OpenAI: Copyrighted Data ‘Impossible’ to Avoid for AI Training

Recently, OpenAI made headlines with its statement to a UK parliamentary committee, claiming that developing leading AI systems today without utilizing significant amounts of copyrighted data is “impossible.” The company emphasized that advanced AI tools, such as ChatGPT, require extensive training datasets, making strict compliance with copyright law impractical.

In its written testimony, OpenAI noted that stringent copyright regulations paired with the vast array of protected online content means that “virtually every sort of human expression” is barred from being used as training data. From news articles to digital images, most online content is neither freely available nor legally usable.

OpenAI warned that trying to limit training data strictly to public domain materials or works created over a century ago would render AI systems incapable of meeting modern requirements. While they assert that their methods adhere to legal standards, OpenAI acknowledged that collaborating with publishers and establishing compensation frameworks might be necessary “to support and empower creators.” However, the company showed no intention of significantly curtailing its data gathering practices, including the use of paywalled content.

This approach has led to several lawsuits against OpenAI, including accusations from major media outlets like The New York Times for copyright violations. Despite this, OpenAI remains reluctant to overhaul its data collection strategies, viewing strict copyright limitations as “impossible” to work within. Instead, the company aims to leverage broad interpretations of fair use to utilize vast amounts of copyrighted material legally.

As AI technology continues to advance and demonstrate remarkable capabilities in mirroring human expression, legal experts predict intense courtroom disputes regarding copyright infringement, given that these systems are inherently designed to consume extensive volumes of proprietary text and media. For now, OpenAI is placing its bets on a model that favors extensive usage of copyrighted content to fuel its ongoing AI evolution.