Skip to main content
Podcast Episode - Who Inserted the Creepy?

Our digital future depends on our ability to access, use, and build on technology. A few media or political interests shouldn’t have unfair technological or legal advantages over the rest of us. Unfortunately, litigious copyright and patent owners can abuse the law to inhibit fair use and stifle competition. Internet service providers can give established content companies an advantage over startups and veto the choices you make in how to use the Internet. The Electronic Frontier Foundation fights against these unfair practices and defends digital creators, inventors, and ordinary technology users. We work to protect and strengthen fair use, innovation, open access, net neutrality, and your freedom to tinker.

In principle, intellectual property laws (or IP law, a catchall term for copyright, patents, and trademarks) should serve the public in a number of ways. Copyrights provide economic incentives for authors and artists to create and distribute new expressive works. Patents reward inventors for sharing new inventions with the public, granting them a temporary and limited monopoly on them in return for contributing to the public body of knowledge. Trademarks help protect customers by encouraging companies to make sure products match the quality standards the public expects.

Unfortunately, our IP regimes have strayed far from their original purposes. Too often, protections for free speech and innovation are seemingly forgotten as soon as someone cries “infringement.” An unproven allegation that your video or blog post infringes copyright, or that your domain name infringes someone’s trademark, can be enough to shut down perfectly lawful speech. A patent troll can kill a small company with a bogus lawsuit based on a questionable patent that shouldn’t have been issued in the first place.

Some of these laws simply haven’t adapted well to modern technology. A warped development in copyright law has made it illegal in many countries to modify or even look at the software built into the products you own, even if you’re doing it for completely lawful purposes. Copyright’s legal reinforcement of digital locks, paired with extreme criminal penalties for infringement, has intimidated a generation of would-be researchers, tinkerers, and inventors. And ongoing expansions of copyright law are often decided in secret, closed-door meetings before the public is ever allowed to debate them 

Imbalanced copyright law—and overzealous enforcement—generally favors powerful voices that have a great deal of influence in culture. Extreme copyright laws can intimidate new types of creators, especially those who use new media techniques to criticize dominant culture or powerful entities. Online platforms that support new, independent creators can only thrive when they don’t risk severe legal repercussions for their users’ activities. Similarly, when only the most privileged members of society have access to up-to-date research, only those members can build on that research to create new ideas and inventions. That’s why EFF supports open access to research, so everyone can build on and contribute to our knowledge commons.

Just as new voices can’t thrive if copyright law doesn’t recognize their rights, new players must also have access to the same resources as established ones. When Internet service providers can give preferential treatment to certain content or hardware companies, those technologies harden to the accelerating effect of competition and users can’t access new sources of information and innovative new services. EFF believes that Internet users should have the freedom to use technology however they like without service providers artificially restricting their experience.

Whether we’re fighting patent trolls in court; arguing in Congress for more balanced copyright laws; or urging governments, funders, and educational institutions to adopt open access policies, EFF is committed to building a society that supports creativity and innovation, where established players in the marketplace for technology and culture aren’t allowed to silence the next generation of creators.

Creativity & Innovation Highlights

Reclaim Invention

When universities invent, those inventions should benefit everyone. Unfortunately, they sometimes end up in the hands of patent trolls, companies that serve no purpose but to amass patents and demand money from other innovators and inventors.
We’re asking universities around the country to protect their inventions from patent trolls...

Copyright Law Versus Internet Culture

Throughout human history, culture has been made by people telling one another stories, building on what has come before, and making it their own. Every generation, every storyteller puts their own spin on old tales to reflect their own values and changing times.
This creative remixing happens today and...

Creativity & Innovation Updates

A robot painting a self-portrait


当任何有奇思妙想和互联网连接的人都可以访问计算机生成的图像时,“人工智能艺术”的创作也引发了质疑和诉讼。 关键问题似乎是 1)它实际上是如何工作的,2)它可以取代什么工作,以及 3)艺术家的劳动如何能在这种变化下得到相应的尊重? 人工智能的诉讼在很大程度上与版权有关。 这些版权问题非常复杂,我们专门用另外一篇的文章来单独讨论它们。此文将专注于更棘手的非法律问题。 AI艺术生成器如何工作? AI 艺术生成器的一声由两个不同的部分组成。 例如,首先是教它“狗”是什么,或者更抽象一点,“愤怒”是什么样子的数据。 接着是系统针对提示进行响应并输出。 早期,当生成器没有接受足够的训练时,这些输出只是松散地反映了提示。但最终,生成器将看到足够多的图像来弄清楚如何正确进行响应(这也是人类的做法)。 AI 生成的创意内容可以涵盖从“根据我在发烧梦中看到的图像提示”到“写得非常糟糕的博客文章”。 AI艺术生成器如何“学习”? AI艺术生成器依赖于“机器学习”。 在机器学习过程中,训练算法接收大量数据并分析其不同方面之间的关系。 AI艺术生成有赖于通过接受图像和描述这些图像的文本进行训练。 在分析了图像数据的特征与文本之间的关系后,生成器就可以使用这组关联来生成新图像。 这就是它如何能够接受文本输入——一个“提示”——比如“狗”,并根据其训练数据生成(即“输出”)与输入文本相关联的像素排列。 这些“输出”的性质取决于系统的训练数据、训练模型以及系统的人类创造者所做的选择。 例如:通过提供带有与公共网页上的图像接近的文本标记的图像来训练模型,效果不如使用带有明确的、人工生成的标签的手动注释图像进行训练。 此过程与婴儿学习事物的过程没有太大区别。 例如,很多孩子起初会将所有的动物都认成“小狗”,直到他们有足够的接触和成人的纠正来区分“小狗”和“马”。 面对那些对人类来说都属于模糊的联系,机器学习当然也会犯类似的错误。 例如,如果图像包含标尺,则癌症分类器可以“学习”该图像显示肿瘤。 AI 学会了一条捷径:放射科医生确定为癌性肿瘤的结构图像上有标尺,用于缩放和跟踪尺寸。 良性增生的训练图像来自不同的集合,它们没有标尺。 除了训练数据质量带来的影响,还有不同训练“模型”的影响。 这些模型的名称类似于“扩散”或“生成对抗网络”(GAN)。这些模型中的每一个都有不同的优点和缺点(在撰写本文时,扩散模型通常被认为是最先进的)。 在模型训练期间,程序员引入变量来确定模型输出与其训练数据中图像的相似性。 其他变量决定了系统是优先为提示创建紧密匹配,还是通过输出模型具有较低“信心”(一种用于描述统计确定性的数学术语)作为用户提示匹配的图像来更具实验性。 一些模型允许用户在向模型输入提示的同时调整这些变量。 训练数据来自何方? 通常,训练数据来自网络抓取:找到含有关联文本的可用图像(在某些情况下,随后会增添注释)。 这意味着图像的创作者或图像中描绘的人可能并不知道或明确同意将版权图像纳入分析。 对于最近涉及两场诉讼的“Stable Diffusion”系统——一起由代表几位视觉艺术家的集体诉讼和另一起由 Getty Images 提起的诉讼——其数据集由非营利组织 LAION 索引的 50 亿张图像组成。 有关与这些训练集相关的版权问题的分析,请参阅我们的其他博客。...

Elon Musk peers out through Twitter logo

بدون الاعتمادية، ما هو هدف تويتر إيلون ماسك؟

يسيء تويتر إيلون ماسك فهم ما جعل تويتر مفيدًا في المقام الأول بشكل جوهري. في محاولة لانتزاع الدم من الحجر، أعلن موقع تويتر أن جميع "العلامات الزرقاء" الأصلية - التي أنشأت في البداية كطريقة للتحقق من هوية الأشخاص الموجودة على تويتر كأشخاص حقيقيين/ات - ستختفي في الأول من أبريل/ نيسان....


Back to top

JavaScript license information