Is paying artists enough to convince them to embrace AI?
Is paying artists enough to convince them to embrace AI?
向艺术家付费，足以说服他们拥抱人工智能吗？
Illustrators have spent years sounding the alarm about generative artificial intelligence startups training their models on artists’ work without permission. They’ve pointed out how the practice is tantamount to theft, and in response, many gen AI boosters have argued that it’s necessary for the technology’s evolution. This has led to contentious legal battles, but it’s also given rise to a new wave of AI startups like Pippa that market their services as being more ethical than the competition.
多年来，插画家们一直在对生成式人工智能初创公司未经许可使用其作品训练模型发出警告。他们指出，这种做法等同于盗窃。对此，许多生成式 AI 的支持者辩称，这是技术演进的必要过程。这不仅引发了激烈的法律诉讼，也催生了像 Pippa 这样的一批新型 AI 初创公司，它们将自己的服务标榜为比竞争对手更具道德感。
Like most companies selling access to text-to-video models, Pippa’s main product is short bursts of user-generated footage that can be cobbled together to tell visual stories. But what sets Pippa apart from other players in the gen AI video space is the way its business model is designed to pay artists directly. Every time a Pippa subscriber generates an image or a video clip whose style is derived from a human artist’s work, that artist is compensated by the platform.
像大多数销售文生视频模型访问权限的公司一样，Pippa 的主要产品是用户生成的短视频片段，这些片段可以拼凑起来讲述视觉故事。但 Pippa 与生成式 AI 视频领域其他参与者的不同之处在于，其商业模式旨在直接向艺术家付费。每当 Pippa 的订阅用户生成一种风格源自某位人类艺术家作品的图像或视频片段时，该艺术家就会获得平台的报酬。
Pippa’s cofounders, Hogan Shrum and Sean Wright, think that their focus on working with artists is what will attract customers. They also believe that their company will prove that AI firms can generate content ethically. But Pippa faces an uphill battle as it seeks to convince artists. Even as it tries to create an ethical spin on AI video generation, the company is still getting off the ground with the same sloppy models that other text-to-video services offer.
Pippa 的联合创始人 Hogan Shrum 和 Sean Wright 认为，他们对与艺术家合作的重视将吸引客户。他们还相信，他们的公司将证明 AI 企业能够以合乎道德的方式生成内容。然而，Pippa 在试图说服艺术家时面临着艰巨的挑战。尽管该公司试图为 AI 视频生成赋予道德色彩，但其起步阶段所使用的模型，与其它文生视频服务所提供的模型并无二致，同样存在粗糙的问题。
When I recently spoke with Shrum and Wright, they told me that they see Pippa as a way to move forward from “the bloody history that the AI industry has been built on.” The duo want to set an example for their peers and show the public how AI companies don’t necessarily have to steal to make their products work.
最近我与 Shrum 和 Wright 交谈时，他们告诉我，他们将 Pippa 视为一种摆脱“AI 行业建立在血腥历史之上”的方式。这对搭档希望为同行树立榜样，并向公众展示 AI 公司并不一定非要通过盗窃才能让产品运作。
“All of these models from the past 18 months have been trained on real people’s art without their permission — that’s how it was built,” Wright said. “I harken it back to when Napster was a thing and everybody was stealing music. But then, Apple came out with the 99-cent song, Napster got shut down, and we figured out a different way to get musical artists paid.”
“过去 18 个月里的所有模型都是在未经许可的情况下，利用真实艺术家的作品进行训练的——这就是它们的构建方式，”Wright 说道。“我把它比作 Napster 盛行、每个人都在窃取音乐的时期。但后来，苹果推出了 99 美分的单曲，Napster 被关闭，我们找到了另一种让音乐艺术家获得报酬的方式。”
Pippa’s payments to artists aren’t very large on an individual basis — especially when you compare them to how much the service charges users for monthly subscriptions (which range from $14.99 to $99.99.) The company says that artists receive $0.005 per image and $0.003 per second of video. The more generations a user makes, the larger an artist’s payments become, and Pippa also gives them access to a cut of a 5 percent royalty pool that’s funded by the platform’s overall subscription revenue. In Pippa’s press materials, the company likens its royalty payment structure to Spotify’s — a curious choice given how many artists have spoken out about being shafted by the music streamer.
Pippa 给艺术家的单笔报酬并不高，尤其是与该服务向用户收取的月费（从 14.99 美元到 99.99 美元不等）相比。该公司表示，艺术家每生成一张图像可获得 0.005 美元，每秒视频可获得 0.003 美元。用户生成的次数越多，艺术家的报酬就越高；此外，Pippa 还允许他们从由平台总订阅收入资助的 5% 版税池中分得一杯羹。在 Pippa 的新闻资料中，该公司将其版税支付结构比作 Spotify——考虑到许多艺术家曾公开抱怨被这家音乐流媒体平台“坑害”，这是一个令人玩味的选择。
Pippa — which launched in May — currently has around 800 paying subscribers, and while the company has only signed licensing / model training agreements with four human artists, Shrum said they are in talks to bring four more into the fold. Artists interested in working with Pippa must first go through a multi-step vetting process to verify that the illustrations they submit are actually theirs. Once an illustrator is approved, models trained on their art become available to Pippa’s subscribers. Additionally, the service provides artists a profile page, which can be used to direct people to their work outside of the platform. But Wright noted that Pippa also gives its partner artists the option to submit their work under pseudonyms if they are concerned about being seen as “crossing the picket line.”
Pippa 于今年 5 月推出，目前拥有约 800 名付费订阅用户。虽然该公司目前仅与四位人类艺术家签署了许可/模型训练协议，但 Shrum 表示他们正在洽谈引入另外四位艺术家。有兴趣与 Pippa 合作的艺术家必须首先通过多步骤的审查流程，以验证他们提交的插图确实属于本人。一旦插画家获得批准，基于其艺术作品训练的模型就会向 Pippa 的订阅用户开放。此外，该服务还为艺术家提供个人资料页面，可用于引导用户访问平台之外的作品。但 Wright 指出，如果艺术家担心被视为“越过纠察线”（即背叛同行），Pippa 也允许他们选择使用化名提交作品。
“We’ve had so many conversations with artists where they’re like, ‘I love this, but I don’t want to get ostracized by my community,’” Wright explained. “We’re very early in this space, so there’s some apprehension to participate in it. But we think that can change as people see the economics of how we’re paying artists and that our intent is good.”
“我们与许多艺术家进行过交流，他们会说：‘我很喜欢这个，但我不想被我的圈子排挤，’”Wright 解释道。“我们在这个领域还处于非常早期的阶段，所以人们参与其中会有一些顾虑。但我们认为，随着人们看到我们向艺术家支付报酬的经济模式，并了解到我们的初衷是好的，这种情况将会改变。”
Though Pippa hopes to fully switch to models that are trained on datasets provided by its in-house artists, that technology is still built with open models that the company says “have initial training on the broader set of content out there.” Put more bluntly, Pippa’s tech still involves some degree of using art that was scraped from the internet without creators’ express consent. The company’s intentions may be good, but it doesn’t exactly have a way to completely solve for the creative theft that’s inherent to gen AI.
尽管 Pippa 希望完全转向使用由其签约艺术家提供的数据集进行训练的模型，但该技术仍然是基于开源模型构建的，公司称这些模型“在更广泛的内容集上进行了初步训练”。直白地说，Pippa 的技术在某种程度上仍然涉及使用从互联网上抓取、且未经创作者明确同意的艺术作品。该公司的初衷可能是好的，但它并没有完全解决生成式 AI 固有的创意盗窃问题。
That issue isn’t unique to Pippa; it’s something that must be considered with every “ethical” AI startup whose bespoke models are trained on information that wasn’t created internally. This is why some companies — like Ben Affleck’s InterPositive — have made a point of emphasizing the fact that they develop wholly original, proprietary datasets of their own. But that requires a lot of upfront capital most startups don’t have access to, and the end result is usually very specialized AI tools that aren’t exactly designed for general consumers.
这个问题并非 Pippa 所独有；任何声称“合乎道德”的 AI 初创公司，如果其定制模型是在非内部创建的信息上进行训练的，都必须面对这一问题。这就是为什么一些公司（如本·阿弗莱克的 InterPositive）特别强调他们开发的是完全原创的专有数据集。但这需要大量的启动资金，而大多数初创公司无法获得，且最终结果通常是针对性极强的 AI 工具，并不完全适合普通消费者。