Microsoft’s AI CEO believes it has agreed to a ‘social contract’ for AI training

Microsoft AI CEO Mustafa Suleiman recently found himself at the center of a heated debate after making some controversial comments at the Aspen Ideas Festival, where he claimed that the internet essentially acts as “freeware” for training AI models, a claim that drew harsh criticism from content creators and everyday users.

Around the 13-minute mark of the interview, the moderator raised concerns about training AI using online content, noting that there were many authors in the audience and that OpenAI uses transcripts of YouTube videos to train its models.

The interviewer questioned who should own the intellectual property (IP) in such cases and how commercial agreements around IP should be structured, suggesting that AI companies may be “stealing” the world’s IP.

Here is Suleiman’s response to this question:

With regards to content that is already on the open web, the social contract around that content since the 90s has been that it is fair use. Anyone can copy it, recreate it with it, reproduce it with it. That’s what “freeware” is, and that’s how it’s been understood. There’s another category where a website, publisher, or news organization explicitly tells me not to crawl or scrape it for any reason other than to index it so others can find this content. This is a gray area, and I think it will be resolved in court.

Suleyman’s remarks suggest that AI developers are free to use the vast amounts of data available online to train their models. This view seems to overlook the complex legal and ethical issues surrounding content ownership and usage rights. Fair use permits limited use of copyrighted material for purposes such as criticism, education, and research. But using vast amounts of content to develop AI models crosses these boundaries, especially when there is a clear commercial motivation involved.

The comment was not well received in the online community, with many X (formerly Twitter) users reposting the video with their own responses to his comments. Tech industry figures like Tom Warren questioned Microsoft’s double standards and asked whether the company would be happy for its Windows operating system to be treated as freeware.

Others, such as artist Denman Luke, stressed that there is a difference between viewing or downloading artwork online and using it commercially without permission, with the latter amounting to theft.

The internet is full of content created by many people who earn money from their work, including journalists and artists. When AI companies use this content without permission to train their models, they are taking away the value of the models without paying the original creators. The interviewer compared this to an author referencing other books when writing their own: they don’t have to pay the referenced authors, but they still have to buy the books or pay library fees.

Suleiman responded by arguing that thanks to AI, the cost of producing information will soon be close to zero: traditionally, creating information has been costly, but AI models could make the cost of producing information close to zero.

Incidentally, OpenAI has recently been actively signing content licensing agreements with major media companies and online platforms, including Reddit, to use their content to train its GPT models.

The debate highlights the urgent need for clear guidelines and ethical standards in the field of AI, and raises broader questions about the future of the information economy and the need to adapt to a rapidly changing technological environment.

What do you think about this issue? Share your thoughts in the comments section below.