Japan's Cabinet Office on Tuesday presented a draft of the "principle code" which establishes rules for protecting intellectual property rights that generative AI providers must comply with.
Presented to an expert panel, the draft focuses on a provision that asks AI providers to disclose AI training data -- an area in which concerns about copyright infringement remain persistent -- at the request of rights holders and under certain conditions.
There were no notable objections from the experts who met on Tuesday. The matter was left to the discretion of panel chair Toshiya Watanabe, vice president of the Institute of Science Tokyo, effectively signifying its acceptance.
The government intends to finalize the basic principles soon and to begin implementing them as early as this autumn.
The formulation of these basic principles is based on the so-called AI Law enacted last year. This marks the first time that rules specifically focused on protecting intellectual property have been established for generative AI.
Aimed at balancing technological innovation with the protection of intellectual property, the principles are intended to encourage companies to ensure the transparency of AI services. To avoid unduly stifling business activities, they do not include penalties and are not legally binding.
The "comply or explain" approach is being used -- companies are urged to fulfill their obligations regarding accountability if they cannot comply.
One legal professional expressed support for the draft at the meeting, stating, "It's important to take this first step," but also pointed out challenges regarding effectiveness and called for ongoing review of the content.
Another legal professional commended the fact that the draft explicitly states that it "does not require the mandatory disclosure of sensitive information, such as trade secrets," taking into account concerns from the business community.
The principle code is centered on three pillars regarding training data: disclosure of an outline; disclosure to rights holders; and disclosure to AI users.
Regarding the disclosure of an outline, companies are required to publish information on their websites regarding the AI training process, the types of training data and collection methods. As to the disclosure to rights holders and AI users, companies are required to disclose -- under certain conditions and in response to requests from rights holders or generative AI users -- whether the URL of a specific creative work is included in the AI's training data.
According to the draft, the principles will apply to developers of generative AI systems that create text, images and other content, and AI service providers.
The draft explicitly states that the principles also apply to foreign companies "when their systems or services are provided to the Japanese market." This was included with such companies in mind as Google LLC and OpenAI Inc., both based in the United States.
The Cabinet Office intends to ensure a certain level of effectiveness by encouraging companies to report their initiatives and by publicizing the results. The basic principles will be revised as needed in light of international trends.
The current Copyright Law in principle allows the use of copyrighted works, such as artworks and articles, in training data. With the emergence of generative AI that can easily produce content strikingly similar to copyrighted works, calls for the protection of rights are growing louder.
----
This article is from The Yomiuri Shimbun. Neither Dow Jones Newswires, MarketWatch, Barron's nor The Wall Street Journal were involved in the creation of this content.
YDN-M0000225977-1
Comments