Tweet
Tweet

Slide 1

Slide 1 text

ҙࢥܾఆ with AI͸ݸਓvsάϧʔϓͰมΘΔͷʁ ౦ژେֶ അ৔ݚڀࣨ D2 ༶໌఩ 2023/12/20 Human-AI Decision making ษڧձ Are Two Heads Beter Than One in AI-Assisted Decision Making? Comparing the Behavior and Performance of Groups and Individuals in Human-AI Collaborative Recidivism Risk Assessment Chun-Wei Chiang, Zhuoran Lu, Zhuoyan Li and Ming Yin CHI 2023 https://dl.acm.org/doi/10.1145/3544548.3581015

Slide 2

Slide 2 text

࿦จ֓ཁ w"*ͷࢧԉ͕͋ΔҙࢥܾఆΛߦ͏࣌ɺݸਓ୯ಠͱάϧʔϓʹΑΔҙࢥܾఆͰ͸ɺ ɹύϑΥʔϚϯε΍ϓϩηεʹҧ͍͕͋Δͷ͔Λௐࠪ w"*ͱਓؒʹΑΔҙࢥܾఆݚڀͷطଘݚڀΛ౿·͑ͯͭͷ؍఺͔Β ɹݸਓͱάϧʔϓͷධՁΛ͓͜ͳ͏ɻ wάϧʔϓ͸ݸਓͱൺ΂ͯ w"*ͷਖ਼͠͞ʹ͔͔ΘΒͣɺ"*ʹґଘ͢Δ܏޲͕͋Δ wެฏੑج४ʹΑͬͯ͸ެฏͳ൑அΛಋ͚Δ wؒҧͬͨ"*ͷ൑அΛࣗ৴Λ࣋ͬͯڋ൱͢ΔɹͳͲͷҧ͍͕͋ͬͨ 3

Slide 3

Slide 3 text

എܠɿ༷ʑͳ෼໺Ͱ"*ͷҙࢥܾఆࢧԉ 4 աڈσʔλ͔Β ൑அͨ͠Α ΞυόΠε ΞυόΠε ΞυόΠε

Slide 4

Slide 4 text

എܠɿػց͕ΞυόΠεɺਓ͕ؒ࠷ऴܾఆ 5 ΞυόΠε ͜ͷ൑அͷํ͕
 ͍͍ͱࢥ͏ ΞυόΠεΛݩʹ࠷ऴ൑அΛͲ͏͢Δ͔ܾΊΔ

Slide 5

Slide 5 text

എܠɿ5XPIFBETBSFCFUUFSUIBOPOF 6 ຊݚڀͷର৅ طଘݚڀͷର৅ άϧʔϓɺݸਓͷҙࢥܾఆ with AI ͰԿ͔ҧ͍͕͋Δͷ͔ʁ

Slide 6

Slide 6 text

എܠɿ5XPIFBETBSFCFUUFSUIBOPOF 7 ຊݚڀͷର৅ طଘݚڀͷର৅ άϧʔϓɺݸਓͷҙࢥܾఆ with AI ͰԿ͔ҧ͍͕͋Δͷ͔ʁ

Slide 7

Slide 7 text

४උɿͭͷ؍఺ʹ஫໨͢Δ 8 1. Decision Accuracy 3. Decision Confidence 5. Fairness 2. Reliance on AI 4. Understanding of AI 6. Accountability

Slide 8

Slide 8 text

४උɿ%FDJTJPO"DDVSBDZ 9 ؒҧͬͨAIʹ௼ΒΕͯؒҧ͏͜ͱʹ஫ҙ… [1,2] ਓؒͷҙࢥܾఆʹ͓͚Δਫ਼౓޲্Λ໨ࢦ͢

Slide 9

Slide 9 text

४උɿ3FMJBODFPO"* 10 ໡໨తʹAIΛ৴པ →Over reliance Algorithm appreciation[4,5] AIΛ৴པ͠ͳ͍ →Under reliance Algorithm aversion[3] AIͷΞυόΠε͕࠷ऴ൑அʹͲΕ͘Β͍ӨڹΛड͚Δ͔ʁ

Slide 10

Slide 10 text

४උɿ%FDJTJPO$POpEFODF 11 AIʹର͢ΔConfidence → AIͷೳྗΛͲΕ͚ͩ৴͍ͯ͡Δ͔ ࣗ෼ʹର͢ΔConfidence → ࣗ෼ͷೳྗΛͲΕ͚ͩ৴͍ͯ͡Δ͔ ৴པ౓߹͍͕ద੾ͳར༻ʹӨڹΛ༩͑Δ[6]

Slide 11

Slide 11 text

४උɿ6OEFSTUBOEJOHPG"* 12 ٬؍తɿ AIͷ༧ଌΛߟ͑ͤͯ͞ AIΛཧղͰ͖͔ͨΛධՁ AIͷ൑அΛ׬ᘳʹ ཧղͰ͖ͨͧ!! ओ؍తɿ ͲΕ͘Β͍ཧղͰ͖͔ͨΛճ౴ͤ͞Δ ৴པͱར༻Λଅͨ͢ΊʹAIͷৼΔ෣͍Λཧղ͍ͤͨ͞[7]

Slide 12

Slide 12 text

४උɿ'BJSOFTT 13 Ӊ஦ਓ͸μϝ ਓؒ͸͍͍Α ಛఆͷଐੑʹରͯ͠ෆ౰ͳ൑அΛ͠ͳ͍Α͏ʹ͢Δ[8] P(Y ∣ A = 0) = P(Y ∣ A = 1) Y͸༧ଌ஋ɺA͸ੑผ΍ਓछͳͲͷଐੑɻଐੑ͝ͱͷ༧ଌൺ཰͕ಉ͡Ͱ͋Δ͜ͱ͕ެฏ

Slide 13

Slide 13 text

४උɿ"DDPVOUBCJMJUZ 14 ͜ͷܾఆ ୭͕੹೚࣋ͭͷʁ άϧʔϓ಺ͷܾఆͰ੹೚͕ͩΕʹ͋Δ͔ௐࠪͨ͠΋ͷ͸গͳ͍

Slide 14

Slide 14 text

࠶ܝɿͭͷ؍఺ʹ஫໨͢Δ 15 1. Decision Accuracy 3. Decision Confidence 5. Fairness 2. Reliance on AI 4. Understanding of AI 6. Accountabilidy

Slide 15

Slide 15 text

࣮ݧλεΫɿ೥ޙͷ࠶൜ϦεΫΛ༧ଌ બ୒ཧ༝ɿ ܐࣄ࢘๏ʹ͓͚Δूஂҙࢥܾఆ͕࣮ࡍʹ͋Δ ܐࣄ࢘๏ʹ"*Λ׆༻͢Δͷ͕ҰൠతʹͳΓͭͭ͋Δ ͜ͷ൑அ͸ಛʹެฏੑ͕໰୊ʹͳ͍ͬͯΔ 16

Slide 16

Slide 16 text

࣮ݧɿશମతྲྀΕ 17 ੹೚ׂ߹Λࣗݾਃࠂ ࿅श໰୊ʹ͓͚Δ
 ϑΟʔυόοΫ

Slide 17

Slide 17 text

࣮ݧɿݸਓPSάϧʔϓʹΑΔҙࢥܾఆ 18 άϧʔϓઃఆɹɹ ݸਓઃఆɹɹɹ 3ਓ1άϧʔϓ AIͷΞυόΠεΛݟ ͨͱ࿩͠߹ͬͯ࠷ऴ ճ౴Λܾఆ AIͷΞυόΠεΛݟ ͨޙʹ࠷ऴܾఆ

Slide 18

Slide 18 text

࣮ݧɿఆྔධՁํ๏ɹͦͷ̍ w%FDJTJPO"DDVSBDZɿݸਓPSάϧʔϓͷ࠷ऴܾఆͷਫ਼౓ΛධՁ w3FMJBODFPO"*ɿ࠷ऴܾఆͱ"*ͷ൑அͷҰக౓߹͍ͰධՁ ɹɹɹɹɹɹɹɹɹ0WFSSFMJBODFˠ"*͕ؒҧ͍ͬͯΔதͰҰக͍ͯ͠Δ൑அ 6OEFSSFMJBODFˠ"*͕ਖ਼͍͠தͰ൑அ͕ޡ͍ͬͯΔ΋ͷ w%FDJTJPO$POpEFODFɿλεΫऴྃޙͷΞϯέʔτͰࣗݾਃࠂ 19

Slide 19

Slide 19 text

࣮ݧɿఆྔධՁํ๏ɹͦͷ w6OEFSTUBOEJOHTPG"*ɿ wΞϯέʔτͰॏཁಛ௃ྔΛऩूɻϞσϧͷॏཁಛ௃ྔͱͷ૬ؔΛܭଌ w'BJSOFTTJOEFDJTJPONBLJOHɿ wෳ਺ͷެฏੑධՁࢦඪΛ༻͍ͯଌఆ͢Δɻ w"DDPVOUBCJMJUZɿ wλεΫऴྃޙʹ࣮ݧࢀՃऀʹʮ୭͕੹೚Λ͍࣋ͬͯΔ͔ʯΛࣗݾਃࠂͰऩू 20

Slide 20

Slide 20 text

࣮ݧ݁Ռɿ%FDJTJPO"DDVSBDZ 21 άϧʔϓઃఆ ݸਓઃఆ ਫ਼౓ ༗ҙࠩͳ͠ ຊ൪λεΫͰͷਫ਼౓ wਫ਼౓ͷ਺஋తʹ͸άϧʔϓͷํ͕ݸਓΑΓ΋ਫ਼౓͕ߴ͍ɻͨͩ͠༗ҙࠩ͸ͳ͍

Slide 21

Slide 21 text

࣮ݧ݁Ռɿ3FMJBODFPO"* wάϧʔϓͷ΄͏͕"*ͷܾఆʹґଘ͍ͯ͠Δʢಉ͡ʹ͢Δʣ܏޲͕͋Δ w0WFSSFMJBODF 6OEFSSFMBJBOEFͷ྆ํͷಛ௃Λ֬ೝ wˠ"*ͷਖ਼͠͞ʹ͔͔ΘΒͣɺ"*Ϟσϧʹґଘ͍ͯ͠Δɻ 22

Slide 22

Slide 22 text

࣮ݧ݁Ռɿ%FDJTJPO$POpEFODF wݸਓ͸൑அ͕ޡ͍ͬͯΔ࣌ʹࣗ৴Λࣦ͏܏޲͕͋Δ wάϧʔϓͷํ͸"*ʹैΘͳ͍࣌΋ࣗ৴Λ࣋ͬͯैΘͳ͍ͱ͍ͯ͠Δ 23 AIʹಉҙ͢Δ AIʹಉҙ͠ͳ͍ શମ

Slide 23

Slide 23 text

࣮ݧ݁Ռɿ6OEFSTUBOEJOHPO"* wάϧʔϓઃఆͱݸਓઃఆͰ"*ͷཧղ౓ʹ͕ࠩͳ͍ wʢͳΜͳΒཧղͰ͖͍ͯͳ͍ͷͰ͸ʁʣ 24 άϧʔϓઃఆ ݸਓઃఆ ϐΞιϯ܎਺ ༗ҙࠩͳ͠ ຊ൪λεΫͰͷཧղ౓

Slide 24

Slide 24 text

࣮ݧ݁Ռɿ%FDJTJPO.BLJOH'BJSOFTT wશମతʹ͸άϧʔϓͷํ͕ެฏͳ൑அΛ΋ͨΒ͍ͯ͠Δɻͨͩ͠༗ҙࠩ͸ͳ͠ w ʹؔͯ͠ͷΈάϧʔϓͱݸਓͰ༗ҙ͕ࠩΈΒΕͨ ΔAcc 25 ूஂެฏੑ ूஂެฏੑ ूஂެฏੑ ूஂެฏੑ ݸผެฏੑ શͯ0ʹ͍ۙ΄Ͳެฏ

Slide 25

Slide 25 text

࣮ݧ݁Ռɿ"DDPVOUBCJMJUZ w࠷ऴܾఆ͕ਖ਼͍࣌͠͸ɺࣗ෼ͨͪʹ੹೚ΛׂΓ౰͍ͯͯΔ w࠷ऴܾఆ͕ؒҧ͏࣌ɺ"*ʹ੹೚ΛׂΓ౰ͯΔ܏޲͸ݸਓͱάϧʔϓͰ͕ࠩͳ͍ 26 ਓؒʹର͢Δ੹೚౓߹͍ AIʹର͢Δ੹೚౓߹͍

Slide 26

Slide 26 text

࣮ݧ݁Ռɿఆྔ݁Ռͷ·ͱΊ 27 1. Decision Accuracy ݸਓͱάϧʔϓͰେࠩͳ͍ 2. Reliance on AI άϧʔϓ͸AIʹґଘ͕ͪ͠ 3. Reliance on AI άϧʔϓ͸ࣗ৴Λ࣋ͬͯ AIΛڋ൱͢Δ 4. Understanding on AI ݸਓͱάϧʔϓͰେࠩͳ͍ 5. Fairness ͋Δج४Ͱ͸άϧʔϓ͸ެฏʹ 6. Accountability ࠷ऴ൑அ͕ؒҧ͏ͱ͖ ݸਓ΋άϧʔϓ΋܏޲͸ಉ͡

Slide 27

Slide 27 text

ఆੑධՁɿάϧʔϓ͸"*ΛͲ͏ར༻͢Δ͔ "*ͷ൑அΛج४఺ͱͯ͠༻͍Δ w"*ͷ൑அΛΞϯΧʔͱͯٞ͠࿦ΛਐΊΔ "*ͷ൑அΛ్தͰৼΓฦΔ w"*ͷ൑அΛਖ਼౰Խɺվળ w"*ͷ൑அͰࣗ෼ͷҙݟΛཪ෇͚Δ ൑அͷλΠϒϨʔΧ wٞ࿦͕፰߅ͨ࣌͠ʹར༻ɻ"*ʹڧ͍൓ର͕͋Δ࣌͸"*͸ڋ൱͞ΕΔ 28

Slide 28

Slide 28 text

ٞ࿦ɿ"*ʹґଘ͢Δ܏޲ "*ͳ͠ͷύΠϩοτݚڀͱൺֱ w"*ͳ͠Ͱಉ༷ͷλεΫΛߦͬͨ࣌ɺਫ਼౓Ͱ͋Γɺճ౴ʹࣗ৴͕ͳ͍ w·ͨάϧʔϓͷٞ࿦Ͱ΋ҙݟͷ৯͍ҧ͍͕໨ཱͬͨ wˠ"*ʹґଘ͢Δ܏޲͸͜ͷΑ͏ͳಛ௃͔Βͩͱߟ͑ΒΕΔ 29

Slide 29

Slide 29 text

ٞ࿦ɿެฏੑʹ͍ͭͯ wެฏੑج४ͷઃఆʹΑͬͯ͸ɺݸਓΑΓ΋άϧʔϓͷํ͕ެฏʹͳͬͨ wطଘݚڀͰ͸ɺݸਓͱൺ΂ɺάϧʔϓͷҙࢥܾఆ͕ެฏʹͳΔ܏޲͕ใࠂ w৺ཧֶ෼໺ͰऔΓ্͛ΒΕ͓ΓɺάϧʔϓͰ࿩͠߹͏ϓϩηε͕େࣄ wάϧʔϓ಺ͷೝ஌ଟ༷ੑ͕ߴ͍άϧʔϓͷํ͕ެฏͳ൑அʹͳΓ͕ͪ 30

Slide 30

Slide 30 text

·ͱΊ w"*Λར༻ͨ͠ݸਓͱάϧʔϓͷҙࢥܾఆʹ͓͚Δҧ͍ʹ͍ͭͯௐࠪ wάϧʔϓ͸ݸਓͱൺ΂ͯ w"*ͷਖ਼͠͞ʹ͔͔ΘΒͣɺ"*ʹґଘ͢Δ܏޲͕͋Δ wެฏੑج४ʹΑͬͯ͸ެฏͳ൑அΛಋ͚Δ wؒҧͬͨ"*ͷ൑அΛࣗ৴Λ࣋ͬͯڋ൱͢ΔɹͳͲͷҧ͍͕͋ͬͨ w·ͨάϧʔϓͰ"*ΛͲͷΑ͏ʹར༻͢Δ͔؍࡯Ͱ͖ͨ 31

Slide 31

Slide 31 text

ࢀߟจݙ <>$IVO8FJ$IJBOHBOE.JOH:JO:PV`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`$IJDBHP`EFDFQUJWF 5PXBSET#VJMEJOH.PEFM%SJWFO5VUPSJBMTGPS)VNBOT <>+VMJB"OHXJO +FG-BSTPO 4VSZB.BUUV BOE-BVSFO,JSDIOFS.BDIJOF#JBT1SP1VCMJDB 32