Microsoft executive says everything on the internet is ‘free’ so could be used to train AI

Microsoft AI CEO Mustafa Süleyman said that everything shared on the internet can be considered “free software”. According to him, all online content can be freely used to train productive AI tools.

The executive made the statement during an interview with NBC News last week. His conclusion is based on an old idea about intellectual property on the internet.

“Regarding the content that is already on the open web, the social contract on this content since the 90s has been that the content should be used honestly. Anyone can copy it, recreate it, duplicate it. You can call it free software. That was the understanding,” Suleiman said.

The AI ​​CEO only opened up citations in certain cases where creators have made it clear that their content can only be used for indexing purposes, so anyone consuming the material will always have to go back to the original source. He said these cases would still be decided in the courts.

Mustafa Süleyman’s statement may not make much sense now. Many artists and creators have challenged the idea that everything on the internet is free. This is a different story than the online landscape of the 1990s, especially since the use of AI can involve making a profit.

There are also many issues that need to be discussed based on Solomon’s idea:

• AI aims to create new content, not just share something that has already been done;
• Is it right for someone to profit from the work of others?
• How could creators and publishers make it clear that their content was not suitable for training AI before it became popular?
• Do AI companies do this kind of preliminary consulting?

copyright issues

In the USA, the Recording Industry Association of America (RIAA) filed a lawsuit against Suno and Udio companies for copyright infringement. The companies would have trained their tools on unauthorized copies of tracks owned by their respective labels.

Concerns over the issue led Sony Music to send an open letter to more than 700 labels worldwide, explicitly stating that its catalogue cannot be used for AI training without prior agreement.