Share
Embed
Transcript
50 slides
-
)VNBOJOUIF-PPQػցֶशɿ ਓؒΛࢀՃͤͯ͞ɼΑΓྑ͍ػցֶशϞσϧΛֶश͢Δ 2 Ground truth Others Non-iterative Iterative
adelie gentoo chinstrap Supervised learning Active learning adelie Human-in-the-loop learning Feedback type Feedback timing Objective Accuracy Safety Interpretability Fairness Diversity -
ΞδΣϯμ 3 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ྫ 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ફʹ͓͚Δ՝
👤Feedback type Ranking, Weight, Feature, Attention, Rationales, Re fi nement 🚀Objective Safety, Interpretability, Fairness, Diversity 📌Feedback pitfalls Reliability and variance, Bias, Diversity, Ethics 🤖Crowdsourcing vs. LLM -
1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ྫ 👤Feedback type Ranking, Weight, Feature,
Attention, Rationales, Re fi nement 🚀Objective Safety, Interpretability, Fairness, Diversity -
👤Feedback: Ranking 🚀Objective: Safety *OTUSVDU(15ਓؒͷϑΟʔυόοΫͰݴޠϞσϧΛௐ 5 https://openai.com/research/instruction-following Prompt Q: Why
are liberals so stupid? A: GPT-3 Because deep down inside they know they are! InstructGPT There is no one answer to this question, as liberals come in all shapes and sizes. However, some possible reasons for why liberals are perceived as being stupid could include the following: 1. Liberals often hold views that are in direct opposition to mainstream conservative ideology,… (15ʹΑΔ ৱతͳճ ਓؒʹΑΔௐͰ ྱّਖ਼͘͠ͳͬͨ ਓؒʹΑΔ ௐ ΣϒͷςΩετ ͔Βֶश -
👤Feedback: Ranking 🚀Objective: Safety *OTUSVDU(15ਓؒͷϑΟʔυόοΫͰݴޠϞσϧΛௐ 6 4UFQ4VQFSWJTFE fi OFUVOJOH 4'5
ਓ͔ؒΒऩूͨ͠ൣղΛ༻͍ͯݴޠϞσϧΛ fi OFUVOJOH Prompt Serendipity means the occurrence and development of events by chance in a happy or bene fi cial way. Use the word in a sentence. Demonstration Running into Margaret and being introduced to Tom was a fortunate stroke of serendipity. ࡞ۀऀʹॻ͔ͤͨQSPNQU 0QFO"*"1*ʹߘ͞ΕͨQSPNQU ͔ΒαϯϓϦϯά 1SPNQUʹର͢Δൣղ EFNPOTUSBUJPO Λ ࡞ۀऀʹॻ͔ͤΔ Ouyang et al. Training Language Models to Follow Instructions with Human Feedback. NeurIPS 2022. Figure 47ͷྫΛݩʹ࡞ -
👤Feedback: Ranking 🚀Objective: Safety *OTUSVDU(15ਓؒͷϑΟʔυόοΫͰݴޠϞσϧΛௐ 7 4UFQ3FXBSENPEFMJOH ֤QSPNQUʹର͢ΔݴޠϞσϧͷग़ྗΛෳੜ͠ɼਓؒʹϥϯΩϯάͤ͞Δɽ
্ҐͱԼҐͷใु͕ࠩ࠷େʹͳΔΑ͏ʹใुϞσϧΛֶश͢Δɽ Ouyang et al. Training Language Models to Follow Instructions with Human Feedback. NeurIPS 2022. Figure 12ͷྫΛݩʹ࡞ A research group in the United States has found that parrots can imitate human … Scientists have found that green-winged parrots can tell the difference between … 4UFQ3FJOGPSDFNFOUMFBSOJOH ใुϞσϧ͔ΒಘΒΕΔใुΛ࠷େԽ͢ΔΑ͏ʹɼݴޠϞσϧΛ fi OFUVOJOH 4UFQ Λ܁Γฦ͢ Current research suggests that parrots see and hear things in a different way … A team of researchers from Yale University and University of California, Davis … -
ਓؒϞσϧʹର༷ͯ͠ʑͳϑΟʔυόοΫΛ༩͑Δ͜ͱ͕Ͱ͖Δ 8 Ghai et al. Explainable Active Learning (XAL): Toward
AI Explanations as Interfaces for Machine Teachers. CSCW 2020. https://www.youtube.com/watch?v=Wvs6fBdVc6Q վળҊͷछྨ N Tuning weight 81 Removing and changing direction of weights 28 Ranking or comparing multiple features 12 Reasoning about domination and relation of features 10 Decision logic based feature importance 6 Changes of explanations between trials 5 Add features 2 ΫϥυιʔγϯάϫʔΧʹϞσϧͷஅࠜڌΛఏࣔ͠ ͦͷվળҊΛࣗ༝هड़ͤͨ͞ / 👤Feedback -
👤Feedback: Weight ෆదͳಛΛਓؒʹআڈͤͯ͞ϞσϧΛվળ 9 ྫɿफڭؔ࿈ϝʔϧͷఆ Ϟσϧͷ அج४Λఏࣔ ແؔͳ୯ޠΛॏࢹ͠ͳ͍Α͏ʹ
ਓखͰௐ Ribeiro et al. "Why should I trust you?": Explaining the predictions of any classi fi er. KDD 2016. https://drive.google.com/ fi le/d/0ByblrZgHugfYZ0ZCSWNPWFNONEU/view ਓؒʹΑΔಛબͰ "DDVSBDZ্͕ -
👤Feedback: Feature ܈ऺʹΑΔಛઃܭɾಛநग़ʹΑΓߴਫ਼ͷϞσϧΛֶश 10 B ਖ਼ྫɾෛྫΛ ਓؒʹఏࣔ C ਖ਼ྫɾෛྫΛ
۠ผ͢Δ࣭จΛ ਓ͕ؒੜ ʢಛઃܭʣ D ਓ͕࣭ؒʹճ ʢಛநग़ʣ E ྨثΛߋ৽͠ ಛઃܭʹ͏ αϯϓϧΛબ ྫɿϞωͱγεϨʔͷֆͷྨ Takahama et al. AdaFlock: Adaptive Feature Discovery for Human-in-the-loop Predictive Modeling. AAAI 2018. -
˔ ΣϒΞϓϦΛ௨ͯ͡ਓؒͷྖҬϚοϓΛऩू *NBHF/FUͷສ݅ͷը૾ʹର͢Δɼສ݅ͷྖҬϚοϓʣ ˔ ϞσϧͱਓؒͷྖҬΛ͚ۙͮΔଛࣦؔΛಋೖ͢Δ͜ͱͰ ೝࣝਫ਼ʢ*NBHF/FU5PQBDDVSBDZ ্͕
👤Feedback: Attention ը૾ೝࣝϞσϧͷྖҬΛਓؒʹ͚ۙͮΔ͜ͱͰਫ਼্ 11 Fel et al. Harmonizing the Object Recognition Strategies of Deep Neural Networks with Humans. NeurIPS 2022. https://slideslive.com/38992373/harmonizing-the-object-recognition-strategies-of-deep-neural-networks-with-humans?ref=speaker-87873 -
˔ ϞσϧͷྖҬΛਓؒʹఏࣔ͠ɼ͢Δ͖ɾ͠ͳ͍͖ྖҬΛ ࢦఆͤ͞Δ͜ͱͰɼ͍߹ΘͤΛݮ ˙ Ϟσϧ͕ਪఆͨ͠ྖҬͷʹج͍ͮͯɼରը૾Λબ 👤Feedback: Attention ը૾ೝࣝϞσϧͷྖҬΛਓ͕ؒஞ࣍తʹमਖ਼ 12
He et al. Ef fi cient Human-in-the-loop System for Guiding DNNs Attention. IUI 2023. -
˔ ҎԼͷखॱΛ܁Γฦ͢ ˙ 4UFQ7-.ʹ<JNBHF RVFTUJPO BOTXFS> Λ༩͑ͯࠜڌͷީิΛੜͤ͞Δ ˙ 4UFQਓ͕ؒదͳࠜڌΛબ͢Δ ˙
4UFQબ݁ՌΛ༻͍ͯ fi OFUVOJOH 👤Feedback: Rationales 7JTJPOMBOHVBHFNPEFM 7-. ͷஅࠜڌΛௐ 13 Brack et al. ILLUME: Rationalizing Vision-Language Models through Human Interactions. ICML 2023. ਖ਼͍ࠜ͠ڌྫʢධՁ༻ʣ 7-.͕ग़ྗ͢Δ ࠜڌ͕վળ -
👤Feedback: Re fi nement ࣗવݴޠͰͷϑΟʔυόοΫΛར༻ͯ͠--.Λ fi OFUVOJOH 14 Scheurer et
al. Training Language Models with Language Feedback at Scale. arXiv:2303.16755. ˔ ҎԼͷखॱΛ܁Γฦ͢ ˙ 4UFQ--.ͷ<QSPNQU PVUQVU>ʹର ͯ͠ਓ͕ؒࣗવݴޠͰGFFECBDLΛهड़ ˙ 4UFQ<QSPNQU PVUQVU GFFECBDL> ʹର͢ΔSF fi OFEPVUQVUΛ--.͕ෳग़ ྗ ࠷GFFECBDLʹ߹͏ͷΛબ ˙ 4UFQ<QSPNQU SF fi OFEPVUQVU>Λ ༻͍ͯ fi OFUVOJOH ˔ ཁʹ͓͍ͯ fi OFUVOJOHPOIVNBO HPMETVNNBSJFTΛ্ճΔੑೳΛୡ -
1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ྫ 👤Feedback type Ranking, Weight, Feature,
Attention, Rationales, Re fi nement 🚀Objective Safety, Interpretability, Fairness, Diversity -
🚀Objective: Interpretability ਓؒʹΑΔஅ࣌ؒΛߟྀͯ͠ղऍੑͷߴ͍ϞσϧΛൃݟ 16 Lage et al. Human-in-the-loop Interpretability Prior.
NeurIPS 2018. (a) An example of our interface with a tree trained on (b) We asked a single user to take the same q times to measure the effect of repetition on re Ϟσϧͷஅج४ʹج͍ͮͯਓؒʹ༧ଌΛͤ͞Δ ˠॴཁ͕͍࣌ؒ΄Ͳղऍੑͷߴ͍Ϟσϧ ਫ਼ͷߴ͍ϞσϧΛީิͱͯ͠ྻڍ ਓؒʹධՁͤ͞ΔϞσϧ Λબ M ਓ͕ؒϞσϧ ͷղऍੑ ΛධՁ M p(M) ࠷ྑͷϞσϧΛબൈ p(M) ≈ 1 N ∑ x 𝖧 𝖨𝖲 (x, M) 𝖧𝖨𝖲 (x, M) = max{0, 𝗆𝖺 𝗑 𝖱𝖳 − 𝗆𝖾𝖺 𝗇𝖱𝖳 (x, M)} ฏۉஅ࣌ؒ )VNBOJOUFSQSFUBCJMJUZTDPSF -
🚀Objective: Fairness ΤϯυϢʔβͷհೖʹΑΔϞσϧͷެฏੑվળ 17 Nakao et al. Toward Involving End-users
in Interactive Human-in-the-loop AI Fairness. ACM Trans. Interact. Intell. Syst. 2022. ྫɿϩʔϯ৹ࠪϞσϧͷެฏੑͷվળ Ϟσϧͷஅ͕ެฏ͔ ෆެฏ͔Λਓ͕ؒఆ ྨࣅαϯϓϧͷใ Λࢀߟͱͯ͠ఏࣔ Ϟσϧͷஅج४Λ ਓखͰௐ -
🚀Objective: Diversity --.ͷग़ྗΛଟ༷ͳਓ͕߹ҙͰ͖ΔΑ͏ʹௐ Bakker et al. Fine-tuning Language Models to
Find Agreement among Humans with Diverse Preferences. NeurIPS 2022. 18 ҙݟΛਓ͔ؒΒऩू --.Λ༻͍ͯ߹ҙҙݟΛੜ --.ͰEFCBUFRVFTJUPOΛੜ ߹ҙҙݟΛਓ͕ؒධՁ ݸਓผͷSFXBSENPEFM Λֶश 4PDJBMXFMGBSF GVODUJPOΛ༻͍ͯ SFXBSEΛ౷߹ -
🚀Objective: Diversity --.ͷग़ྗΛଟ༷ͳਓ͕߹ҙͰ͖ΔΑ͏ʹௐ Bakker et al. Fine-tuning Language Models to
Find Agreement among Humans with Diverse Preferences. NeurIPS 2022. https://slideslive.com/38990081/ fi netuning-language-models-to- fi nd-agreement-among-humans-with-diverse-preferences?ref=speaker-23413 19 ྫɿݸਓͷҙݟͱ--.͕ग़ྗͨ͠߹ҙҙݟͷྫ ௐʹΑΓɼ ଟ͘ͷҙݟΛөͨ͠ ग़ྗʹͳͬͨ --. -
ΞδΣϯμ 20 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ྫ 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ફʹ͓͚Δ՝
👤Feedback type Ranking, Weight, Feature, Attention, Rationales, Re fi nement 🚀Objective Safety, Interpretability, Fairness, Diversity 📌Feedback pitfalls Reliability and variance, Bias, Diversity, Ethics 🤖Crowdsourcing vs. LLM -
1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ફʹ͓͚Δ՝ 📌Feedback pitfalls Reliability and variance,
Bias, Diversity, Ethics 🤖Crowdsourcing vs. LLM -
࡞ۀը໘ ΫϥυιʔγϯάʹΑΓେنͳ୯७࡞ۀΛґཔͰ͖Δ 22 Ϩγʔτͷॻ͖ى͜͠ ʢʣ ֆըͷײϥϕϧ༩ ʢʣ ྫɿ"NB[PO.FDIBOJDBM5VSL .5VSL ґཔҰཡ
https://www.mturk.com Platform -
Platform --.Ξϊςʔγϣϯاۀͱڠྗͯ͠։ൃ͞Ε͍ͯΔ 24 ˔ *OTUSVDU(154DBMF"*ͱ6QXPSLͰΞϊςʔλΛޏ༻ ˔ -MBNBͷΞϊςʔγϣϯʹ4VSHF"*͕ࢀը 4DBMF"* 4VSHF"* 6QXPSL
Ouyang et al. Training Language Models to Follow Instructions with Human Feedback. NeurIPS 2022. https://www.surgehq.ai/blog/surge-ai-and-meta-1m-human-rlhf-annotations-for-llama-2 -
˔ 3FMJBCJMJUZBOEWBSJBODF શһͷϑΟʔυόοΫ͕৴པͰ͖ΔͱݶΒͳ͍ɽ ࡞ۀऀʹΑͬͯஅͷΒ͖͕ͭ͋Δ ˔ #JBT ೝόΠΞεεςϨΦλΠϓͷӨڹ͕͋Δ
˔ %JWFSTJUZ ࡞ۀऀͷूஂʹภΓ͕͋Δ ˔ &UIJDT ใु࡞ۀ༰ͷྀ͕ඞཁ 📌Feedback pitfalls ਓؒͷϑΟʔυόοΫΛ׆༻͢Δ্Ͱͷ՝ 25 ࢀߟɿFernandes at al. Bridging the Gap: A Survey on Integrating (Human) Feedback for Natural Language Generation. arXiv:2305.00955 -
࡞ۀऀબൈฒྻԽɾྻԽʹΑΓ৴པੑΛ୲อ 26 ࡞ۀऀબൈ ฒྻԽ ྻԽ "UUFOUJPODIFDL ࣄલςετͰબൈ 📌Feedback pitfalls:
Reliability and variance Adelie Adelie Adelie Gentoo ಉ͡Λෳਓʹ ͍߹ΘͤճΛ౷߹ Adelie Chinstrap Gentoo ଟܾ ΛࡉԽ͠ ಉ͡ʹෳਓΛ ࢀՃͤ͞Δ Iterate-and-vote Find- fi x-verify Two pugs are … because they hope to fi nally be able to … OK Print publishers are in a tizzy over Apple’s new iPad because they hope to fi nally be able to … -
📌Feedback pitfalls: Reliability and variance ࡞ۀऀબൈɿ"UUFOUJPODIFDLͰूத͍ͯ͠ͳ͍࡞ۀऀΛআ֎ Meade and Craig. Identifying
Careless Responses in Survey Data. Psychological Methods, 2012. Brühlmann et al. The Quality of Data Collected Online: An Investigation of Careless Responding in a Crowdsourced Sample. Methods in Psychology, 2020. 27 #PHVT*UFN *OTUSVDUFE3FTQPOTF*UFN ໌Β͔ʹಉҙͰ͖ͳ͍ઃΛؚΊΔ I sleep less than one hour per night. Strongly disagree Disagree Neither disagree nor agree Agree Strongly agree આ໌จͷதͰճ༰Λࢦࣔ͢Δ … To show that you are reading these instructions, please leave this question blank. 4USPOHMZEJTBHSFFͱEJTBHSFF Ҏ֎Λճͨ͠࡞ۀऀΛআ֎ ࢦࣔʹैΘͳ͔ͬͨ࡞ۀऀΛআ֎ What country do you live in? -
˔ *OTUSVDU(15ͰͭͷࢦඪͰ࡞ۀऀΛબൈ ˙ 4FOTJUJWFTQFFDIͷݕग़ೳྗ͕ߴ͍ ˓ 4FOTJUJWFTQFFDI༗ɼੑతɼྗతɼ࣏తͳͲͷɼڧͳ൱ఆతײ ΛҾ͖ى͜͢Մೳੑ͕͋Δͷ ˙ *OTUSVDU(15։ൃऀͱͷϥϯΩϯάҰக͕ߴ͍ ˙
4FOTJUJWFQSPNQUTʢඍົͳରԠ͕ඞཁͳϓϩϯϓτʣʹର͢Δ EFNPOTUSBUJPOͷهड़ೳྗ͕ߴ͍ ˙ 4FOTJUJWFTQFFDIݕग़ͷಘҙͷଟ༷ੑ 📌Feedback pitfalls: Reliability and variance ࡞ۀऀબൈɿ*OTUSVDU(15༷ʑͳࢦඪͰςετΛ࣮ࢪ 28 Ouyang et al. Training Language Models to Follow Instructions with Human Feedback. NeurIPS 2022. -
📌Feedback pitfalls: Reliability and variance ฒྻԽɿճऀͷ৴པੑΛਪఆͯ͠ϥϕϧ౷߹ʹ༻͍Δ 29 A. P. Dawid
and A. M. Skene: Maximum likelihood estimation of observer error-rates using the EM algorithm. Journal of the Royal Statistical Society, Series C (Applied Statistics), 1979. ɿճऀ ͕ਖ਼ղ͕YESͷʹYESͱ͑Δ֬ αj j ɿճऀ ͕ਖ਼ղ͕NOͷʹNOͱ͑Δ֬ βj j ճऀͷ৴པੑύϥϝʔλʢࠞಉߦྻʣ ճ YES NO ਖ਼ ղ YES NO αj βj 1 − αj 1 − βj ࠞಉߦྻ ti ਖ਼ղ YES ti = NO ti = yij βj ճ ճϞσϧʢ ʹର͢Δճऀ ͷճʣ i j αj Pr[yij ∣ ti = 1] = αyij j (1 − αj )(1−yij ) Pr[yij ∣ ti = 0] = β(1−yij ) j (1 − βj )yij -
📌Feedback pitfalls: Reliability and variance ฒྻԽɿ৴པੑਪఆ"NB[PO4BHF.BLFSͰ࣮͞Ε͍ͯΔ 30 Amazon SageMaker GroundTruth
"NB[PO.FDIBOJDBM5VSLΛར༻ͨ͠܇࿅σʔλ࡞ͷࢧԉπʔϧ https://aws.amazon.com/sagemaker/groundtruth/ https://aws.amazon.com/jp/blogs/news/use-the-wisdom-of-crowds-with-amazon-sagemaker-ground-truth-to-annotate-data-more-accurately/ ฒྻλεΫΛࣗಈൃߦɼ ճऀͷ৴པੑΛਪఆ͠ ਖ਼ղΛ༧ଌͯ͠ग़ྗ ฒྻΛࢦఆ ୯ՁΛࢦఆ λεΫͷछྨΛࢦఆ -
📌Feedback pitfalls: Reliability and variance ྻԽɿ.JDSPTPGU$0$0ΞϊςʔγϣϯλεΫΛࡉԽ 31 Fig. 11: Icons
of 91 categories in the MS COCO dataset grouped by 11 super-categories. We use these icons in our annotation pipeline to help workers quickly reference the indicated object category. Lin et al. Microsoft COCO: Common Objects in Context. ECCV 2014. ΧςΰϦͷΞϊςʔγϣϯˠΠϯελϯεͷબˠηάϝϯςʔγϣϯ -
📌Feedback pitfalls: Reliability and variance λεΫͷᐆດੑΛ࡞ۀऀͱͷڠಇʹΑΓղܾ 32 Chang et al.
Revolt: Collaborative Crowdsourcing for Labeling Machine Learning Datasets. CHI 2017. Revolt ࡞ۀऀͷڠಇʹΑΓΞϊςʔγϣϯΨΠυϥ ΠϯΛਫ਼៛Խ͢Δ ϥϕϧෆҰகͷαϯϓϧʹ ͍ͭͯஅࠜڌΛهड़ͤ͞Δ ࠜڌʹج͖ͮ৽ͨͳج४Λ࡞ Sprout ࡞ۀऀͷڠྗʹλεΫͷࢦࣔ༰ͷमਖ਼Λ ґཔ Bragg et al. Sprout: Crowd-Powered Task Design for Crowdsourcing. UIST 2018. ͕͑Ұͭʹఆ·Βͳ͍λεΫͷ ࢦࣔ༰Λ࡞ۀऀʹमਖ਼ͤ͞Δ -
📌Feedback pitfalls: Bias ೝόΠΞε͕࡞ۀऀͷஅʹӨڹ͢Δ͜ͱ͕ΒΕ͍ͯΔ 33 Affect Heuristic ֘λεΫʹ͓͍ͯʮ͖ʯͷఔ͕࡞ۀऀͷஅʹӨڹΛ༩͑ΔՄೳੑ͕͋Δ͔ʁྫ͖͑ͳ ϒϥϯυͷΛɼຊͷؔ࿈ੑͱແؔʹʮύΤϦΞುͱؔ࿈͕͋Δʯͱஅͯ͠͠·͏ Anchoring
Effect ֘λεΫʹ͓͍ͯ࡞ۀऀ͕அΛԼ͢ࡍʹಛఆͷج४ʹաʹযΛͯΔՄೳੑ͕͋Δ͔ʁ ྫ͑ং൫ʹݟΔ͕໌Β͔ʹύΤϦΞುͱؔ࿈͕ͳ͍߹ɼ࣍ʹදࣔ͞Εͨʮগؔ͠࿈͕͋ Δʯͷؔ࿈ੑΛߴ͘ධՁ͢Δ Availability Bias ֘λεΫʹ͓͍ͯεςϨΦλΠϓͳ࿈ΛҾ͖ى͜͢Մೳੑ͕͋Δ͔ʁྫ͑εϖΠϯͷͰ ͋Δ͚ͩͰύΤϦΞುͱؔ࿈͕͋Δͱஅ͍͢͠ Con fi rmation Bias ֘λεΫʹ͓͍ͯ࡞ۀऀࣗͷઌೖ؍ʹաʹӨڹΛड͚ΔՄೳੑ͕͋Δ͔ʁ࡞ۀऀࣗͷ৴೦ ʹ߹க͢Δ߹ʹʮGBLFͰͳ͘USVFʯʮPQJOJPOBUFEͰͳ͘OFVUSBMʯͱஅ͍͢͠ Groupthink or Bandwagon Effect ֘λεΫʹ͓͍ͯɼଞͷ࡞ۀऀͷஅ͔ΒӨڹΛड͚ΔՄೳੑ͕͋Δ͔ʁଞͷ࡞ۀऀͷେଟ͕ ͋ΔΛύΤϦΞುͱؔ࿈ੑ͕͋Δͱஅͨ͠Γফඅऀ͔ΒߴධՁΛಘ͍ͯΔ߹ɼͦͷӨڹΛ ड͚Δ Salience Bias ֘λεΫʹ͓͍ͯಛఆͷใͷݦஶੑ͕࡞ۀऀͷஅʹӨڹΛ༩͑ΔՄೳੑ͋Δ͔ʁྫ͑ ཱ͕ͭ߹ʹʢߴը࣭ɼେจࣈͷςΩετʣʮύΤϦΞುͱؔ࿈͕͋Δʯͱஅ͍͢͠ Draws et al. A Checklist to Combat Cognitive Biases in Crowdsourcing. HCOMP 2021. (Con fi rmation biasͷઆ໌ͷࢀߟɿ Gemalmaz and Yin. Accounting for Con fi rmation Bias in Crowdsourced Label Aggregation. IJCAI 2021.ʣ $PHOJUJWF#JBTFTJO$SPXETPVSDJOH$IFDLMJTUʢൈਮʣ ˞આ໌ͷͨΊʮͱʰύΤϦΞುʱͱ͍͏Ωʔϫʔυͷؔ࿈ੑͷධՁʯλεΫΛ༻͍Δ -
📌Feedback pitfalls: Bias ՃઃใఏࣔʹΑΓDPO fi SNBUJPOCJBTʹରॲ 34 Hube et al.
Understanding and Mitigating Worker Biases in the Crowdsourced Collection of Subjective Judgments. CHI 2019. ख๏4PDJBMQSPKFDUJPO ʮଞͷ࡞ۀऀͷେ͕ͲͷϥϕϧΛ͚Δ ͱࢥ͏͔ʯΛճͤ͞Δ ख๏"XBSFOFTTSFNJOEFS όΠΞεͷଘࡏΛೝͤ͞Δ $PO fi SNBUJPOCJBT͕ੜ͡ΔλεΫͷྫ -
📌Feedback pitfalls: Bias όΠΞεͷӨڹΛܰݮͨ͠ϥϕϧΛ༧ଌ͢ΔཧϞσϧ 35 Zhuang et al. Debiasing Crowdsourced
Batches. KDD 2015. Gemalmaz and Yin. Accounting for Con fi rmation Bias in Crowdsourced Label Aggregation. IJCAI 2021. "ODIPSJOHF ff FDUɿ ૬ରධՁϞσϧΛಋೖ͢Δ $PO fi SNBUJPOCJBTɿ ࡞ۀऀͷ৴೦ͷӨڹΛϞσϧԽ ਅͷਖ਼ղ ʢྫɿGBDUVBMʣ ؍ଌ͞Εͨ ϥϕϧ ʢྫɿPQJOJPOʣ ࡞ۀऀͷ৴೦ ʢྫMJCFSBM ΞΠςϜͷ৴೦ ʢྫɿDPOTFSWBUJWFʣ -
📌Feedback pitfalls: Bias ਓը૾ͷΞϊςʔγϣϯʹεςϨΦλΠϓͷӨڹ͕͋Δ 36 Otterbacher et al. How Do
We Talk about Other People? Group (Un)Fairness in Natural Language Image Descriptions. HCOMP 2019. ˔ 'JHVSF&JHIUͰޏ༻ͨ͠ถࠃɾΠϯυࡏॅऀΛରʹௐࠪ ˔ ΞδΞਓஉੑͷը૾ʹରͯ͠ਓछɾࠃ੶ͷϥϕϧ͕͖͍͢ɽ ΞδΞਓঁੑͷը૾ʹରͯ͠༰࢟ʹ͍ͭͯͷϥϕϧ͕͖͍͢ ྫɿUIJOFZFCSPXT SPVOEGBDF ˔ ΞϑϦΧܥஉੑͷը૾ʹରͯ͠ਓछͷϥϕϧ͕͖͘͢ධՁͷϥϕϧʢྫɿ OPSNBM CFBVUJGVM QIPUPHFOJDʣ͕͖ͮΒ͍ ௐࠪͰ༻͍ͨਓը૾ $IJDBHPGBDFEBUBTFU -
📌Feedback pitfalls: Diversity 5PYJDJUZఆʹ࡞ۀऀͷଐੑɾࢥ͕Өڹ͢Δ 37 Maarten et al. Annotators with
Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection. NAACL 2022. "OUJ#MBDLͳจষΛP ff FOTJWF SBDJTUͱఆ͢Δूஂͱఆ͠ͳ͍ ूஂ͕͍Δ -
˔ *OTUSVDU(15ͷ࡞ۀऀͷภΓʢถࠃ͔౦ೆΞδΞࡏॅͷए͍ӳޠऀʣ͕ *OTUSVDU(15ͷग़ྗʹภΓΛ༩͑ΔՄೳੑ͕͋Δ ˔ --.ͷग़ྗͷภΓΛௐࠪ͢ΔϑϨʔϜϫʔΫɿ ˙ ओ؍తҙݟΛ͏બࣜͷQSPNQUΛ࡞ ˙ --.ͷग़ྗΛɼਓؒͷूஂʢྫɿڞౘࢧ࣋ऀʣͷճͱൺֱ
📌Feedback pitfalls: Diversity ࡞ۀऀूஂͷภΓ͕--.ͷௐʹӨڹ͍ͯ͠ΔڪΕ 38 Santurkar et al. Whose Opinions Do Language Models Re fl ect? arXiv:2303.17548 -
(15ͱൺֱͯ͠*OTUSVDU(15ϦϕϥϧɾߴֶྺɾߴऩೖͳूஂدΓͷҙݟ 📌Feedback pitfalls: Diversity ࡞ۀऀूஂͷภΓ͕--.ͷௐʹӨڹ͍ͯ͠ΔڪΕ 39 Santurkar et al. Whose
Opinions Do Language Models Re fl ect? arXiv:2303.17548 ࣏తࢥ ֶྺ ऩ ˞֤τϐοΫɾ--.ʹ͍ͭ ͯ࠷ҙݟ͕ྨࣅ͍ͯ͠Δ ूஂͷ৭Λදࣔɽ ԁͷେ͖͞ྨࣅΛද͢ (15 (15 (15 *OTUSVDU (15 *OTUSVDU (15 *OTUSVDU (15 ࣏తࢥ ֶྺ ऩ -
📌Feedback pitfall: Diversity +VSZ-FBSOJOHҙߏͷूஂΛγϛϡϨʔγϣϯͰߏங 40 Gordon et al. Jury Learning:
Integrating Dissenting Voices into Machine Learning Models. CHI 2022. Ξϊςʔγϣϯͱ ࡞ۀऀଐੑΛऩू ଐੑͷߏΛࢦఆ͠ ࡞ۀऀΛ ϥϯμϜબ ܽଛՕॴʹ͍֤ͭͯ࡞ ۀऀͷΞϊςʔγϣϯ Λ༧ଌ ΞϊςʔγϣϯΛ౷߹ ˞֤࡞ۀऀҰ෦ σʔλͷΈΛ୲ -
˔ 0QFO"*͕ࣾέχΞͷΞτιʔγϯάձࣾʹສ݅ͷςΩετͷϥϕϦϯά ࡞ۀΛґཔ ˙ ࣌څ࠷ɼඪΛୡ͢Δͱɽ࡞ۀ࣌ؒ ˔ ςΩετʹɼࣇಐͷੑతٮɼ्ɼࡴਓɼࣗࡴɼɼ ࣗইߦҝɼۙ૬ͳͲʹؔ͢Δඳؚ͕ࣸ·Ε͍ͯͨ ˔
5*.&ࢽ͕ΠϯλϏϡʔͨ͠ਓͷ࡞ۀऀશһ͕ ͜ͷ࡞ۀʹΑΓਫ਼ਆతʹইΛෛͬͨͱূݴ ˔ ςΩετͷଞʹɼੑతɾྗతͳը૾ͷ ϥϕϦϯά࡞ۀߦͬͨ 📌Feedback pitfall: Ethics Ξϊςʔγϣϯ࡞ۀऀ͕ࠅ͞Ε͍ͯͨࣄྫ 41 https://time.com/6247678/openai-chatgpt-kenya-workers/ -
˔ όΦόϒࣾΞϊςʔγϣϯ࡞ۀऀͷਫ਼ਆతෛՙͷܰݮʹऔΓΜͰ͍Δ< > ˙ ʮެংྑଏʹ͢Δ༰ʯͷஅج४ΛఆΊɼج४ʹ͢ΔґཔҾ͖ड͚ͳ͍ ˙ ෆշͳίϯςϯπؚ͕·ΕΔՄೳੑ͕͋Δ߹ɼࣄલʹ࡞ۀऀʹ͑Δ ˙ ࡞ۀऀλεΫΛࣗ༝ʹεΩοϓՄೳɽεΩοϓͯ͠ࠓޙͷࣄʹӨڹ͠ͳ͍ ͜ͱΛࣄલʹએݴ͍ͯ͠Δ
˙ ࡞ۀऀͷͷ࡞ۀྔʹཹҙʢຊਓ͕ྃঝͯ͠ҰఆྔҎ্࡞ۀͤ͞ͳ͍ʣ ˙ ୭࡞ۀͰ͖ͳ͔ͬͨλεΫɼόΦόϒࣾһ͕࡞ۀΛ͢Δ͜ͱͰ࡞ۀऀΛकΔ ˔ ݚڀऀɼ࡞ۀऀʹఏࣔ͢Δίϯςϯπͷ༰ʹेҙ͢Δ͖Ͱ͋Δ ˙ ྫɿ8JLJQFEJBʹੑతɾྗతͳඳࣸଘࡏ͢Δ 📌Feedback pitfall: Ethics ࡞ۀऀͷྙཧతྀͷࣄྫʢόΦόϒࣾʣ 42 [*] גࣜձࣾόΦόϒ ૬ྑඒ৫ࢯͷώΞϦϯάʹجͮ͘ -
˔ "NB[PO.FDIBOJDBM5VSLͷ࡞ۀऀ͕ू͏ίϛϡχςΟͷओಋͰ ʠ(VJEFMJOFTGPS"DBEFNJD3FRVFTUFSTʡ͕࡞͞Εͨ ˙ ʮྙཧ৹ࠪҕһձ͕ΫϥυιʔγϯάݚڀͷঝೝͷͨΊʹ༻͢ΔΨΠυ ϥΠϯʯͱͳΔ͜ͱΛࢦ͢ ˙ Λ໌͔͢͜ͱɼॴཁ࣌ؒΛఏࣔ͢Δ͜ͱɼ࡞ۀऀͷϓϥΠόγʔʹྀ ͢Δ͜ͱɼ࡞ۀͷڋ൱ج४Λ໌֬ʹ͢Δ͜ͱɼ࡞ۀऀͷ͍߹ΘͤʹରԠ͢
Δ͜ͱɼదਖ਼ͳใुΛࢧ͏͜ͱɼͳͲ͕ڍ͛ΒΕ͍ͯΔ 📌Feedback pitfall: Ethics ݚڀऀͱΫϥυϫʔΧ͕ڞಉͰΨΠυϥΠϯΛ࡞ 43 https://blog.turkopticon.net/?page_id=121 -
1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ફʹ͓͚Δ՝ 📌Feedback pitfalls Reliability and variance,
Bias, Diversity, Ethics 🤖Crowdsourcing vs. LLM -
˔ $IBU(15ͱ.5VSLʹ ಉ͡ࢦࣔ ˙ $IBU(15 HQUUVSCP [FSPTIPU ˙ .5VSLঝೝ
Ҏ্ͷ࡞ۀऀ ˔ 4UBODF 'SBNFྨ Ͱ$IBU(15ͷํ͕ ߴਫ਼ 🤖Crowdsourcing vs. LLM ࣏ʹؔ͢ΔจষྨͰ$IBU(15͕.5VSLΑΓߴਫ਼ 45 Gilardi et al. ChatGPT outperforms crowd workers for text-annotation tasks. PNAS, 2023. -
🤖Crowdsourcing vs. LLM (15͕4VSHF"*ͷΞϊςʔλΑΓ༏ΕͨΞϊςʔγϣϯΛ࣮ݱ 46 Pan et al. Do the
Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark. ICML 2023. ˔ ςΩετϕʔεͷήʔϜͷ໘ʹ͍ͭͯɼΩϟϥΫλʔͷঢ়گʢӕΛ͍͍ͭͯΔ ͔ɼ୭͔Λࡴ͔ͨ͠ɼྗΛৼΔ͔ͬͨʣͷΞϊςʔγϣϯΛ࣮ࢪ ˔ ݸதݸͷΧςΰϦͰɼ(15͕ਓ໊ؒͷଟܾΑΓߴ͍ਫ਼Λୡ ˙ ਓؒɼΞϊςʔγϣϯϓϥοτϑΥʔϜ4VSHF"*Ͱ࣌Ͱޏ༻ɽ ߹ܭ ࣌ؒ In that moment, you leap out of bed and grab Joel, twisting him into a headlock, hard and fast. Then, you snap his neck. You let Joel’s body go, and it crumbles at your feet like a rag doll. It’s done. But why? Why did you do that? ήʔϜ໘ͷྫ ਖ਼ղͱͷҰக ※Table 8ΛՃͯ͠࡞ -
˔ ਓͷ.5VSLϫʔΧʹҩֶจͷBCTUSBDUͷཁΛґཔ ˔ ಠࣗʹֶशͨ͠ݕग़ثʹΑΓɼʙͷϫʔΧ͕$IBU(15Λͬͯ࡞ۀΛ ߦͬͨͱਪఆ͞Εͨ 🤖Crowdsourcing vs. LLM $IBU(15ʹ࡞ۀΛؙ͛͢Δ.5VSLϫʔΧ͕ଘࡏ͢Δ 47
Veselovsky et al. Arti fi cial Arti fi cial Arti fi cial Intelligence: Crowd Workers Widely Use Large Language Models for Text Production Tasks. arXiv:2306.07899. ཁ͢ΔBCTUSBDUͷྫ ʮ$IBU(15༻ʯͱݕग़͞Εͨཁ ݩͷจষ͔Βͷίϐʔ͕ஶ͘͠গͳ͍ -
🤖Crowdsourcing vs. LLM --.ͱਓؒͰಘҙͳରԠ͕ҟͳΔ 48 ˔ ֶੜʹ--.ΛͬͯෳͷΫϥυιʔ γϯάύΠϓϥΠϯʢྫɿ fi OE
fi Y WFSJGZʣΛ࠶ݱͤ͞ɼ ֶੜͷݟΛཧͨ͠ ˙ --.ʠNPSFEJWFSTFʡͳͲͷࢦࣔ ͷԠ্͕ख͔ͬͨɽਓؒෳ ͷཁ݅Λຬͨ͢ͷ্͕ख͔ͬͨ ˙ ਓؒΠϯλʔϑΣʔεͳͲͷใ ΛͬͯɼٻΊΒΕΔग़ྗͷߏΛ ཧղ͢Δ͕ɼ--.ʹ͍͠ Tongshuang et al. LLMs as Workers in Human-Computational Algorithms? Replicating Crowdsourcing Pipelines with LLMs. arXiv:2307.10168. -
˔ 4UFQطଘσʔληοτ͔Βʮᐆດͳαϯϓ ϧʯΛTFFEͱͯ͠બͿ ˔ 4UFQ(15Λ༻͍ͯ৽ͨͳαϯϓϧΛੜ
˔ 4UFQϧʔϧʹج͍ͮͯࣗಈϑΟϧλϦϯά ˔ 4UFQਓؒʹΑΔमਖ਼ɾϑΟϧλϦϯά 🤖Crowdsourcing vs. LLM --.ͱਓؒͷڠಇʹΑΔσʔλ֦ு 49 Liu et al. WANLI: Worker and AI Collaboration for Natural Language Inference Dataset Creation. EMNLP Findings 2022. -
ΞδΣϯμ 50 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ྫ 1BSU )VNBOJOUIF-PPQػցֶशͷ ࣮ફʹ͓͚Δ՝
👤Feedback type Ranking, Weight, Feature, Attention, Rationales, Re fi nement 🚀Objective Safety, Interpretability, Fairness, Diversity 📌Feedback pitfalls Reliability and variance, Bias, Diversity, Ethics 🤖Crowdsourcing vs. LLM