微软：2023年生成式人工智能心理健康和安全技术（英文版）（76页）.pdf

1、Microsoft confidentialGenerative AI,Mental health,and Safety TechnologiesSeptember 2023Microsoft confidential|2 STUDY METHODOLOGYWeb survey,Teens 13-17,Parents of 617,Adults 18-6417 countries,N=16,795Timing:July 27-August 22,2023Microsoft confidential|3 TOPICS COVERED IN WAVE 8Microsoft confidential

2、|4 THEME 1:GOOD CIVILITY RATINGS HELD STEADY The decline in good online civility ratings stopped this past year Bad online civility ratings only increase 1-pt.since last year although still slightly higher than good ratings The gap between good and bad civility ratings varied considerably by geograp

3、hySingapore and India showed the largest positive gaps at 46%and 34%respectivelyFrance and Mexico were on the other end of the spectrum with bad ratings outpacing good-43%and-28%respectively46%34%25%17%4%-2%-3%-8%-12%-13%-15%-16%-22%-24%-26%-28%-43%SingaporeIndiaAustraliaCanadaUnited KingdomS.Africa

4、DenmarkUnited StatesCzechiaItalyGermanyS.KoreaBrazilSpainColombiaMexicoFranceGap between good and bad civility ratings38%34%31%31%22%27%32%33%0%5%10%15%20%25%30%35%40%45%50%2020202120222023Change in online civility*2022-2023Microsoft confidential|5 THEME 2:THE RISK LANDSCAPE WAS SLIGHTLY IMPROVED IN

5、 2023Any risk experienced fell 2-pts.from 69%to 67%.The drop was 3-pts.among countries that were in both Wave 7&8Driving risks lower was a drop in Sexual Risks(-4-pts.)with each component of the category contributing to the fallPersonal and violent content risks were reported more often by parents o

6、f younger children(6-12)vs.parents of 13-17Exposure to risks and consequences peaked at age 18-19Respondents having a college degree or higher were slightly more likely to experience Personal,violent content or sexual risksMicrosoft confidential|6 THEME 3:YOUNG ADULTS 18-24 LED THE ADOPTION OF GENER