There is no doubt that Gen AI platforms are revolutionizing the way we search information or give us human-like personalized answers. However they are referenced from the multiple datasets created through laborious human effort. And we are witnessing instances where creators are not happy with Gen AI platforms for scraping their content without consent or licensing deals.
Companies like OpenAI and Anthropic have scraped billions of web pages in order to train their LLM models. Most of the scraping is done without consent. Its just very recently that they have started to license deals with some of the credible publishers in order to use their creative work to train their AI models.
But smaller creators often don’t have the money or resources to legally prevent Gen AI companies from scraping their work. This scenario may change very soon especially for the music creators as researchers have now developed an AI tool that effectively prevents Gen AI platforms for producing quality output even if they scrape the music.
What is HarmonyCloak?
It is an AI tool that is designed to prevent Gen AI platforms to produce quality music output even if they are trained on the music which is protected by Harmony Cloak. This tool is developed by the reserachers at University of Tennessee, Knoxville and Lehigh University.
How does Harmony Cloak prevent AI Agents from scraping music content for quality output?
Harmony Cloak claims to work by embedding a layer of noise in the original music composition that is inaudible to human ear but prevents the agents of Gen AI platforms to effectively filter out that particular layer of noise in order to have a clean source of music to train themselves for quality music generation.
The tool effectively creates a layer of noise that gets embedded at various timeslines of the music file. This layer of noise remains below the threshold of human audible noise and the AI agents can have a hard time in deciphering the multiple bits of embedded noise that needs to be removed in order to have clean source music to train themselves. The result is erratic output by the Generative AI platforms if they are trained on the AI protected music.
A video published by New Atlas shows remarkable difference in the outputs of Gen AI platforms when they are trained on non-protected music vs. protected music by Harmony Cloak. You can watch the video below to see the difference yourself.
Why Harmony Cloak could be a savior for the music artists from the Gen AI Agents?
The distinctive feature about Harmony Cloak is that different parameters can be used in order to create an artificial noise layer for every track which will make it extremely hard for AI to crack the code for removing the protected noise layer.
Imagine having a database that is encrypted with multiple algorithms. While in theory, AI could figure out the password with the help of enormous computing power, it will require extreme amount of time and money and therefore such an arms race may not be possible for AI agents to crack the code for every single track in order to gain access to the original music composition.
And this is why tools like Harmony Cloak may become a pioneering example for similar tools that could be developed in future in order to prevent AI agents from effectively scraping original human-generated content whether its text, video, Images or music.
Multiple instances of music creators expressing discontent and protest at the Gen AI platforms have been observed in different parts of the world. In 2023, UK Music expressed strong discontent at the fact that tech giants are freely scraping the creative work in order to train their LLMs. They called for stricter regulations to prevent unauthorized access of their work by Gen AI Agents.
You may read more about Harmony Cloak from this source.
