Generative AI has been making waves in the tech world, with models like ChatGPT, Claude, Bard, and LLaMA producing coherent text by training on vast amounts of data, mostly scraped from the internet. However, this practice of data scraping has recently come under fire, sparking debates about copyright, privacy, and the ethics of AI training.
OpenAI, the organization behind ChatGPT, has been hit with two lawsuits. The first alleges that OpenAI unlawfully copied book text without obtaining consent from copyright holders or offering them credit and compensation. The second claims that OpenAI's ChatGPT and DALL·E collect personal data from across the internet, violating privacy laws.
Twitter has also been in the news, taking measures to protect its data by limiting access to it. To curb the effects of AI data scraping, Twitter temporarily prevented individuals who were not logged in from viewing tweets and set rate limits for how many tweets can be viewed.
Google, on the other hand, confirmed that it scrapes data for AI training and updated its privacy policy to include Bard and Cloud AI in the list of services where collected data may be used.
The public's understanding of generative AI models has grown significantly over the past year. People are now more aware of where the data for these models comes from and the implications of using such data. This has led to renewed debates around data scraping, with experts suggesting that companies need to shift their thinking about data from ownership to access and usage.
The use of personal data in AI models presents unique privacy issues. One such issue is the lack of transparency; it's difficult to know if personal data was used, how it is being used, and what the potential harms are from that use. Another issue is that once a model is trained on data, it may be impossible to "untrain it" or delete or remove data.
The discussion around data scraping for AI training also revolves around whether or not copyrighted works can be determined to be "fair use" according to U.S. copyright law. However, fair use is a defense to copyright infringement and not a legal right, and it can be difficult to predict how courts will rule in any given fair use case.
In conclusion, the controversy surrounding data scraping for generative AI training is complex and multifaceted. It involves issues of copyright, privacy, and ethics, and it's clear that the debate is far from over. As AI continues to evolve, so too will the discussions around the best and most ethical ways to train these models.
Comments