Urgent.News

What's breaking now, across thousands of outlets.

AI

Code of Principles for AI: Can This Really Protect Intellectual Property?

In a series of cases, characters, musical compositions and other works created by generative artificial intelligence have born a striking resemblance to existing copyrighted works, raising suspicions of copyright infringement.

In the wake of recent cases where AI-generated characters, musical compositions, and other works have eerily mirrored existing copyrighted material, raising concerns over potential copyright infringement, the Japanese government has taken preliminary steps. The final draft of a code of principles, which service providers for generative AI must adhere to, signifies progress in setting regulations to curb intellectual property rights violations. However, the draft falls short of fully addressing the issue.

The draft mandates that providers disclose on their websites the data utilized to train their AI models and the rationale behind data collection. It further encourages providers to craft their own rules for safeguarding intellectual property rights. This draft applies to overseas providers like Google and OpenAI, offering generative AI services in Japan.

A formal decision on the code is anticipated soon, with implementation slated to commence in the autumn. However, it's crucial to note that the code lacks legal enforceability.

If a provider fails to meet specific requirements, they may be granted an exemption upon publicly justifying their inability to comply. The code's shortcomings are glaring, especially concerning manga, music, novels, and other works. Copyright holders have been severely restricted in the specifics they can request from service providers to disclose.

Despite initially including provisions for detailed disclosure covering data types, nonpublic data, and decision-making processes, these were removed for the final draft at the behest of service providers.

A key provision in the final draft asserts that "immediate criticism should be avoided" even if a service provider fails to disclose certain items. This appears to reflect provider concerns about being unduly burdened, which could harm their competitiveness in AI development. The government, still grappling with the aftermath of "digital defeat," a term coined due to processing delays for relief payments during the COVID-19 pandemic, seems eager to rectify past mistakes but overly favors AI development and business interests.

If copyright infringement cases persist following the code's implementation, the government must not shy away from enacting legislation to impose penalties. Indeed, the European Union has already established a system enabling it to penalize AI service providers that infringe copyrights. The recent revision of Japan's Copyright Law in 2018, which allowed AI to train on content without the copyright holder's consent, has led to a surge in cases of suspected intellectual property infringement. To effectively curb infringement, the law must be revised once again.

Written by urgent.news from The Japan News by The Yomiuri Shimbun's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at japannews.yomiuri.co.jp →

More in AI

Audit the Server, Not the Model: A Free-Tier Test for AI Coding Tools

The model decides what the tool says, but the server decides what the tool does, and most developers audit the first and ignore the second.

  • Server performance crucial for AI coding tools to execute generated code
  • Free server audit evaluates tool's capabilities and reliability
  • Metrics include time to first edit, run, and error-to-fix cycles

Debugging a Flaky LLM Pipeline: Timeouts, Truncation, and a 40-Line Probe

The failure had nothing to do with the free server, and everything to do with two assumptions I had baked into my harness. I moved a small LLM batch pipeline to MonkeyCode's free server to cut costs, and within an hour the same prompts that worked on a paid endpoint started returning empty completions and hanging requests.

  • Batch job processed 40 requests, then returned 200 with empty content for next 3-4 requests
  • ReadTimeout occurred after 30-second hang, indicating real issue with input length
  • Fixed context-window overflow and dead keep-alive connection in author's harness

More from Tuesday 25 August →