
å°éå®¶ã®èŠç¹ã§èªã¿è§£ãïŒãã¯ãããžãŒãšç€ŸäŒãç¹ããããŸèªãã¹ãæŽæž5éž
ããã¯ãããžãŒã®æ¬ãèªãã§ããããšèšããšããããŠãã®äººã¯ããã°ã©ãã³ã°ã®æç§æžããèµ·æ¥å®¶ã®èªäŒãæ³åããã
ã§ãããã¯ãããžãŒãšç€ŸäŒã®é¢ä¿ãèããããã®æ¬ã¯ããã®ã©ã¡ãã§ããªããã³ãŒãã®æžãæ¹ã§ããæåè ã®æŠåäŒã§ããªãããæè¡ã瀟äŒãã©ãå€ãã瀟äŒãæè¡ã«ã©ã圢ã¥ããããŠãããããåãçŽãæ¬ã ã
ãã®çš®ã®æ¬ãè±èªåã§ã¯æ¥éã«å¢ããŠãããAIã®ãªã¹ã¯ãè°è«ããæ¿çç«æ¡è ãããã¯ã³ã³ãã³ã»ãŒã·ã§ã³ãåãããšã³ãžãã¢ããããã¯ãéçºã®å€æãè¿«ããããããŒãžã£ãŒââç«å Žã®éããè¶ ããŠèªãŸããŠããäžçŸ€ã®æžç±ããããæ¥æ¬ã§ã¯ãŸã ã»ãšãã©ç޹ä»ãããŠããªããããã¯ãããžãŒã«æºãããã¹ãŠã®äººéã«ãšã£ãŠãäžèªã®äŸ¡å€ãããã
æ¥æ¬ã§AIãæè¡ã«ã€ããŠæžãããæ¬ã®å€ãã¯ããã©ã䜿ãããããè åšãåŠãããšããäºé 察ç«ã§èªããããè±èªåã®ãã®çš®ã®æ¬ã¯ããå°ãè€éãªåããç«ãŠããæè¡ã¯èª°ã®å©çã®ããã«èšèšãããŠããã®ãããã®èšèšã«ã¯èª°ã®å£°ãåæ ãããŠããªãã®ããèªååãé²ããšããæ©æµãåããã®ã¯èª°ã§ãäžå©çãæŒãä»ããããã®ã¯èª°ãããããã£ãåãã«ãããŒã¿ãšãã£ãŒã«ãã¯ãŒã¯ã§åãåãæ¬çŸ€ã ã
ä»åã¯ããã®äžãã5åãéžãã ããå°éå®¶ã§ãªããšèªããªããæ¬ã§ã¯ãªããæè¡ã®æèãæã¡ãªãã瀟äŒãžã®åããæ·±ããæ¬ã ãã©ããè±èªåèã®ã¿ã§æ¥æ¬èªç¿»èš³çã¯æªåè¡ã ããè±èªã¯å¹³æã§ããã¯ãããžãŒã®å®åçµéšãããã°ååã«èªã¿é²ããããã
AIã¯ãªãã䟡å€èгããééããã®ã
The Alignment Problem: Machine Learning and Human ValuesïŒBrian ChristianïŒ
AIãã人éã®æå³éãã«åããªããåé¡ã¯ãåãªããã°ã§ã¯ãªãã人éãæã€äŸ¡å€èŠ³ãæ©æ¢°ã«æ£ç¢ºã«äŒããæ¹æ³ããŸã ååšããªãããšããæ ¹æ¬çãªå°é£ã ãæ¬æžã¯ãã®åé¡ââã¢ã©ã€ã³ã¡ã³ãåé¡ââã«åãçµãç ç©¶è ãã¡ã®çŸå Žã䞹念ã«è¿œã£ãã«ãã«ã¿ãŒãžã¥å Œè§£èª¬æžã ã
èè ã®Brian Christianã¯ããæã人éãããæ©æ¢°ãšãæãæ©æ¢°ããã人éãèŠã€ããããã¥ãŒãªã³ã°ãã¹ãã«å®éã«åå ããçµéšãæã€ã©ã€ã¿ãŒã§ãç§åŠãšå²åŠãšå«çã亀差ããå°ç¹ãé®®ããã«æããæ¬æžã§ã¯åŒ·ååŠç¿ã®çœ ãå ±é ¬èšèšã®é£ããããã§ã¢ãã¹ã®å®çŸ©ãè€æ°ãã£ãŠäºãã«ççŸãããšããçŸå®ããå ·äœçãªã±ãŒã¹ã¹ã¿ãã£ãšãšãã«èªãããã
reward hackingãvalue alignmentãspecification gamingââChatGPTãLLMå šçã®çŸä»£ããããã®æŠå¿µã¯å®åã§ãé »ç¹ã«äœ¿ããããèªåãã¡ãäœãã·ã¹ãã ã瀟äŒã«äœãããã®ããåãç¶ããããšã³ãžãã¢ã«ããŸãæã«åã£ãŠã»ããäžåã ã2020幎åè¡ã§è±èªåã§ã¯é«ãè©äŸ¡ãåããŠããããæ¥æ¬ã§ã®ç¥å床ã¯é©ãã»ã©äœãã
AIã®ãèŠããªãã³ã¹ãããå°å³ã§æã
Atlas of AI: Power, Politics, and the Planetary Costs of Artificial IntelligenceïŒKate CrawfordïŒ
AIã¯ãã¯ã©ãŠãã®äžã«ããç¥æ§ãã§ã¯ãªããã³ã³ãŽã®ã³ãã«ãé±å±±ãã¢ããŸã³å庫ã®ãŠã§ã¢ããŠã¹åŽåè ãèšå€§ãªé»åãæ¶è²»ããããŒã¿ã»ã³ã¿ãŒââAIãåããããã«å°çäžã®ã©ããã§äœããç ç²ã«ãªã£ãŠãããæ¬æžã¯ãã®ãèŠããªãã³ã¹ããããå°çã»è³æºã»åŽåã»ããŒã¿ã»åé¡ã»ææ ã»åœå®¶ãšãã7ã€ã®è»žã§è§£åããã
èè ã®Kate Crawfordã¯Microsoftã®AIå«çç ç©¶è ã§ãããNew York倧åŠã®ææã§ããããç ç©¶è ãšããŠã®å³å¯ããšããã£ãŒã«ãã¯ãŒã¯ã§åŸãçã ããçŸå®æåãçµã¿åããã£ãæ¬æžã¯ãAIãããã¯ãããžãŒã®åé¡ãã ãã§ãªããæ¿æ²»çµæžã®åé¡ããšããŠæãçŽãèŠåº§ãäžããŠãããã
ãèªåãäœããããã¯ãã¯ã©ãããæ¥ãŠããã®ãããåãå§ãããšããã®æ¬ã®åããæ¥ã«åå®ã«ãªããããŒã¿ã»ã³ã¿ãŒã®ç°å¢è² è·ã話é¡ã«ãªãçŸä»£ãæ¬æžã¯çºå£²åœæïŒ2021幎ïŒããä»ã®ã»ãã宿ãæã£ãŠèªããããµã¹ããããªãã£ãAIã¬ããã³ã¹ã«é¢ãã人ã«ã¯ç¹ã«åºããã¯ãã ã
AIãé°ã§æ¯ããã幜éåŽåè ãã®å®æ
Ghost Work: How to Stop Silicon Valley from Building a New Global UnderclassïŒMary L. Gray, Siddharth SuriïŒ
Amazon Mechanical Turkã§ã¯ãªãã¯ä»åãããã人ãã³ã³ãã³ãã¢ãã¬ãŒã¿ãŒãšããŠæŽååç»ãäžæ¥äžå¯©æ»ãã人ãAIã®åŠç¿ããŒã¿ã«ã©ãã«ãä»ãç¶ãã人ââãããã®ä»äºã¯ãAIãèªååããããšèšãããäžæ¹ã§ãå®éã«ã¯äººéãåŠçããŠãããèè ãã¡ã¯ãã®çŸè±¡ãããŽãŒã¹ãã¯ãŒã¯ããšåŒã¶ã
Microsoftã®ãªãµãŒãã£ãŒã§ããèè 2人ã¯ãæ°çŸäººã®ãŽãŒã¹ãã¯ãŒã«ãŒãžã®ã€ã³ã¿ãã¥ãŒãéããŠããã©ãããã©ãŒã çµæžã®è£åŽãèšé²ãããä»äºã®äžå®å®ããå ±é ¬ã®äžéæããçªç¶ã®ã¢ã«ãŠã³ã忢ââã·ãªã³ã³ãã¬ãŒã®æé·ã®èã«é ããåŽåè ã®çŸå®ããçµ±èšãšèšŒèšã§æããã«ãããã
human-in-the-loopãmicro-taskingãcrowd-sourced labelingââAIã®ç²ŸåºŠãäžããããã«æ¬ ãããªãæŠå¿µã ãããã®åããã«äººéã®åŽåãããããšãããšã³ãžãã¢ã¯æèãã«ãããæ¬æžãèªããšãèªåãã¡ãæ¥åžžçã«äœ¿ãAPIã®åããåŽãèŠããããã«ãªããAIéçºã«æºãããšã³ãžãã¢ãšãããã¯ããããŒãžã£ãŒã«ãç¹ã«èªãã§ã»ããã
ã¢ã«ãŽãªãºã ã¯èª°ã®å³æ¹ã
Automating Inequality: How High-Tech Tools Profile, Police, and Punish the PoorïŒVirginia EubanksïŒ
ã¢ã¡ãªã«ã®ã€ã³ãã£ã¢ãå·ã§ã¯ãå»ç絊ä»ã®å¯©æ»ãèªååããã·ã¹ãã ãå°å ¥ãããŠãããæ°åäžäººã絊ä»ãæã¡åããããå€ãã¯æžé¡äžåãæè¡çãšã©ãŒãåå ã ã£ãããç³ãç«ãŠã®æç¶ãã¯è€éã§ã被害ãåããã®ã¯å£°ãäžãã«ããè²§å°å±€ã ã£ããããã¯ç¹æ®ãªã±ãŒã¹ã§ã¯ãªãã
èè ã®Virginia Eubanksã¯æ¿æ²»åŠè ãšããŠãã¢ã¡ãªã«3éœåžã®çŠç¥ã»å»çã»åäºåžæ³ã·ã¹ãã ã«ããããé«åºŠãªèªååããŒã«ãããè²§å°å±€ã«å¯ŸããŠã©ã®ããã«æ©èœããŠãããã調æ»ãããæ¬æžãåãã®ã¯ã·ã³ãã«ã ââããã¯ãããžãŒã¯äžç«ãããšããåãã ã
ãã¯ãããžãŒã¯ãã人éã®å€æã®ãã€ã¢ã¹ãæé€ããŠå®¢èгçãªçµè«ãåºãããã®ãšããŠå°å ¥ãããããšãå€ããã§ããã·ã¹ãã ãèšèšããã®ã¯äººéã§ãåŠç¿ããŒã¿ãäœãã®ã人éã ãéå»ã®æææ±ºå®ã«ãã€ã¢ã¹ããã£ããªãããã®ããŒã¿ã§èšç·Žãããã·ã¹ãã ã¯ãã€ã¢ã¹ãåŒãç¶ãããããŠèªååã®ç®ããã¶ã£ãŠããã¶ãããã·ã¹ãã ããã倿ããããšããŠäººéã®è²¬ä»»ãææ§ã«ãªãã
çµè«ã¯æå¿«ã§ããã€éããããŒã¿ã¯éå»ã®å·®å¥ãåæ ããã¢ã«ãŽãªãºã ã¯ãããèªåçã«åçŸãããpredictive analyticsãrisk scoringãautomated decision-makingââããããèšèšãããšã³ãžãã¢ãæ¬æžãèªããšããæ£ç¢ºãªã·ã¹ãã ãäœãããšããšãå ¬æ£ãªã·ã¹ãã ãäœãããšãã¯å¿ ãããåãã§ã¯ãªããšããçŸå®ããå ·äœçãªäºäŸãšãšãã«è ã«èœã¡ãã2018幎åè¡ã ãå 容ã®åå®ãã¯å¢ãã°ããã ã
ã¢ã«ãŽãªãºã ãã奜ã¿ããåäžåãããšã
Filterworld: How Algorithms Flattened CultureïŒKyle ChaykaïŒ
ãªãäžçäžã®ã«ãã§ã䌌ããããªå è£ ã«ãªã£ãã®ãããªãã¹ããªãŒãã³ã°ã§ããããã鳿¥œã®æ§é ãåæããŠããã®ãããªãInstagramã§äººæ°ã®ããåçã¯åããããªè²èª¿ãšæ§å³ãæã€ã®ããæ¬æžã¯ãã®ãæåã®å質åãã®åå ãã¬ã³ã¡ã³ããŒã·ã§ã³ã¢ã«ãŽãªãºã ã«æ±ããã
èè ã®Kyle Chaykaã¯The New Yorkerã®ã¹ã¿ããã©ã€ã¿ãŒã§ããã¯ãããžãŒãšæåã®äº€å·®ç¹ãé·å¹ŽåæããŠãããæ¬æžïŒ2024幎åè¡ïŒã§ã¯ãSpotifyãNetflixãTikTokãInstagramãšãã£ããã©ãããã©ãŒã ãããŠãŒã¶ãŒã®è¡åããŒã¿ãããšã«ããšã³ã²ãŒãžã¡ã³ãããæé©åãç¶ããçµæã倿§ã ã£ãã¯ãã®æåããã©ããåããŠããéçšã远ãã
engagement optimizationãrecommendation loopãfilter bubbleââãããã¯æ¥åžžçãªæè¡æŠå¿µã ããæ¬æžã¯ãã®å ã«äœãèµ·ããŠããããæåæ¹è©ã®è§åºŠããåããã¢ã«ãŽãªãºã ãèšèšããåŽã®äººéããèªåãã¡ã®éžæãæåã«ã©ããªåœ±é¿ãäžããŠããããèãããã£ãããšããŠã2024å¹Žã®æŽæžã®äžã§ã¯ç¹ã«ç€ºåã«å¯ãäžåã ã
5åã®äžã§ã¯æããããã¯ç³»ãã³ãã£ã¯ã·ã§ã³ãããé ãæäœã§æžãããŠãããThe New Yorkerãããèªã¿ãããããããAIãã·ã¹ãã èšèšã®æ¬ã§ç²ãããšããå¥ã®è§åºŠãããã¯ãããžãŒãšç€ŸäŒã®é¢ä¿ãèããããšãã«æã«åã£ãŠã»ãããæè¡ã®äžã«ãã人éã»ã©ãå€ããæè¡ãçºããèŠç¹ã®äŸ¡å€ããããã
ãã®5åãã©ãèªãã
ãã®5åã«å ±éããã®ã¯ããã¯ãããžãŒããããèªäœã§éããåé¡ããšããŠæ±ããªãããšã ãAIã®ã¢ã©ã€ã³ã¡ã³ãåé¡ã¯å²åŠãšå«çã®åé¡ã§ãããAIã®ç°å¢ã³ã¹ãã¯å°æ¿åŠãšè³æºçµæžã®åé¡ã§ããããŽãŒã¹ãã¯ãŒã¯ã¯åŽåæ³ãšæ Œå·®ã®åé¡ã ãæè¡ã®åé¡ã¯ã瀟äŒã®åé¡ã ã
èªã¿æ¹ã®ãã³ããã²ãšã€ãåç« ã®åé ãæ«å°Ÿã«çœ®ãããã±ãŒã¹ã¹ã¿ãã£ããèªã¿å§ããã®ã广çã ãçè«çãªèª¬æãããå ·äœçãªäºäŸãå ã«æ¥ããšãè±èªã®è§£å床ãäžããããããå®éã®ã€ã³ã·ãã³ãã人ç©ã®èªããæãããã«ãããšãå°éçšèªãåºãŠããŠãæèã§æå³ãæŽãããæè¡çãªè±èªãšã¯ç°ãªãããã®ã«ããŽãªã®æ¬ã¯ç©èªæ§ãé«ããã³ãŒãã®ããã¥ã¡ã³ããèªããšãã®è±èªåãšãããã§æ±ããããè±èªåã¯æ¹åæ§ãéããèªã¿ç©ãšããŠæ¥œããããã©ããã§ãè±èªåèãç¶ããç¿æ £ãå€ãã£ãŠããã
ãŸãããããã®æ¬ã§åºãŠããèªåœââaccountabilityãtransparencyãbiasãfairnessãstakeholderââã¯AIå«çãESGãã³ãŒãã¬ãŒãã¬ããã³ã¹ã®è°è«ã§ãã®ãŸãŸäœ¿ãããè±èªã ãèªæžãéããŠæŠå¿µãšèªåœãåæã«æã«å ¥ãããšãè±èªåã®ã«ã³ãã¡ã¬ã³ã¹ãè«æãèªãé床ãæç¢ºã«äžãããå€è³ç³»äŒæ¥ã®äŒè°ããã°ããŒãã«ããŒã ãšã®è°è«ã§ãããã®èšèã䜿ããããšãããã 翻蚳ã§ããã ãã§ãªããã©ããªæèã§äœ¿ãããèšèããç¥ã£ãŠãããã©ããã§ãè°è«ãžã®åå ã®è³ªããŸãã§å€ããã
æè¡ãšç€ŸäŒã®æ¥ç¹ãè±èªã§èãããã人éã¯ããŸã å€ããªãããã®åžå°æ§ããä»åŸã®ãã£ãªã¢ã§å·®å¥åã«ãªãã
AIã瀟äŒå€é©ãããŒãã«ããæŽæžã«ææŠããã人ã¯ãã¡ãã â ãåããŠã®1åãèªã¿åã£ãïŒããšããèªä¿¡ãããããããŒãžæ°ãå°ãªãã§èªã¿ãããæŽæžãªã¹ã
è±èªã®èªã¿æ¹ãã®ãã®ãå€ããã人ã¯ãã¡ãã â è±èªããèªãããã«å€ããç¬éïœTOEIC 600ç¹å°ããå§ããæŽæžå€èªã®æç§æž
æŽæžå€èªã§è±èªåãã©ãå€ãããç¥ããã人ã¯ãã¡ãã â æŽæžå€èªã§çޝèš100äžèªãèªãã ãäœãå€ãã£ããïœ1幎éã®å šèšé²ãšæ¬é³ã¬ãã¥ãŒ
â»æ¬èšäºã«ã¯Amazonã¢ãœã·ãšã€ããªã³ã¯ãå«ãŸããŠããŸããAmazonã®ã¢ãœã·ãšã€ããšããŠã驿 Œè²©å£²ã«ããåå ¥ãåŸãŠããŸãã
èè ãããã£ãŒã«
åœå å€ã§10å¹Žè¶ ã®ãã£ãªã¢ãæã€ITãšã³ãžãã¢æå¿ã«ããèªæžãŠããããå šå¡ãå€èªãéãTOEIC 950è¶ ãé§åšãçµéšã
ãITÃæé€ãã軞ã«ãæè¡æžãããã³ãã£ã¯ã·ã§ã³ãŸã§ãåã¡ã³ããŒãå°éèŠç¹ã§å³éžããæŽæžãã¬ãã¥ãŒããŸãããšã³ãžãã¢ãè±èªãæŠåšã«ããããã®ç¥èŠãå ±åçºä¿¡äžã