As generative AI becomes more popular, Fairly Trained wants to highlight good students. Several companies have even already obtained its certification.
Labels are commonplace in many sectors: Protected designation of origin (PDO) for food products, Energy Star for electronic devices, Ecoprod for the cinema… There is no shortage of examples and you just have to look behind any packaging to find at least one.
So, at a time when many artists are worried about what tools like ChatGPT or Midjourney do with their creations, and this, without any remuneration, wouldn’t a label of circumstance be welcome?
A label to see more clearly
It is difficult to know whether a generative AI has not been trained using copyrighted creations and without the consent of their authors. While the giants of the sector consider themselves entitled to do so, other companies are trying to follow a more ” fair » towards artists and other creators. Either by respecting the rights of the licenses under which their works are recorded, or by turning exclusively to the public domain or to material under free licenses.
However, how can we recognize actors who respect these principles? It is to answer this thorny question that Fairly Trained was created. This non-profit organization’s mission is to control the training data of companies wishing to obtain its label. This can then certify that an AI has been “ fairly trained “, thus allowing users to use tools that correspond to their values, at least if they can and if they want to.
To date, nine generative artificial intelligences – eight specialized in audio and one in images – have managed to obtain certification from the organization. It is :
- Beatoven.AI ;
- Boomy, who had some disagreements with Spotify last year;
- BRIA AI;
- Endel
- LifeScore, found in the Deezer Zen application;
- Rightsify;
- Somms.ai ;
- Soundful;
- Tuney.
Of the ” Copyright-friendly AI “, it’s possible ?
The people behind Fairly Trained didn’t come from nowhere. His advisors include Maria Pallante, CEO of the Association of American Publishers (AAP), Tom Gruber, co-founder of Siri, and Max Richter, a composer and pianist much loved by Sam Altman, CEO of OpenAI. Its creator, Ed Newton-Rex, was responsible for the development of Stable Audio within Stability AI, the company behind Stable Diffusion.
Of course, if he is there today, it is because he left his former employer in November 2023. According to him, the latter “ exploits the creators » using their works to drive its products. “ As long as generative AI companies say “Yes, it’s fair use, we can scrape whatever we want”, I think there will be a battle between two camps », Newton-Rex said at the time.
Will such a label really be useful? It’s hard to imagine that the giants of the industry will seek to obtain it, even though they defend the use of all works, even those subject to copyright. If several legal battles on this subject are underway around the world, it will certainly take stronger political action than the European AI Act to change things.
Several months ago, a friend asked me if it was conceivable that one day we would have “ Copyright-friendly AI “. At the time, it still seemed difficult to answer in the affirmative, as tools like ChatGPT or Midjourney need a colossal amount of qualitative data to be so effective. Ed Newton-Rex, however, is encouraging: “ There is a mutually beneficial system that works for everyone “. And this is precisely what his label must prove to us.
Source : Bloomberg
0