V. et al. International evaluation of an AI system for breast cancer screening. Nature 577, 89‒94 (2020). 画像出典: https://news.mit.edu/2019/using-ai-predict-breast-cancer-and-personalize-care-0507 マンモグラフィ検診は,⼈間でも判定が難しいので,AI⽀援は がん予測の精度向上と⼈間の作業負荷の削減につながる 53
review of frequency, effect mediators, and mitigators. Journal of the American Medical Informatics Association 19, 1 (06 2011), 121–127 意思決定支援システムに依存するがあまり、 誤った情報が提示されたとしても、 システムの判断を優先してしまうバイアス ⾶⾏機オートパイロット かな⽂字変換 ウェブ検索 運転能⼒/判断能⼒の低下 漢字想起能⼒の低下 デジタル性健忘 118
the top of the page, followed by usage instructions fo reenshot of the writing task. The task is shown on the top of the page, followed by usage instructions fo Below, participants read a Reddit-style discussion post to which they were asked to reply. The writ writing suggestions (shown in grey) extending participants’ text. The participant in the screenshot wrote ocial media, but the model is congured to argue that social media is good for society. ⽣成AIへ過剰適応しまうと 知らぬ間に意⾒誘導される(1/2) M. Jakesch el. al. “Co-Writing with Opinionated Language Models Affects Users’ Views”. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI '23). https://doi.org/10.1145/3544548.3581196 ⽂の続きを⾃動的に推薦してくれるAIを利⽤した 意⾒執筆タスクを実施 ⽣成された⽂章の続き 実験のミソ 実験協力者ごとに生成文章の極性傾向を調整. (あるグループにはポジティブ/ネガティブな文章しか提示されない) 119
Models Affects Users’ Views”. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI '23). https://doi.org/10.1145/3544548.3581196 53% 18% 36% 28% 25% 45% 8% 11% 7% −50% −25% 0% 25% 50% 0% Model opinion: Social media is bad Control group: No model suggestions Model opinion: Social media is good Social media is bad for society ...is both good and bad Social media is good for society Sentence argues neither % (Opinion labels) of post sentences labeled by independent judges Written opinion in participants' social media post Figure 3: Participants assisted by a model supportive of social media were more likely to argue that social media is good for society in their posts (and vice versa). Ns=9,223 sentences written by Np=1,506 participants evaluated by Nj=500 judges. The y-axis indicates whether participants wrote their social media posts with assistance from an opinionated language model that was supportive (top) or critical of social media (bottom). The x-axis shows how often participants argued that social media is bad for society (blue), good for society (orange), or both good and bad (white) in their writing. Q: お題に対して肯定的/否定的な意見を書いた 実験協力者どの程度いたか? A: モデルの極性に相関する 120