Upgrade to PRO for Only $50/Year—Limited-Time Offer! 🔥

人間と人工知能の協働

Yukino Baba
January 20, 2023

 人間と人工知能の協働

2023年1月20日(金)「高校生と大学生のための金曜特別講座」

Yukino Baba

January 20, 2023
Tweet

More Decks by Yukino Baba

Other Decks in Education

Transcript

  1. 2 അ৔ઇ೫ʢ͹͹Ώ͖ͷʣ ౦ژେֶ૯߹จԽݚڀՊ޿ҬՊֶઐ߈ɾ।ڭत ത࢜ʢ৘ใཧ޻ֶʣ ઐ໳˔ਓ޻஌ೳ ػցֶश ݚڀτϐοΫ˔)VNBO$PNQVUBUJPO 
 ɹɹɹɹɹɹɹ)VNBO"*$PMMBCPSBUJPO 

    ɹɹɹɹɹɹɹ)VNBOJOUIF-PPQ.BDIJOF-FBSOJOH ஶॻ˔ʮώϡʔϚϯίϯϐϡςʔγϣϯͱΫϥ΢υιʔγϯάʯ ɹɹɹɹʢߨஊࣾαΠΤϯςΟϑΟοΫʣ
  2. ͍·ͷਓ޻஌ೳ͸ෳࡶͳจষͷཧղ͕ۤखͰ͋Δ 11 ࣭໰จͷग़యɿS. Baron-Cohen et al.: Does the Autistic Child

    have a “Theory of mind” ? Cognition, Vol. 21, Issue 1, 1985. ճ౴จ͸https://beta.openai.com/playground Λར༻ͯ͠ߨࢣ͕࡞੒ɻ
  3. 7J[8J[ࢹ֮ো͕͍ऀͷ࣭໰Ԡ౴γεςϜ 15  -tap to take a photo.  -tap

    to begin recording your question and again to stop.        side,     User     ? Web Server - Database - Local Client Remote Services and Worker Interface 4UFQϢʔβ͕࣭໰Λ౤ߘ ίʔϯͷ؈͸ͲΕʁ 4UFQγεςϜ಺෦ͷਓ͕ؒճ౴ Ұ൪ӈͷ؈Ͱ͢ J. P. Bigham et al.: VizWiz: Nearly Real-time Answers to Visual Questions. In Proceedings of the 23rd annual ACM symposium on User interface software and technology (UIST), 2010.
  4. ;FOTPSTਓؒΛ࢖ͬͨ؂ࢹγεςϜ 17 G Laput et al. Zensors: Adaptive, Rapidly Deployable,

    Human-Intelligent Sensor Feeds. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI), 2015. How many glasses need a re fi ll? or… I can’t Tell https://www.dreamstime.com/empty-white- wine-glasses-table-restaurant-bar-setting- close-up-alcohol-image215682647 ෼͓͖ʹਓؒʹը૾Λૹ৴ͯ͠ 
 ໰͍߹ΘͤΔ
  5. ;FOTPSTਓؒΛ࢖ͬͨ؂ࢹγεςϜ 18 ਓؒͷճ౴Λ༻͍ͯ 
 ਓ޻஌ೳΛֶश 
 
 
 े෼ͳਫ਼౓ʹୡͨ͠Β 


    ਓؒͰ͸ͳ͘ਓ޻஌ೳΛར༻ G. Laput et al. Zensors: Adaptive, Rapidly Deployable, Human-Intelligent Sensor Feeds. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI), 2015.
  6. SF$"15$)"ɿॻ੶ͷจࣈೝࣝγεςϜʹਓؒΛ૊ΈࠐΉ 23 morning 4UFQ݁Ռ͕ෆҰகͷͱ͖ʹਓؒʹ໰͍߹ΘͤΔ moroiog morpipg 4UFQॻ੶தͷจࣈΛ̎ͭͷਓ޻஌ೳγεςϜʹೝࣝͤ͞Δ L. von Ahn

    et al.: reCAPTCHA: Human-Based Character Recognition via Web Security Measures. Science, Vol. 321, Issue 589, 2008. ਓ޻஌ೳʹ͸೉͍͠จࣈΛਓؒʹೝࣝͤ͞Δ
  7. ʮਓ͔ؒϘοτ͔ͷ൑ఆʯͱʮॻ੶ͷจࣈೝࣝʯΛಉ࣌ʹߦ͏ 25 ᶃਓ͔ؒϘοτ͔ͷ൑ఆɹᶄॻ੶ͷจࣈೝࣝ SF$"15$)"ͷ໨త Type the word Type the word

    reCAPTCHA ೋͭͷจࣈΛೝࣝ͢ΔΑ͏ʹࢦࣔ͢Δ ˞ͲͪΒ͕ਖ਼ղط஌͔͸఻͑ͳ͍ ਖ਼ղط஌ ਖ਼ղະ஌ L. von Ahn et al.: reCAPTCHA: Human-Based Character Recognition via Web Security Measures. Science, Vol. 321, Issue 589, 2008. ᶄʮਓؒɹʯͷ݁Ռ͚ͩΛ 
 ɹೝࣝ݁Ռͱͯ͠࠾༻ ᶃɹਖ਼ղͳΒˠਓؒ 
 ɹෆਖ਼ղͳΒˠϘοτ
  8. ඼࣭อূͷͨΊʹʮਓؒʯͷճ౴͚ͩ࠾༻͢Δ 26 overlooks morning reCAPTCHA ਖ਼ղط஌ ਖ਼ղະ஌ L. von Ahn

    et al.: reCAPTCHA: Human-Based Character Recognition via Web Security Measures. Science, Vol. 321, Issue 589, 2008. ਖ਼ղͳͷͰ 
 ʮਓؒɹʯͱ൑ఆ ʮਓؒɹʯͷೝࣝ݁ՌͳͷͰ 
 ࠾༻⭕
  9. ʮϘοτʯͷճ౴͸࠾༻͠ͳ͍ 27 overfooRs morpipg reCAPTCHA ਖ਼ղط஌ ਖ਼ղະ஌ L. von Ahn

    et al.: reCAPTCHA: Human-Based Character Recognition via Web Security Measures. Science, Vol. 321, Issue 589, 2008. ෆਖ਼ղͳͷͰ 
 ʮϘοτɹʯͱ൑ఆ ʮϘοτɹʯͷճ౴ͳͷͰ 
 ෆ࠾༻❌
  10. &41ήʔϜɿը૾΁ͷϥϕϧ෇͚ΛήʔϜԽ 29 ਓ޻஌ೳͷֶशʹ༻͍ΔͨΊɼը૾ͷλάΛਓ͔ؒΒूΊΔ ˔ ૬खϓϨΠϠʔͱಉ͡λάΛ౴ ͑ͨΒϙΠϯτ͕΋Β͑Δ 
 ʮήʔϜʯʹ͢Δ͜ͱͰ 
 ࢀՃΛଅ͢

    ˔ σλϥϝͳճ౴ͷ๷ࢭʹ΋ 
 ໾ཱͭ L. von Ahn et al.: Designing games with a Purpose. Communication of the ACM, Vol.51, Issue 8, 2008
  11. ੜెͷճ౴͚͔ͩΒࢼݧͷਖ਼ղΛ༧ଌ͢ΔΑ͏ͳ΋ͷ ճ౴͔Β֤ࣗͷ৴པੑΛਪఆ֤͠໰୊ͷਖ਼ղΛ༧ଌ͢Δ 34 ਖ਼ղ ໰୊ NO YES YES YES YES

    YES NO YES YES YES YES NO ? ? ? “Is a bird in 
 the picture?" https://commons.wikimedia.org/wiki/File:American_ fl amingo_(Phoenicopterus_ruber).JPG
  12. ti ਖ਼ղ YES ti = NO ti = yij βj

    ճ౴ ճ౴Ϟσϧʢ໰୊ ʹର͢Δճ౴ऀ ͷճ౴ʣ i j αj Pr[yij ∣ ti = 1] = αyij j (1 − αj )(1−yij ) Pr[yij ∣ ti = 0] = β(1−yij ) j (1 − βj )yij ճ౴ϞσϧΛߏஙͯ͠ਖ਼ղ༧ଌʹ༻͍Δ 35 ɿճ౴ऀ ͕ਖ਼ղ͕YESͷ໰୊ʹYESͱ౴͑Δ֬཰ αj j ɿճ౴ऀ ͕ਖ਼ղ͕NOͷ໰୊ʹNOͱ౴͑Δ֬཰ βj j ճ౴ऀͷ৴པੑύϥϝʔλʢࠞಉߦྻʣ ճ౴ YES NO ਖ਼ 
 ղ YES NO αj βj 1 − αj 1 − βj ࠞಉߦྻ 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.
  13. ˔ ଞਓͷ౴͑Λ࢖͍ղ͔ͤΔ͜ͱͰਖ਼͍͠౴͑ΛҾ͖ग़͢ ˙ 4PZMFOUਓؒΛ࢖ͬͨจষߍਖ਼γεςϜ  'JOE໰୊ͷ͋ΔՕॴͷݕग़  'JYߍਖ਼ͷ࣮ࢪ  7FSJGZߍਖ਼͕ਖ਼͍͔֬͠ೝ

    ௚ྻ໰߹ͤʹΑΔ඼࣭อূ 36 When the crowd is finished, Soylent calls out the edited sections with a purple dashed underline. If the user clicks on the error, a drop-down menu explains the problem and offers a list of alternatives. By clicking on the desired alter- native, the user replaces the incorrect text with an option of Figure 2. Crowdproof is a human-augmented proofreader. The drop-down explains the problem (blue title) and suggests fixes (gold selection). M. S. Bernstein et al.: Soylent: a Word Processor with a Crowd Inside. Communications of the ACM, Vol. 58, Issue 8, 2015. Find Fix Verify
  14. ˔ %"31"3FE#BMMPPO$IBMMFOHF੺͍෩ધΛ୳͢ίϯςετ ˙ શถՕॴʹ੺͍෩ધ🎈Λઃஔ ˙ ࠷ॳʹશͯͷ੺͍෩ધΛݟ͚ͭͨνʔϜʹ৆͕ۚࢧ෷ΘΕΔ ˙ ༏উνʔϜ͸ʮ෩ધΛݟ͚ͭͦ͏ͳਓʯΛଞऀ͔ΒͷਪનͰൃݟ ਪનʹΑΔ඼࣭อূ 37

    contributed articles platform for viral collaboration that used recursive incentives to align the public’s interest with the goal of win- ning the Challenge. This approach was inspired by the work of Peter S. Dodds et al.5 that found that success in us- ing social networks to tackle widely distributed search problems depends on individual incentives. The work of Mason and Watts7 also informed the use of financial incentives to motivate crowdsourcing productivity. The MIT team’s winning strategy was to use the prize money as a finan- cial incentive structure rewarding not only the people who correctly located balloons but also those connecting the finder to the MIT team. Should the team win, they would allocate $4,000 in prize money to each balloon. They promised $2,000 per balloon to the Figure 1. Locations in the DARPA Red Balloon Challenge. Figure 2. Example recursive incentive-structure process for the MIT team. J. C. Tang et al.: Re fl ecting on the DARPA Red Balloon Challenge. Communications of the ACM, Vol. 54, Issue 11, 2011.
  15. ˔ ώϡʔϚϯίϯϐϡςʔγϣϯ͸ɼਓ޻஌ೳͱਓؒͷڠಇʹΑΓ 
 ͜ͱͰͲͪΒ͔Ұํ͚ͩͰ͸ղܾ͕೉͍͠໰୊Λղܾ͢Δ ˔ ώϡʔϚϯίϯϐϡςʔγϣϯͷ՝୊͸ಈػ͚ͮͱ඼࣭อূ ˙ ಈػ͚ͮ 
 ήʔϜԽ΍ɼΞΫηε੍ޚɾ޷ح৺ɾۚમͳͲΛใुʹ͢Δ͜ͱͰ

    ਓؒͷࢀՃΛଅ͢ ˙ ඼࣭อূ 
 ௚ྻ໰߹ͤɾฒྻ໰߹ͤʹΑΓෳ਺ਓΛ૊Έ߹ΘͤͨΓɼ 
 ਪનʹΑΓ৴པͰ͖ΔਓΛݟ͚ͭͯɼਖ਼͍͠ճ౴ͷ֫ಘΛ໨ࢦ͢ ·ͱΊ 39