scour or "scrape" online networks, or in some cases, are provided by the owners of the networks themselves. Direct informed consent is rarely, if ever, sought for such uses. Instead, automated software routinely extracts an individual's visual likeness from these photos, translates it into data (often in the form of one's facial geometry, sometimes called a "vector" or "faceprint'), and then uses that data to train FR machinelearning systems.'¢ Yet mass appropriation techniques were not always the default approach. In the early days of FR development, most images for datasets were compiled either through photoshoots, with full informed consent of participants, or through acquisition of government-provided "mugshots."!7 This was the case even in the nascent stages of FR system development in the 1960s; though these early versions of FR were not automated or nearly as powerful as today's technology-they still depended on the appropriation of individual images and likenesses. Despite this consistent need for images, the process for gathering these images was entirely different. For example, when creating one of the first foundational FR systems in 1968, Woody Bledsoe, a pioneer in artificial intelligence, used photos of men who agreed to have their photos taken by him and his research team.!* The same was true for the FERET program, another early 383, 389 (1960);Fairfield v. Am. Photocopy Equip. Co., 291 P.2d 194, 197 (Cal. Ct. App. 1955) (explaining that "[o]ne concept of the right of privacy is the right of a person to be free from unauthorized and unwarranted publicity'); see also SARAH E. IGO, THE KNOWN CITIZEN 34 (2018) (noting the ROP emerged as a response to "the potent trinity of press, photography, and publicity in the late nineteenth century'). [Vol. 88:4 public eye, but who were then later exposed via new media.?! Yet even from these origins, the ROP was more than a "right 'to be let Finally, there is the output of the FR system. When a "match" is made, FR systems will typically output an official image and identification of the likely subject. The display of these images and associated information not only relies on the uses identified above but also constitutes its own independent use as an output for consumption by the FR system customer.
Customers purchase FR systems for these outputs, so without them, the entire commercial value of the system would be dramatically diminished.*' Thus, such outputs constitute a third violation of individuals' ROP. . This would also address some of the concerns about the ongoing harm of FR systems, as each "search" would constitute an additional ROP violation, whether or not the search produced a particular individual's image as the result, given that any target image would need to be checked against all known (and appropriated) identities to ensure the "accuracy" that companies like Clearview AI promise to their customers.
87 Oddly, some courts have suggested that the ROP does not apply to uses of likenesses where the likeness itself is or is a part of the product, rather than an advertisement for a product. See Brooks v.
The commercial advantages of mass image appropriation are also influenced by the network effects of the image database in which they reside. Companies like Clearview have made explicit their intention to have multiple photos of every single person in the world in their database in order to improve the accuracy of their system. Much as having a complete collection of images and identities for sports stars in trading card sets or as avatars in videogame rosters adds to the holistic commercial value of those products, the addition of any single photo or face to an FR system adds value in terms of alleged accuracy, , 815 F. Supp. 2d 1108, 1113(C.D. Cal. 2011), aff'd, 489 F. App'x 155 (9th Cir. 2012) (citing Newton v. Thomason, 22 F.3d 1455, 1461(9th Cir. 1994)).
103 Td. at 1114. 104 See Cohen v. Facebook, Inc., 798 F. Supp.2d 1090, 1096(N.D. Cal. 2011) (holding that users did not consent to the use of their identity for commercial purposes using Facebook's "Friend Finder" service because the terms of service were too ambiguous to find specific consent). where in order to send invitation emails to their users' contacts, LinkedIn gathered explicit consent to collect user's Gmail contacts, showed users matches between their contacts and all LinkedIn users, and then asked users to "invite some people" 109 Jd. at 1215 (finding that LinkedIn's disclosures were "clear enough to alert
[p]laintiffs" of the ensuing use of their names, and that a reasonable person who saw such disclosures and continued was consenting to the emails).
110 See Mark A. Lemley, Terms of Use, 91 MINN. L. REV. 459, 463 (2006).
111 See No Doubt v. Activision Publ'g, Inc., 122 Cal. Rptr. 3d 397 (Ct. App. 2011).
3 Special thanks to Jessica Silbey for this insight and connection.
14 Moore v. Regents of Univ. of Cal., 793 P.2d 479, 480-81 (Cal. 1990). 20 See Waits v. Frito-Lay, Inc., 978 F.2d 1093, 1103(1992) 142 Id. at 574; see also ROTHMAN,supra note 11, others could broadcast an entire play, boxing match, song, or symphony without permission" under the pretext of news reporting.'43 One justice went so far as to state at oral argument that the issue was "reproduction" rather than "reporting."!4
While plays, songs, and symphonies would also be covered by uncopyrightable "hot news" information by wire services using new transmission technology to be illegal); Shyamkrishna Balganesh, "Hot News": The Enduring Myth of Property in News, 111 COLUM. L. REV. 419, 442 (2011). 47 Pora v. Frontline Video, Inc., 18 Cal. Rptr. 2d 790, 792 (Ct. App. 1993) (holding that the makers of a state history documentary featuring a noncelebrity , Inc., 989 F.2d 1512, 1513(9th Cir. 1993) (denial of petition for rehearing and rejection of suggestion for rehearing en banc) (Kozinski, J., dissenting) ("Creativity is impossible without a rich public domain. Nothing today, likely nothing since we tamed fire, is genuinely new: Culture, like science and technology, grows by accretion, each new creator building on the works of those who came before.
Overprotection stifles the very creative forces it's supposed to nurture.").
149 Haelan Lab'ys, Inc. v. Topps Chewing Gum, Inc., 202 F.2d 866, 868 (2d Cir. 1953).
150 Gionfriddo v. Major League Baseball, 114 Cal. Rptr. 2d 307, 314 (Ct. App. 2001).