From the Myth of Babel to Google Translate: Confronting Malicious Use of Artificial Intelligence—Copyright and Algorithmic Biases in Online Translation Systems
Many of us rely on Google Translate and other Artificial Intelligence and Machine Learning (AI) online translation daily for personal or commercial use. These AI systems have become ubiquitous and are poised to revolutionize human communication across the globe. Promising increased fluency across cultures by breaking down linguistic barriers and promoting cross-cultural relationships in a way that many civilizations have historically sought and struggled to achieve, AI translation affords users the means to turn any text—from phrases to books—into cognizable expression. This Article discusses the burgeoning possibilities in the 3A Era (Advanced, Autonomous, AI systems) of AI online translation as accessible tools, whose users are data suppliers and feedback providers, and hence, contributors to the programing and improvement processes of these translation tools. On the other hand, this Article also acknowledges the real concerns this new realm raises, stemming from malicious uses of AI, which are most often concealed from the public. Such hidden aspects include built-in algorithmic biases, such as race, sex and gender, color, religion, or national origin biases, which this Article addresses in a discussion of AI systems’ systemic shortcomings. Because AI translation systems learn and function through the data that they receive from data providers, they are vulnerable to societal biases. When users offer feedback, these systems may perpetuate sexist, racist, or otherwise objectionable expressions, of which other users, when consulting the systems, are unaware. Furthermore, examining the current copyright regime, this Article claims, for the first time, that we, as users, have become inadvertent infringers of legal rights since a translation, according to copyright law, is a derivative work owned and controlled by the author. As such, an author’s permission is necessary prior to the creation of a translation, with the author in a position to collect payment when due. Moreover, under current law, the Fair Use Doctrine is frequently inapplicable. This Article claims that the legal and academic communities and policy makers have failed to address AI translation systems within the copyright regime, and that this failure renders the current copyright regime outdated and ill-equipped to handle the advent of sophisticated AI tools. Additionally, this Article states that the present inability of AI technology to routinely capture the nuance of human prose gives rise to another concern. The ubiquitous role such (as yet) flawed AI online translation systems play in translation services, for personal or commercial purposes, should be better balanced with the concerns of authors—who may worry about the linguistic integrity of an AI translation of their work—and their rights, in certain circumstances, to control translations of their work and object to unauthorized AI translations. Understanding the concerns attending unauthorized AI translations under the current copyright regime, while still recognizing that users should be able to profit from the wellspring of literacy which AI translation offers, this Article argues for a harmonization of AI translation with amended copyright protection. To that end, this Article calls on policymakers to adjust the current legal regime to include advanced technologies and suggests new principles for combining legal tools with existing technology. Such an approach would better balance the benefits of accessible AI translation systems with the requirements of a modified, modern copyright regime via the implementation of a method coined “fair use and equality by design.” Additionally, by recognizing the conflicting interests at stake, this Article invites international policymakers, such as WIPO (the World Intellectual Property Organization), to promote the development of international standardized guidelines for the use of AI translation systems, and possibly other AI systems, by emphasizing fair use exemptions and limitations. This Article concludes that by understanding the significant drawbacks of AI translation systems and adopting the suggested principles, policymakers can promote access to an evolving AI technology, while also recognizing the integrity of authors’ linguistic choices and preserving the beauty of linguistic diversity—which, as the ancient story of Babel hinted, is valuable because of, not despite, the challenges it poses.
Shlomit Yanisky-Ravid and Cynthia Martens, From the Myth of Babel to Google Translate: Confronting Malicious Use of Artificial Intelligence—Copyright and Algorithmic Biases in Online Translation Systems, 43 SEATTLE U. L. REV. 99 (2019).