Shutterstock是互联网上最大的库存照片和插图来源之一，现在为其客户提供了生成自己的人工智能图像的选择。去年10月，该公司宣布与OpenAI合作，后者开发了广受欢迎且备受争议的DALL-E人工智能工具。现在，这项交易的结果正在beta测试中，所有付费Shutterstock用户都可以使用。
根据该公司的一份新闻声明，新平台提供“该网站提供的所有语言”，并包括在客户现有的许可包中。而且，根据Gizmodo自己的测试，你输入Shutterstock的机器的每一个文本提示都会显示四张图片，表面上是根据你的要求定制的。在页面底部，该网站还建议“更多来自Shutterstock库的人工智能生成的图像”，这些图像提供了对空白的不相关一瞥。
但是，在你抓住机会用人工智能结构取代你所有标准的库存图像最爱之前，请注意：使用人工智能来创造“艺术”的想法是一个越来越有争议的想法。生成式人工智能是一个充满潜在法律和伦理复杂性的领域。
一、为什么要这么担心?
所有的人工智能都是在数据集上训练的，即大量的材料聚合，教会它目标是什么。对于人工智能图像生成器，这些训练集包含由人类制作的图像——通常是人类艺术家，他们的工作是他们的生计。
最近有多起针对AI美术生成器Stable Diffusion和其他版权侵权的诉讼。对于如何处理这些案件，目前还没有明确的法律先例。
Shutterstock的主要竞争对手之一Getty Images表示，该公司短期内不会涉足人工智能的浑水。该网站禁止在其平台上使用人工智能生成的图像。关于这项技术，Getty的首席执行官克Craig Peters说:“我认为这是危险的。我不认为它有责任。我认为这可能是非法的，”他在接受The Verge采访时表示。
很明显，人工智能的“灵感”一定来自于真人的工作。但很难确定AI生成器何时何地窃取了视觉艺术家的作品。解读艺术风格似乎有些主观。另一方面，在人工智能生成的文本中，人工智能的剽窃行为更加明显——尽管并不更加令人震惊。显然，如果不谨慎对待，人工智能可能会为创意领域的盗窃危机铺平道路。
二、Shutterstock是如何解决这个问题的呢?
为了消除人们对版权法和艺术道德的担忧，Shutterstock表示，它使用“从Shutterstock授权的数据集”来训练其DALL-E和LG exaone支持的人工智能。该公司还声称，它将向作品被用于其人工智能一代的艺术家支付报酬。Shutterstock计划通过一个“贡献者基金”来实现这一目标。
该公司在其网站的常见问题解答部分解释道，“如果Shutterstock的贡献者的知识产权被用于开发人工智能生成模型(如OpenAI模型)，该基金将通过对Shutterstock库数据的授权，直接补偿他们。”该公司还表示:“Shutterstock将继续为未来通过Shutterstock AI内容生成工具获得人工智能生成内容许可的贡献者提供补偿。”
向贡献者支付的第一笔款项原定于12月，也就是该公司2022年最后一个财季结束时分发。目前还不清楚上个月有多少贡献者得到了报酬，如果有的话，分配了多少。Gizmodo联系了Shutterstock，询问有关这一过程的问题，但没有立即得到回复。
此外，Shutterstock在他们的人工智能图像使用指南中包含了一个聪明的警告。该公司指出：“您不得使用生成的图像侵犯、挪用或侵犯任何第三方的知识产权或其他权利，不得生成垃圾邮件、虚假、误导性、欺骗性、有害或暴力图像。”而且，虽然我不是法律专家，但这一条款似乎把避免陷入麻烦的责任放在了客户身上。如果生成的图片包含了一些可识别的商标材料，或者是名人的肖像，Shutterstock工具的用户就会注意到并避免重新发布问题内容。
三、但这有用吗?
就效果而言，Gizmodo花了5个不同的提示，类似于“机器人画机器人的图片”，人工智能才真正吐出了一些接近这个概念的东西。同样，每个结果页面都提供了四个人工智能生成的选项。在机器生成的20张图片中，只有这篇文章顶部的一张清晰地展示了一个机器人拿着一幅画的表现形式。其他的……好坏参半。
就目前而言，以及在可预见的未来，我想我还是会坚持使用Shutterstock更标准的产品。
原文：
Shutterstock, one of the internet’s biggest sources of stock photos and illustrations, is now offering its customers the option to generate their own AI images. In October, the company announced a partnership with OpenAI, the creator of the wildly popular and controversial DALL-E AI tool. Now, the results of that deal are in beta testing and available to all paying Shutterstock users.
The new platform is available in “every language the site offers,” and comes included with customers’ existing licensing packages, according to a press statement from the company. And, according to Gizmodo’s own test, every text prompt you feed Shutterstock’s machine results in four images, ostensibly tailored to your request. At the bottom of the page, the site also suggests “More AI-generated images from the Shutterstock library,” which offer unrelated glimpses into the void.
But, be warned before you jump on the chance to replace all your standard stock image favs with AI constructs: The idea of using artificial intelligence to pump out “art” is an increasingly divisive one. Generative AI is a landscape fraught with potential legal and ethical complications.
Why all the worry?
All AI is trained on datasets, i.e. massive aggregations of material that teach it what to aim for. And for AI image generators, those training sets contain images made by humans—often human artists for whom their work is their livelihood.
Multiple recent lawsuits have been levied against the AI art generator, Stable Diffusion, and others for copyright infringement. And there’s not yet a clear legal precedent for how these cases will be handled.
One of Shutterstock’s main competitors, Getty Images, has said it wouldn’t be wading into the murky waters of AI anytime soon. The site banned AI-generated images on its platform. And, with regards to the technology, Getty’s CEO, Craig Peters said, “I think that’s dangerous. I don’t think it’s responsible. I think it could be illegal,” in an interview with The Verge.
It’s obvious that AI must be pulling its “inspiration” from the work of real, live people. But it’s difficult to pin down exactly when and where AI generators steal from visual artists. Interpreting artistic style can seem subjective. On the other hand, AI’s acts of plagiarism are much more apparent—though no more egregious—in AI-produced text. Clearly, if not approached carefully, artificial intelligence could pave the way for a theft crisis in creative fields.
How is Shutterstock trying to get around the issue?
In an attempt to pre-empt concerns about copyright law and artistic ethics, Shutterstock has said it uses “datasets licensed from Shutterstock” to train its DALL-E and LG EXAONE-powered AI. The company also claims it will pay artists whose work is used in its AI-generation. Shutterstock plans to do so through a “Contributor Fund.”
That fund “will directly compensate Shutterstock contributors if their IP was used in the development of AI-generative models, like the OpenAI model, through licensing of data from Shutterstock’s library,” the company explains in an FAQ section on its website. “Shutterstock will continue to compensate contributors for the future licensing of AI-generated content through the Shutterstock AI content generation tool,” it further says.
The first pay-out to contributing creators was scheduled to be distributed in December, at the end of the company’s last fiscal quarter of 2022. It’s unclear how many contributors were paid last month, and how much was distributed, if any. Gizmodo reached out to Shutterstock with questions about this process, but did not immediately receive a response.
Further, Shutterstock includes a clever caveat in their use guidelines for AI images. “You must not use the generated image to infringe, misappropriate, or violate the intellectual property or other rights of any third party, to generate spam, false, misleading, deceptive, harmful, or violent imagery,” the company notes. And, though I am not a legal expert, it would seem this clause puts the onus on the customer to avoid ending up in trouble. If a generated image includes a recognizable bit of trademarked material, or spits out celebrity’s likeness—it’s on the user of Shutterstock’s tool to notice and avoid republishing the problem content.
But does it work?
As far as effectiveness goes, it took Gizmodo five different prompts similar to “robot drawing a picture of a robot” before the AI actually spit out something close enough to that concept. Again, each results page provides four AI-generated options. Of the twenty total images the machine generated. Only the one included at the top of this post clearly showed some representation of a robot holding a drawing/painting. The others were… a mixed bag.
For now, and for the foreseeable future, I think I’ll be sticking to Shutterstock’s more standard offerings.
本文由数字化转型网（www.szhzxw.cn）翻译而成，作者：Lauren Leffer；翻译：数字化转型网宁檬树；翻译审核：数字化转型网默然。
免责声明: 本网站(https://www.szhzxw.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。
本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。
