Who pays for an AI’s training? | WARC | The Feed

Who pays for an AI’s training?

ChatGPT and Google’s Bard need lots of information, much of it gathered from professional publishers who need to protect their businesses, raising questions of who owns the data that trains large language models. 

Solving this problem will be a critical aspect of the coming phase of commercial AI.

Why it matters

Generative artificial intelligence chatbots don’t yet have an obvious business model, but the potential costs of providing them continue to grow. OpenAI has piloted a subscription service for power users, but will that scale? Google’s Bard and Microsoft’s ChatGPT-integrated Bing are chasing the next phase of search, albeit one in which the model of ad-funded sponsored links will come under significant strain. What happens then?

A bigger question surrounds the information that was used to train the AI, often professionally created images and writing, and who pays for it when the point of access is a chatbot’s response rather than a site visit. Who gets to profit from the results?

In its totality, the story indicates how pervasive the effect of successful AI could be. For publishers, it may or may not change the writing operation, but it’s likely to affect its core advertising businesses if successful.

Visitations

The big issue for publishers is the potential evisceration of a page-visit-dependent online advertising model. “Clearly, [the AI companies] are using our proprietary content – there should be, obviously, some compensation for that,” said Robert Thompson, CEO of News Corp, at an investors’ conference reported by the Wall Street Journal. 

In action

The question has already reached the courts. In January, Getty Images took legal action in the UK and US against Stability AI, alleging that its text-to-image Stable Diffusion tool had infringed on its intellectual property rights. The suit is likely to be foundational to the question of fair use in the AI era.

The question gets really knotty when third parties, like brands or TV production companies or even rival publishers, start to integrate AI into their work. The cases are likely to take months if not years to take place, likely far too slow for the revolution that might take place. But users in commercial settings should understand that these questions will need answering eventually, even amid the fog of AI excitement.

The way of the internet? 

This is not just a question for the AI era. During the tussle over Australia’s News Media Bargaining Code – a set of collective rules for publishers that other governments around the world are also considering – the issue of platforms like Google listing answers to users’ searches as well as links to websites was a vital argument in favour of payments for the use of publishers’ content.

Google (through Google News Showcase) and Microsoft (through MSN) are likely to have some existing relationship with publishers, but these tend to be major publishers with clout; the fate of smaller players is much more uncertain. The US is now considering a bill to allow publishers to bargain collectively and is expected to contain AI provisions.

Sourced from WSJ, WARC, OpenAI