尊敬的用户您好，这是来自FT中文网的温馨提示：如您对更多FT中文网的内容感兴趣，请在苹果应用商店或谷歌应用市场搜索“FT中文网”，下载FT中文网的官方应用。
For a still nascent technology, generative artificial intelligence already has an impressive resume. It can compose music, summarise wads of legal documents in seconds and generate television adverts based on minimal descriptive input. To become even cleverer, weed out errors and broaden its uses, AI models will need to continuously ingest human-generated content to train on. But the legal framework required to facilitate this symbiosis between man and machine has fallen woefully behind. That puts the long-term development of the technology, and the individuals and companies who feed it with unique data and insights, in harm’s way.
对于一项仍处于发展初期的技术，生成式人工智能已经拥有令人印象深刻的履历。它可以创作音乐、在几秒钟内总结大量法律文件，并根据极少的描述性输入生成电视广告。为了变得更聪明、消除错误并拓宽其用途，AI模型需要不断摄取人类生成的内容进行训练。然而，促进这种人机互惠互利关系所需的法律框架却严重滞后。这使得该技术的长期发展，以及为其提供独特数据和见解的个人和公司处于危险之中。
Generative AI models owe their capabilities, so far, to the reams of text, sounds, images and videos posted online. Much of this has been scraped without the consent of the original creators. A lack of clarity over how copyright laws apply to gen-AI training has also fomented protests and litigation battles around the world. Model developers tend to argue that “fair use” exemptions, which allow the use of copyrighted material under specific conditions, for instance by researchers using short, cited excerpts, are applicable. Artists, musicians and the media strongly disagree. They allege that AI companies are breaching their rights to intellectual property protections, since they go beyond merely excerpting their data.
生成式AI模型的能力迄今为止依赖于线上发布的大量文本、声音、图像和视频。其中大部分是在未经原创作者同意的情况下被抓取的。围绕版权法如何适用于生成式AI训练缺乏明确性，也在全球各地引发了抗议和诉讼。模型开发者倾向于辩称这适用“合理使用”豁免，这种豁免允许在特定条件下使用受版权保护的材料，例如研究人员使用简短的引用摘录。艺术家、音乐家和媒体则强烈反对。他们指控AI公司侵犯了他们的知识产权保护权，因为这些公司不仅仅是摘录他们的数据。
With legal cases ensuing across America and disagreements in Europe over how the EU’s AI Act applies, Britain has taken a welcome initiative to end the ambiguity. Last week it closed a consultation into plans for the future of copyright and AI. But the UK government is also caught between wanting to be attractive for AI companies to scale and drive economic growth, while also protecting its world-class creative industry.
随着法律案件在美国各地接踵而至，以及欧洲内部在如何实施欧盟人工智能法案方面出现分歧，英国采取了一项值得欢迎的举措来结束这种模棱两可的局面。上周，英国结束了对未来版权和人工智能计划的意见征询。但是，英国政府也陷入了两难境地，一方面希望吸引人工智能公司扩大规模并推动经济增长，另一方面又希望保护其世界一流的创意产业。
Though Prime Minister Sir Keir Starmer last week suggested that plans are not set in stone, the consultation did indicate that the government favours allowing AI firms to use copyrighted work to train their models without consent, unless the owner opted out. That approach would be a mistake. It tilts the playing field against content creators, overturning a default right, which has stood for centuries, that no one should be able to profit from another’s established IP. Though the opt-out mechanism is used in the EU, the systems needed to process and enforce them, across numerous platforms and use cases, are patchy.
尽管首相基尔•斯塔默爵士(Sir Keir Starmer)上周表示计划尚未确定，但意见征询显示政府倾向于允许AI公司在未经同意的情况下使用受版权保护的作品来训练他们的模型，除非版权所有者选择不参与。这种做法将是一个错误。它使得竞争环境对内容创作者不利，推翻了几个世纪以来一直被尊重的默认权利，即任何人都不能从他人已确立的知识产权中获利。虽然欧盟采用了不参与机制，但在众多平台和使用案例中，处理和执行该机制所需的系统并不完善。
Legislators around the world should recognise that building a sustainable and fast-growing gen-AI ecosystem depends on the strength and trust of those that produce the source data. Indeed, enabling tech firms to absorb their content, against their will, to build highly scalable competitors against them, undermines the creative and innovative incentives of individuals and companies in the first place.
全球各地的立法者应认识到，建立一个可持续且快速成长的生成式人工智能生态系统依赖于源数据生产者的数量和信任。事实上，让科技公司可以违背他们的意愿吸收他们的内容，以打造高度可扩展的竞争对手，首先会破坏个人和公司的创造和创新动力。
There is a better way forward: supporting licensing markets. Remunerated consent between creators and AI companies gives content makers control over their copyright (it is opt-in by design) and compensation for their work, which incentivises their efforts. It also gives AI models sustained access to high-quality data, free from legal wrangling. Many creative businesses, including this newspaper, have already struck individual content licensing deals with AI companies. Moving from ad hoc deals to a broader market for training licenses is the next step. Governments can help by supporting industry-led transparency standards for how training data is used alongside the development of software to process and track licenses.
有一个更好的办法：支持许可市场。创作者和人工智能公司之间的有偿同意，使内容创作者能够控制他们的版权（有意为之的选择参与）并为他们工作提供补偿，从而激励他们的努力。它还使人工智能模型能够持续获得高质量的数据，免受法律纠纷影响。包括本报在内的许多创意企业已经与人工智能公司达成了个人内容许可协议。下一步是从临时协议转向一个更广泛的训练许可市场。政府可以通过支持行业主导的透明度标准来提供帮助，这些标准涉及培训数据如何被使用，以及处理和追踪许可的软件的开发。
As it reviews its consultation responses, Britain’s government now has an opportunity to set a global standard for how AI and human creativity can coexist. If it wants to create a competitive environment to attract AI companies that actually endures, developing a free and fair market for data is a win-win solution.
在审查其意见反馈之际，英国政府现在有机会为AI与人类创造力如何共存设定一个全球标准。如果英国政府想创造一个有竞争力的环境来吸引人工智能公司，并使其真正经久不衰，那么建立一个自由公正的数据市场是一个双赢的解决方案。