adjudication by algorithm

Monday’s issue of The Globe and Mail describes new initiatives to secure better returns for the music industry when musical content is used via radio or internet. Under a joint initiative between the University of Toronto and The Society of Composers, Authors and Music Publishers of Canada (SOCAN), students are investigating how technology “… can parse through audio and video to find media using SOCAN member songs that should be paying royalties to creators and publishers.”

If a reader parses that sentence, the word “should” stands out. Merely using a SOCAN member’s song, or anyone’s song, does not automatically indicate that payment is required. While it is plausible that artificial intelligence can develop a capacity to engage in the contextual analysis required to determine whether a use is legitimate or an infringement, much will depend on the human input.

(As I write this, I recall undergraduate days and a computer science professor who was fond of saying, “garbage in, garbage out.”)

Music collectives invest in AI to better scan online content, finding songs to license them “whether they’re in the background of a cat video or an amateur singer covering Drake.” Surprisingly, no mention of fair dealing or lawful user-generated content. https://t.co/JsTsg2azJ0

— Carys Craig (@CraigCarys) January 2, 2018

In her remarks about the article Carys Craig draws on the work of Niva Elkin-Koren, who has written at length about the perils of copyright adjudication by algorithm. For instance, in Fair Use by Design (2017), Elkin-Koren argues that: “… for fair use to serve its role in the twenty-first century, the checks that it intends to create on the rights of authors must also be embedded in the design of online systems.” She reveals some disturbing findings following analysis of 10,000 removal requests sent to Google, to the conclusion that “an algorithmic regime, which is neither overseen by the public nor by any judicial entity, is extremely vulnerable to misuse.”

Misuse may be deliberate, but misuse also occurs through confusion with respect to the very nature of copyright. Too many people believe that copyright means an absolute right of control; which it never has been, nor has it ever functioned in this manner. From its implementation into statutory law (1710), copyright has been structured as a set of limited rights. But despite this 300+ year ancestry, contemporary articles rarely provide any explanation of where control begins or where control ends.

That story is told through the Copyright Act. Section 3.1 states:

For the purposes of this Act, copyright, in relation to a work, means the sole right to produce or reproduce the work or any substantial part thereof in any material form whatever, to perform the work or any substantial part thereof in public …

From 3.1 we see that copyright exists only when a substantial amount of work is being reproduced. Any algorithm that deems infringement by only identifying use, has vastly overstepped its bounds. Copyright may not even have arisen, let alone finding infringement. (For more about substantial/insubstantial, see here and here.)

If a substantial reproduction has occurred, copyright owners (which may include the writers, musicians, artists, etc. that created the work) are entitled to control the use of the work, through the measures enumerated in the Copyright Act. But that control is not absolute. It is limited, not only by time (Canada maintains the life+50 copyright duration mandated by international treaty) but also by many statutory exceptions. That list begins with fair dealing:

Section 29, fair dealing “Fair dealing for the purpose of research, private study, education, parody or satire does not infringe copyright.”

Sections 29.1 and 29.2 – which provide fair dealing for “criticism or review” and “news reporting” under conditions of attribution. Writers and publishers (perhaps those associated to national newspapers) might appreciate this exception.

(Over the last fifteen years Canada’s treatment of fair dealing has evolved into a measured, progressive exception and ensures that the system of copyright remains balanced and does not devolve purely into a means of rent-seeking. For instance, see here, here, and here.)

Canada’s jewel in the crown – S29.21 “Non-commercial user-generated content,” is more colloquially known as The MashUp Exception. With conditions (amateur creation, attribution, legitimate source material, and a consideration of market effect), creativity at its most nascent is protected as lawful activity. While the scope is vast, at the very least S29.21 seems tailor-made to protect video involving a dancing cat. (For more on 29.21, see here and here.)

Or if the musical accompaniment to the cat was unintended, the unsung heroic exception of S30.7 “Incidental Use” comes to mind:

It is not an infringement of copyright to incidentally and not deliberately (a) include a work or other subject-matter in another work or other subject-matter; or (b) do any act in relation to a work or other subject-matter that is incidentally and not deliberately included in another work or other subject-matter.

Incidental use is not limited to amateur creation, nor is it confined to any specific purpose of use. That said, it has provided Canadians with some bragging rights in a particular genre; as Howard Knopf wrote over a decade ago, “This section is the envy of American documentarians … .”

The entire list of exceptions is extensive and should be part of any algorithmic effort to pronounce judgement on use of copyrighted works. In this regard, artificial intelligence could lead to better outcomes for copyright owners and users alike, if such systems are appropriately seeded, capable of learning from existing and ongoing court decisions, and attuned to the nuance that permeates application of the law. To rephrase my former professor’s words: comprehensive information in, contextual decisions out.