
å°éå®¶ã¯AIçææç« ãèŠæãããïŒ
ð æç®æ å ± ãš æé²åèš³
å€§èŠæš¡èšèªã¢ãã«æä»£ã®ç§åŠçæç« ïŒAIãšäººéãäœæããã³ã³ãã³ãã®èšç®åæ
ðKhera, Rohan, et al. "Scientific Writing in the Era of Large Language Models: A Computational Analysis of AI-Versus Human-Created Content." Stroke 56.10 (2025): 3078-3083. https://doi.org/10.1161/STROKEAHA.125.051913
ð DOI, PubMed, Google Scholar ðµããããã£ã¹ã >>> Not applicable
[èæ¯ã»ç®ç] å€§èŠæš¡èšèªã¢ãã«ïŒLLMïŒã¯ã人éã®å°éå®¶ã®ãããªã³ã³ãã³ããçæããç§åŠæç®ã®åæãå éããããã«äœ¿çšã§ãã人工ç¥èœïŒAIïŒããŒã«ã§ãããã誀解ãæãã³ã³ãã³ããçæããããšã«ãã£ãŠèª€ã£ãæ å ±ãåºããå¯èœæ§ããããæ¬ç ç©¶ã§ã¯ãAIãçæããç§åŠããã¹ããšäººéãäœæããç§åŠããã¹ããåºå¥ããããã®èšèªçç¹åŸŽãæããã«ãããã®ã¿ã¹ã¯ã«å¯ŸããAIæ€åºããŒã«ã®æ§èœãè©äŸ¡ããããšãç®çãšããã
[æ¹æ³] è³è¡ç®¡ã®ãããã¯ã«é¢ãã34ã®æç« ïŒå€§èŠæš¡èšèªã¢ãã«[Generative Pre-trained Transformer 4ãGenerative Pre-trained Transformer 3.5ãLlama-2ãBard]ã«ãã£ãŠçæããã12æ¬ãšã人éã®ç§åŠè ã«ãã22æ¬ïŒã®èšç®åæãè¡ã£ããåå°è«æã¯ãStrokeç·šéå§å¡äŒã®æå€§38人ã®ã¡ã³ããŒã«ãã£ãŠãAIãäœæãããã®ããŸãã¯äººéãå·çãããã®ãšããŠè©äŸ¡ããããå°éå®¶ã®éå£çããã©ãŒãã³ã¹ãšãåºã䜿çšãããŠãããªã³ã©ã€ã³AIæ€åºããŒã«ã§ããGPTZeroãæ¯èŒãããAIãäœæããã³ã³ãã³ããšäººéãäœæããã³ã³ãã³ãã®èšèªçå·®ç°ãç¹åŸŽä»ããããã«ãæ§æïŒåèªæ°ãè€éããªã©ïŒãæå³ïŒæ¥µæ§ïŒãèªã¿ãããïŒãã¬ãã·ã¥ã¹ã³ã¢ïŒãã°ã¬ãŒãã¬ãã«ïŒãã¬ãã·ã¥ã»ãã³ã±ã€ãïŒãèšèªã®åœææ§ïŒãŸãã¯äºæž¬å¯èœæ§ïŒã«ãããèšèªçç¹åŸŽãæœåºããæ¯èŒããã
[çµæ]
â å°éå®¶ã¯AIçææç« ãèŠæãããïŒ

ã»AIçææç« ãAIçæãšå€æïŒN=10, 83.3%ïŒïŒããšããšAIçæã®å°è«ããå°éå®¶ãæ£ããAIãšå€å®ããã±ãŒã¹ïŒçéœæ§ïŒTPïŒãAIèŠæãã®æåºŠã瀺ãã
ã»AIçææç« ã人éå·çãšå€æïŒN=2, 16.7%ïŒïŒAIçæãªã®ã«äººéãšå€å®ããã±ãŒã¹ïŒåœé°æ§ïŒFNïŒãAIãå°éå®¶ãããæããäŸã§ããã
ã»äººéå·çæç« ãAIçæãšå€æïŒN=7, 31.8%ïŒïŒäººéå·çãªã®ã«AIãšå€å®ããã±ãŒã¹ïŒåœéœæ§ïŒFPïŒãäººéææžãAIãšèª€èŠå ±ããäŸã§ã鿀åºã®åŸåã瀺ãã
ã»äººéå·çæç« ã人éå·çãšå€æïŒN=15, 68.2%ïŒïŒäººéå·çãæ£ãã人éãšå€å®ããã±ãŒã¹ïŒçé°æ§ïŒTNïŒãAIæ€ç¥ã®ç¹ç°åºŠã«çžåœããã
âãã®é åããå°éå®¶ããã«ã®æŠèŠæçžŸã¯ãæåºŠ=10/12=83.3%ãç¹ç°åºŠ=15/22=68.2%ãå šäœæ£è§£ç=25/34â73.5%ãšãªããè§£éãšããŠãå°éå®¶ã¯AIãèŠæãåã¯é«ãã ããäººéææžãAIãšèª€ãçïŒ31.8%ïŒãç®ç«ã€ããšãã嫿ã«ãªãã
GPTZeroã¯ãAIãäœæãã12æ¬ïŒ100ïŒ ïŒãšäººéãæžãã21æ¬ïŒ95.5ïŒ ïŒãæ£ç¢ºã«åé¡ããããããããã®ããŒã«ã¯ãåé¡ã®ããã«æ°åã®éèŠãªæç« ã«ã®ã¿äŸåããã人éã®ãšãã»ã€ãšæ¯èŒããŠãAIãçæããã³ã³ãã³ãã¯ãåèªæ°ãšè€éããäœããææã«äœãperplexityïŒäžå€®å€ã15.0察7.2ïŒP<0.001ïŒãäœãå¯èªæ§ã¹ã³ã¢ïŒFleschäžå€®å€ã42.1察26.4ïŒP<0.001ïŒãããã³é«ãã°ã¬ãŒãã¬ãã«ïŒFlesch-Kincaidäžå€®å€ã13.1察14.8ïŒP=0.006ïŒã瀺ããã
[çµè«] å€§èŠæš¡ãªèšèªã¢ãã«ã¯ã人éãæžããããã¹ããšæ¯èŒããŠæž¬å®å¯èœãªå·®ç°ãæã€ç§åŠçã³ã³ãã³ããçæãããã人éã®å°éå®¶ã§ãäžè²«ããŠèå¥ã§ããªãç¹åŸŽã衚ããŠãããè€éãªAIæ€åºããŒã«ãå¿ èŠãšãããå°éå®¶ãAIãšäººéã®ã³ã³ãã³ããåºå¥ããéã«çŽé¢ãã課é¡ãèæ ®ãããšãç§åŠçã³ãã¥ãã±ãŒã·ã§ã³ã®å®å šæ§ãå®ãããã«äººéã®èšŒæãäžå¯æ¬ ã§ããå Žåã«ã¯ãæè¡æ¯æŽããŒã«ãäžå¯æ¬ ã§ããã
ð± So What?ïŒäœãé¢çœããšæãããïŒ
æè¿ãNature Human Behaviourèªã«æ²èŒãããè«æãæèªããã
ãã®è«æã¯ãåŠè¡è«æäžã®çæAI䜿çšçã調æ»ããAIçæããã¹ããæ¥å¢ããŠããããšãæããã«ããã
æè¿ã®çæAIã®é²æ©ã¯ãæ¬åœã«åãŸããã
ã¡ãã£ãšããæç« ã®æ ¡æ£ãããäžæãããIntroductionå
šäœãŸã§æ§ç¯ããŠããŸãç Žå£åãæã£ãŠãããšæãã
ãããªAIçææç« ã ããæãããŠå°éå®¶ã®ç®ã¯ã©ã®çšåºŠèŠæãããã®ãªã®ã ãããã
ãããªçŽ æŽãªçåãä»åæèªããç ç©¶ã¯æããã«ããŠãããã
çµè«ãšããŠã¯ãå°éå®¶ã¯AIãèŠæãåã¯é«ãã ããäººéææžãAIãšèª€ãçïŒ31.8%ïŒãç®ç«ã£ãã
ä»ã®ãšããã¯ãŸã ãå°éå®¶ã®ç®ã¯ãŸããŸãAIçææç« ãèŠæããããã ã
ã ãããããé
åã®åé¡ããããšæãã
äŸãã°ãä»åã®ç ç©¶ã«ãããŠã¯ãããããïŒFull textèŠããªãã®ã§æšæž¬ã ãïŒçœãé»ãã®äžçã ããã
ã€ãŸããå
šæäººéãå·çããæç« ããå
šæAIãçæããæç« ãããšããäžçã
ã ããäžã®äžã«ãããçŽãªçœãé»ãå°ãªãããã«ãçæAIã®äœ¿çšã«ãé
åéãå€ãããç°è²ããååšãããã
äŸãã°ãIntroductionã«ãããŠé¢é£è«æAã«é¢ããèšèŒã®äžæã ããAIçæããŠãããããããå
šäœãšããŠã¯äººéãå·çããæç« ã®äžã«éåæã®ãªãããã«çµã¿èŸŒãã
æãããŠãããã¯AIçæã«åºåãããã®ã ãããããããšã人éå·çã«åºåãããã®ã ãããã
ã©ããŸã§ããããã¯ã©ã®ããã«çæAIã䜿çšããããããã¯AIçææç« ã«åºåãããã®ã ãããã
ã»ã»ã»ããšãŠãé£ããç·åŒãã§ããã
ããã蟺ã¯ã©ããŸã§ãèããŠããããã
â¬ïž é¢é£ note & ðã§ã®æçš¿âš
ðå°éå®¶ã¯AIçææç« ãèŠæãããïŒ
â çåŠçæ³å£«_海接éœäž Ph.D. (@copellist) October 7, 2025
ã»è³è¡ç®¡ãããã¯ã«é¢ããæç«
ãâ22æ¬ã¯äººéå·ç, 12æ¬ã¯AIçæ
ã»å°éå®¶(éèªç·šéè )38åãå·çè ãå€å®
ð¹AIçææç« ïŒæ£è§£83.3%, äžæ£è§£16.7%
ð¹äººéå·çæç« ïŒæ£è§£68.2%, äžæ£è§£31.8%
AIæç« ã¯çµæ§èŠæããã
äžæ¹, 人éæç« ãAIãšèª€ãçãããé«ãã§ããð² pic.twitter.com/vY3skeS98I
âââââââââââââã»ã»ã»â¥ ⥠⥠â¥
è¯è³ªãªãªãå»åŠé¢é£ã»è±è«ææèªãã¢ãªïŒARIã
ãã¡ãããâª
âââ

⥠⥠⥠â¥ã»ã»ã»ââââââââââââ
#ïžâ£ #çåŠçæ³ #èšåºç ç©¶ #ç ç©¶ #ãªãããªããŒã·ã§ã³ #è±è«æ #æç®æèª #è±ææèª #ãšããã³ã¹ #ãµã€ãšã³ã¹ #æ¯æ¥æŽæ° #æè¿ã®åŠã³