫϯੜ֬ͷߴͷ৭͚Ͱ λεΫͷਖ਼͕վળ 29 Spatharioti et al. Effects of LLM-based search on decision making: Speed, accuracy, and overreliance. CHI 2025. --.࣭Ԡ ैདྷܕݕࡧ -
--.આ໌ɺݕূΑΓೲಘΛଅ͢ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ ٬؍తࣄ࣮ʹؔ͢ΔͰͷ࣮ݧɻ --.ਖ਼ޡઆ໌༗ແιʔε༗ແͰൺֱ ˔ આ໌͕͋Δͱਖ਼ޡʹΑΒͣSFMJBODFΛߴΊͨ ˔ ιʔε͕͋ΔͱPWFSSFMJBODFΛԼ͛Δ ˔
--.ޡ࣌ͷ࠷ऴਖ਼ ιʔεͷΈઆ໌ ιʔε ˙ આ໌͕ઌʹʮೲಘʯΛଅͨ͠Մೳੑ ˔ આ໌͕͋Δͱਖ਼ޡʹΑΒͣʮࠜڌ͚͕ेʯ ʮஅʹཱͭʯͱ͍͏ҹΛ༩͑ͨ 30 Kim et al. Fostering appropriate reliance on large language models: The role of explanations, sources, and inconsistencies. CHI 2025. આ໌ ιʔε -
--.BHFOUͷߦಈϩάɺਖ਼֬ੑͷอূʹͳΒͳ͍ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ --.BHFOU׆༻ͷఆੑੳɻ ʮߴྸͷͷ݈߁Λࢧԉ͢ΔͨΊʹ ใΛௐΔʯλεΫͰ࣮ݧɻࢀՃऀ ςοΫاۀۈɺෳҬɻ ˔ ࢀՃऀɺιʔεͷछྨɺߦಈϩ άɺճͷ͞ɾจମ͕--.BHFOUʹ
ର͢Δ֬৴ͷख͕͔Γͩͱޠͬͨ ˙ ͜ΕΒ࣮ࡍͷਖ਼֬ੑΛอূͤͣɺ BHFOU͕ޡͰࢀՃऀ͕BHFOUग़ྗ ͷߴ͍֬৴Λࣔ͢໘͕͋ͬͨ 31 Brachman et al. Building appropriate mental models: What users know and want to know about an agentic AI chatbot. IUI 2025. Ϣʔβͷ࣭ʹର͠ɺ ࣗͰݕࡧߦಈΛܾΊͯԠ͢Δ --.BHFOU -
--.ͷจମ༰ͱผʹஅΛಈ͔͢ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ "*ॿݴͷτʔϯ͕ਓͷҙࢥܾఆʹ༩͑ΔӨڹΛௐࠪɻ ࡐөըਪનɺҙݟܗɺ࠶൜༧ଌɻ ˙ --.Λ༻͍ͯॿݴͷτʔϯΛ'PSNBM DBTVBM SPNBOUJDʹม ˔
ϢʔβଐੑผͷޮՌͷҧ͍͕ݟΒΕͨɻ ˙ ߴྸͳਓ΄Ͳ'PSNBMͳ τʔϯʹै͍͢͠ɻ ࠶൜༧ଌͰࣗͷ ࠷ऴஅͷ֬৴ ڧ·ͬͨ 32 Okoso et al. Do expressions change decisions? Exploring the impact of AI's explanation tone on decision-making. CHI 2025. ॿݴͷ༰ಉ͡Ͱ τʔϯ͚ͩม -
--.ཧ༝͚ͮΛ͛Δ૬खʹͳΔ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ ࢿϙʔτϑΥϦΦ࡞ࢧԉ"*Ͱ࣮ݧɻࢀՃऀࢿʹेৄ͍͠ɻ ˙ 3FDPNNFOE"*ࢿΛਪન ˙ &YUFOE"*ࢀՃऀࢿํͱཧ༝ΛࣗવݴޠͰॻ͘ɻ"*͕Ϩ Ϗϡʔ͠ɺࢀՃऀͦΕΛ౿·͑ͯ࠷ऴܾఆ ˔
྆ํͷ"*Ͱɺࢿઌ͕Ұ෦ͷࠃۀछʹภΓʹ͘͘ͳͬͨɻಛʹ &YUFOE"*ͰɺࢿઌͷࠃɾҬ͕ΑΓ͕Δ͕͋ͬͨ ˔ ݁Ռͷຬ&YUFOE"*ͷํ͕ߴ͔ͬͨɻ 3FDPNNFOE"*৽͍͠ํੑΛ༩͑͘͢ɺೝෛՙখ͔ͬͨ͞ 33 Reicherts et al. AI, help me think-but for myself: Assisting people in complex decision-making by providing different kinds of cognitive support. CHI 2025. -
ूஂҙࢥܾఆʹհೖ͢Δ"* --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ "*அΛ౿·͑ͯूஂͰٞ͢Δ ໘Ͱɺଞऀͷಉௐ ରཱճආͰ"*ͷޡΓΛݟಀ͢ڪΕ ˔ "*அଟҙݟʹ͢Δ ٞհೖCPUl%FWJM`TBEWPDBUFz ΛఏҊ
˔ ࠶൜༧ଌΛࡐʹ࣮ݧɻٞհೖʹ ΑΓਫ਼্͕ɻಛʹ"*அͷ ͕༗ޮɻࢀՃऀ͕ࠜڌͷ໌֬ԽଞऀʹൃݴΛଅ༷͢ࢠ͋ͬͨɻ ˙ ͨͩ͠νʔϜϫʔΫࣗͨͪͷਖ਼͠͞ͷओ؍ධՁѱԽ 34 Chiang et al. Enhancing AI-assisted group decision making through LLM-powered devil's advocate. IUI 2024. 🔴☠ "*அଟҙݟʹ Ϙοτ͕ -
৮Λସ͢Δ"*หऀ --.ΛͲ͏͍ٙɺͲ͏ׂ͚ͮΔ͔ ˔ ҟͳΔഎܠͷूஂؒʢྫɿࠃֶੜWTཹֶੜʣͷٞෆ҆Λ͏ ˔ ଟࢹऔಘΛଅͨ͢Ίूஂٞʹհೖ͢Δ֎ूஂͷหऀ"*ΛఏҊ ˙ ֎ूஂͷΈͷٞϩάΛ༻͍ͯหऀ"*Λ࣮ ˔ ʮࠃࡍྈͰͷτϥϒϧճආͷͨΊͷ׆ಈܭըࡦఆʯʹ͍ͭͯ
ࠃֶੜʢूஂʣ͕ཹֶੜʢ֎ूஂʣͷหऀ"*ͱٞ͢Δ࣮ݧ ˔ কདྷతͳ৮ͷෆ͕҆ ݮ͞ΕΔ͕ݟΒΕͨ ˔ ·ͨɺࢀՃऀ͕ࣗ֎ूஂࢹΛ औΓೖΕΔέʔεͱหऀ"*ʹ ֎ूஂࢹΛҕͶΔέʔε͕͋ͬͨ 35 Hata et al. GroupEnvoy: A Conversational Agent Speaking for the Outgroup to Foster Intergroup Relations. CUI 2026 (accepted). +4"*ൃද <݄ ɺ04> ֎ूஂͷٞϩάΛ༻͍ หऀ"*Λ࣮ ूஂ"*ͷٞʹหऀ"*͕ࢀՃ -
4FDPOEPQJOJPO"*ैΛݮΒ͕͢ਫ਼্͛ͳ͍ ෳͷॿݴɺࢹʹѹྗʹͳΔ ˔ "*அʹՃ͑ͯଞऀPSଞ"*ͷஅ TFDPOEPQJOJPO Λఏࣔ͢Δ߹ ΛөըϨϏϡʔͷײۃੑఆΛࡐ࣮ͯ͠ݧ ˔ ଞऀஅͷఏࣔɺ"*ͷSFMJBODFΛ༗ҙʹԼ͛ͨɻ ಛʹ"*ͱଞऀ͕ରཱ͢Δͱɺେ͖͘Լ͕ͬͨ
˙ "*͕ਖ਼͍࣌͠"*ʹै͠ͳ͍ ͨΊਖ਼վળ͠ͳ͔ͬͨ ˙ ॴཁ࣌ؒ૿Ճͨ͠ ˔ ଞऀஅΛʮଞ"*ͷஅʯͱఏࣔ ͯ͠ɺޮՌมΘΒͳ͔ͬͨ 37 Lu et al. Does more advice help? The effects of second opinions in AI-assisted decision making. CSCW 2024. "*அ ଞऀͷஅ -
ಉ͡ਫ਼Ͱਓؒͷॿݴ͕ॏࢹ͞Ε͍͢ ෳͷॿݴɺࢹʹѹྗʹͳΔ ˔ ਓҎ্ͷਓؒBOEPS"*͕ॿݴ͢Δ ߹ΛՍۭͷҩྍஅͰ࣮ݧ ˔ ਓؒͱ"*͕ಉͷਫ਼ͳΒ ਓ͕ؒঢ়ใΛࢧ࣋ͨ࣌͠ͷํ ͕ɺ"*͕ࢧ࣋ͨ࣌͠ΑΓஅʹө ͞Ε͔ͬͨ͢
˔ ॿݴऀʹΑͬͯਫ਼͕ҟͳΔͱ͖ ਓ͔ؒ"*͔ͷӨڹখ͍͞ 38 Zhong et al. Drivers and influence of social conformity on decision making in human-AI teams. Scientific Reports 2026. ॿݴऀͱͦͷਫ਼ ॿݴऀͷஅ -
"*ͷΑΓ߹ҙͷํ͕ैʹӨڹ ෳͷॿݴɺࢹʹѹྗʹͳΔ ˔ ෳ"*͕ॿݴ͢Δ໘Ͱ"*ͷ ͱ"*߹ҙͷӨڹΛௐࠪɻ ऩ༧ଌɺ࠶൜༧ଌɺσʔτ༧ଌ͕ࡐɻ--.ੜͷ"*આ໌༩ɻ ˔ Ұ෦λεΫͰ"*͕"*ΑΓ༗ҙʹਖ਼ΛߴΊͨ ˙
"*Ͱ͜ͷޮՌͳ͔ͬͨ ˔ "*ͷ૿Ճ"*ଟҙݟͷ 4XJUDIʹӨڹ͠ͳ͔ͬͨɺ "*߹ҙ͕ߴ͍΄Ͳ"*ଟҙݟ ͷ4XJUDI্͕͕ͬͨ ˙ ࢀՃऀ͔Βʮ"*ಉ͕࢜ҙݟ ෆҰகͩͱࠞཚͨ͠ʯͱ͕͋ΓɺೝෛՙΛͨΒ͢ڪΕ 39 Tsuchiya et al. More isn't always better: Balancing decision accuracy and conformity pressures in multi-AI advice. CHI 2026. +4"*ൃද <݄ ɺ(4> -
ෳΤʔδΣϯτʹΑΔଟࢹ୳ࡧ ෳͷॿݴɺࢹʹѹྗʹͳΔ ˔ ෳ"*ΛଟܾͰͳ͘ଟࢹ୳ࡧʹ͏γεςϜl$IPJDF.BUFTz ˙ ʮझຯ༻ͷ৽͍͠ΧϝϥΛ୳͍ͯ͠Δʯͷ࣭ʹରͯ͠ʮը࣭ॏ ࢹʯʮॳ৺ऀ͚ʯʮܰྔॏࢹʯͷΤʔδΣϯτ͕༻ҙ͞ΕΔ ˙ ΤʔδΣϯτ܈ͱϢʔβͷ ରͰग़ͯ͘Δ0QUJPOT
$SJUFSJBΛཤྺͱͯ͠ه ˙ Ϣʔβཤྺ͔ΒॏཁͳͷΛ 1SFGFSFODFTQBDFͱͯ͠هɺ ࠷ऴஅ ˔ ࣗಈԽෳΤʔδΣϯτΑΓ ͍୳ࡧ͕ՄೳͱߴධՁ 40 Park et al. ChoiceMates: Supporting unfamiliar online decision-making with multi-agent conversational interactions. IUI 2026. ෳ"*͕ҟͳΔࢹΛ୲͏ ཤྺ͔ΒॏཁͳͷΛ Ϣʔβ͕நग़ ෳ"*ͱάϧʔϓνϟοτ