人工智能、反垄断和隐私（英）

flawed policies. 2INTRODUCTIONHow will generative artificial intelligence (hereinafter AI) likely impact your life in 2040?1Whenasked this question in 2023, most Americans and AI experts were negative on many parameters(with the AI experts especially negative):•79% of the AI experts expected AI to harm personal privacy (which was higherthan the two-thirds of polled Americans who expressed that);•54% of AI experts expected AI to harm basic human rights (versus 41% of the•67% of AI experts expected AI to harm politics and elections (versus 51% of the•52% of AI experts expected AI to worsen the level of civility in society (versus40% of the polled Americans).2Notably absent from that survey was AI’s impact on competition.In parallel with the debate over AI’s broader effects, antitrust scholars and enforcers are debatinghow the emerging AI foundation model supply chain may evolve and whether the technology mayentrench the market power of a few firms. Competition authorities are concerned about theincreasing concentration in this emerging supply chain. In particular, the digital economy hasseveral factors and characteristics that can lead to concentrated markets.Are there similar factorsin the emerging AI foundation model supply chain that will lead to“winner-take-most-or-all”?Could AI herald new business models and innovations that disrupt the dominant ecosystems ofGoogle, Apple, Meta, Amazon, and Microsoft (GAMAMor data-opoliesfor short)? Or will theseecosystems also dominate key segments of the AI foundation model supply chain?Competitionauthorities naturally seek to promote competition in this supply chain.1As used herein, generative artificial intelligence means machine learning models thatleverage deep neural networks to emulate human intelligence (i.e. by imitating informationprocessing of neurons in the human brain) by being exposed to data (training) and finding patternsthat are then used to process previously unseen data. This allowsthe model to generalise based onprobabilistic inference (i.e., informed guesses) rather than causal understanding. Unlike humans,who learn from only a few examples, deep neural networks need hundreds of thousands, millions,or even billions, meaning thatmachine learning requires vast quantities of data.OECD, AI, Data Governance & Privacy: Synergies and Areas of International Co-Operation, OECD ArtificialIntelligencePapers No.22,at 18(June 2024),https://www.oecd.org/en/publications/ai-data-governance-and-2Elon University,The Impact ofArtificial Intelligenceby 2040:National public opinion poll findings(Feb. 2024),https://imaginingthedigitalfuture.org/wp-content/uploads/2024/02/AI2040-Report-public-opinion-poll-white-paper-1.pdf;Lee Rainie&Janna Anderson,A New Age of Enlightenment? A New Threat to Humanity?, Elon Universityhttps://imaginingthedigitalfuture.org/wp-content/uploads/2024/02/AI2040-FINAL-White-Paper-2- polled Americans);polled Americans); andprivacy_2476b1a4-en.html.(Feb.2024),2.29.24.pdf. 3The twoongoingdebates over AI (increased competition and protecting privacy) are largely siloedfrom each other. Competition officials focus on AI's impact on competition without consideringAI's broader implications (which theydeem beyond antitrust’s scope). Those debating AI's widerimplications for privacy, human rights, elections, and civility are not considering the role ofcompetition. Missing frombothdebatesis how exactly will AI affect the relationship betweencompetition and privacy. Specifically, how will AI impact profiling and behavioral advertising?Will more competition in the AI supply chain improve (or harm) privacy, autonomy, and well-being? What are the broader implications on democracy, social discourse, and civility whencompetition among the foundation models increases? Understanding this relationship betweencompetition and privacy is important for several reasons.First, neither privacy nor competition concerns can be addressed independently; privacy andcompetition policies must work in tandem. The conundrum is this: In many digital markets wherethe product or service is ostensibly free (think video streaming, Internet search engines, maps,social networks), privacy can be a critical non-price parameter of competition. However, manydigitalfirms often fail to provide the privacy protections that individualsdesire.If themarketfailure is due to a lack of meaningful competition, such as a dominant firm exercising its marketpower by eroding privacy protections, then in that case, current or more advanced antitrust toolsmay address the problem. However, if the market failure is due to misaligned incentives (i.e.,where firms collect personal dataaboutus but notfor ourbenefit), thenmorecompetition will notfixthe issue. Instead, injecting more competition in the AI supply chain can worsen privacy,autonomy, well-being, and democracy.Second, competition can often enhance privacy when the incentives of market participants alignwiththose of individuals.This alignment,however,d