定位与保护：人工智能的法律现状.pdf

1、1POSITION&PROTECT:THE LEGAL STATE OF PLAY IN AIAI in Finance SummitApril 19,2024THE IMF PREDICTS:“FINANCIAL INSTITUTIONS ARE FORECAST TO DOUBLE THEIR SPENDING ON AI BY 2027”2AGENDA 2Protecting AI3Regulatory Considerations4Best Practices1Training AI3TRAINING AI14 Sometimes called data extraction or a

2、ggregation,web scraping is the practice of collecting vast amounts of online data by automated means.Web scraping(or purchasing scraped data from a vendor)is one common method of obtaining training data.TRAINING AI:WEB SCRAPING 56Is web scraping legal?CFAA imposes both civil and criminal liability f

3、or computer hacking.CFAA prohibits accessing a computer without authorization.Scraping data behind a login screen or paywall could be a CFAA violation if the scraper is not authorized to access the data.CFAA does not apply to publicly-available data.“Giving companies like LinkedIn free rein to decid

4、e,on any basis,who can collect and use datadata that the companies do not own,that they otherwise make publicly available to viewers,and that the companies themselves collect and userisks the possible creation of information monopolies that would disserve the public interest.”hiQ v.LinkedInTRAINING

5、AI:SCRAPING&THE COMPUTER FRAUD AND ABUSE ACT(CFAA)7 Many websites Terms of Service(TOS),on their face,prohibit data scraping.Many websites TOS,on their face,also prohibit commercial use of data.TOS may constitute a binding and enforceable contract.TRAINING AI:SCRAPING AND BREACH OF CONTRACT 8Backgro

6、und:Bright Data sells data scraped from Metas Facebook and Instagram platforms.Data is scraped in a“logged-off”state no account logged in at the time of collection.Meta sued,claiming this scraping violates Metas TOS.In January 2024,the California Federal Court ruled that the TOS did not prohibit“log