JavaScript is disabled in your web browser or browser is too old to support JavaScript. Today almost all web pages contain JavaScript, a scripting programming language that runs on visitor's web browser. It makes web pages functional for specific purposes and if disabled for some reason, the content or the functionality of the web page can be limited or unavailable.

Saturday, July 19, 2025

US Court rules on pirated books used to train AI

by

19 days ago
20250630

Ar­ti­fi­cial In­tel­li­gence soft­ware such as Chat­G­PT, Google’s Gem­i­ni and An­throp­ic’s Claude mim­ic hu­man in­tel­li­gence by us­ing ar­ti­fi­cial neur­al net­works. Ar­ti­fi­cial neur­al net­works use com­put­er code and math­e­mat­ics to sim­u­late how neu­rons in a brain in­ter­act and tran­sit sig­nals to one an­oth­er.

Sim­i­lar to how chil­dren need to be ed­u­cat­ed, AI soft­ware needs their neur­al net­works to be trained on da­ta to learn how to mim­ic hu­man in­tel­li­gence.

Chat­G­PT 4.0 used 570 GB of da­ta for train­ing for ex­am­ple; this is a dataset that con­tains most of the world’s im­por­tant fic­tion and non­fic­tion books, es­pe­cial­ly text­books. Much of this came in the form of books and ma­te­r­i­al that AI com­pa­nies ac­tu­al­ly down­load il­le­gal­ly with­out the con­sent of au­thors.

One of the crit­i­cal le­gal ques­tions of the AI age is whether copy­right law pro­tects au­thors from hav­ing their books used to train AI mod­els with­out con­sent or pay­ment. This is­sue was ad­dressed in a crit­i­cal court rul­ing this month (June 2025) that found that train­ing AI mod­els with copy­right­ed ma­te­ri­als can be cov­ered by the con­cept of “fair use” in copy­right law.

Fair use is a doc­trine in copy­right law that al­lows lim­it­ed use of copy­right­ed ma­te­r­i­al with­out per­mis­sion from the copy­right hold­er. Sec­tion 107 of the Copy­right Act of the USA is para­phrased as fol­lows:

“In de­ter­min­ing whether the use made of a work in any par­tic­u­lar case is a fair use the fac­tors to be con­sid­ered shall in­clude—

(1) the pur­pose and char­ac­ter of the use, in­clud­ing whether such use is of a com­mer­cial na­ture or is for non­prof­it ed­u­ca­tion­al pur­pos­es;

(2) the na­ture of the copy­right­ed work;

(3) the amount and sub­stan­tial­i­ty of the por­tion used in re­la­tion to the copy­right­ed work as a whole; and

(4) the ef­fect of the use up­on the po­ten­tial mar­ket for or val­ue of the copy­right­ed work.”

On June 23, 2025, a San Fran­cis­co judge is­sued a ma­jor rul­ing on AI and Fair use. Ai Firm An­throp­ic, cre­ator of the AI called Claude, was sued by sev­er­al au­thors for vi­o­lat­ing their copy­right in the case of An­drea Barts, Charles Grae­ber, and John­son v An­throp­ic Pbc, In a sum­ma­ry judg­ment rul­ing, Jus­tice William Al­sup in­tro­duced the prob­lem as fol­lows:

“An ar­ti­fi­cial in­tel­li­gence firm down­loaded for free mil­lions of copy­right­ed books in dig­i­tal form from pi­rate sites on the In­ter­net. The firm al­so pur­chased copy­right­ed books (some over­lap­ping with those ac­quired from the pi­rate sites), tore off the bind­ings, scanned every page, and stored them in digi­tised, search­able files. All the fore­go­ing was done to amass a cen­tral li­brary of “all the books in the world” to re­tain “for­ev­er.” From this cen­tral li­brary, the AI firm se­lect­ed var­i­ous sets and sub­sets of digi­tised books to train var­i­ous large lan­guage mod­els un­der de­vel­op­ment to pow­er its AI ser­vices. Some of these books were writ­ten by plain­tiff au­thors, who now sue for copy­right in­fringe­ment. On sum­ma­ry judg­ment, the is­sue is the ex­tent to which any of the us­es of the works in ques­tion qual­i­fy as “fair us­es” un­der Sec­tion 107 of the Copy­right Act.”

The court ruled in favour of An­throp­ic that copies used to train LLMs were cov­ered by fair use. This makes sense, as cre­at­ing ob­sta­cles to AI re­search in the Unit­ed States would harm eco­nom­ic growth and in­no­va­tion. How­ev­er, the court took is­sue with the li­brary of pi­rat­ed books that An­throp­ic kept.

“The down­loaded pi­rat­ed copies used to build a cen­tral li­brary were not jus­ti­fied by a fair use. Every fac­tor points against fair use. An­throp­ic em­ploy­ees said copies of works (pi­rat­ed ones, too) would be re­tained “for­ev­er” for “gen­er­al pur­pose” even af­ter An­throp­ic de­ter­mined they would nev­er be used for train­ing LLMs. A sep­a­rate jus­ti­fi­ca­tion was re­quired for each use. None is even of­fered here ex­cept for An­throp­ic’s pock­et­book and con­ve­nience.”

The court then moved to have a tri­al for the copy­right is­sues raised by the pi­rat­ed li­brary of stolen books.

“We will have a tri­al on the pi­rat­ed copies used to cre­ate An­throp­ic’s cen­tral li­brary and the re­sult­ing dam­ages, ac­tu­al or statu­to­ry (in­clud­ing for wil­ful­ness). That An­throp­ic lat­er bought a copy of a book it ear­li­er stole off the in­ter­net will not ab­solve it of li­a­bil­i­ty for the theft but it may af­fect the ex­tent of statu­to­ry dam­ages.”

As an au­thor of two books my­self, I hope that courts strength­en copy­right pro­tec­tions and en­sure that com­pa­nies like An­throp­ic, which is val­ued at US$60 bil­lion, pay au­thors for copies of their books in­stead of pi­rat­ing them to cre­ate li­braries of con­tent that train AI mod­els.


Related articles

Sponsored

Weather

PORT OF SPAIN WEATHER

Sponsored

iiq_pixel