Chatbot applications: "Decisive step towards the development of responsible AI"
Photo: OLIVIER MORIN / AFP
Tech companies that are important for the development of artificial intelligence (AI) have committed to labeling AI-generated content, according to the U.S. government. This pledge is a "decisive step towards the development of responsible AI," the White House said on Friday. Representatives of Amazon, Anthropic, Google, Inflection, Meta, Microsoft and OpenAI were expected to meet President Joe Biden later in the day to announce the details.
At the latest with the success of the chatbot ChatGPT developed by OpenAI, there has been increased concern that AI-generated texts, image or sound recordings could be used for fraud and disinformation. In the United States, also with a view to the 2024 presidential election, ways are being sought to detect whether audio or image material has been artificially generated in order to prevent deception and falsification. Such unlabelled content is already being used in election campaigns.
The corporations "are committed to creating a more comprehensive regulation that makes it easier for consumers to identify whether or not content has been artificially generated," a White House official said. There's still technical work to be done, but the point is that it applies to audio and visual content and will be part of a broader system."
In addition to the digital watermarks for AI-generated content, the voluntary commitments include, according to the AP news agency:
(partially) external security audits of commercial AI systems,
the establishment of reporting channels for security researchers who discover vulnerabilities in the systems,
the publication of reports on technical vulnerabilities and societal risks.
In doing so, the companies are pre-empting a legal regulation on which Congress has yet to agree.
At the EU level, a far-reaching labeling requirement for AI-generated content is currently being discussed as part of the comprehensive "AI Act". Federal Digital Minister Volker Wissing (FDP) called for the introduction of an AI seal for texts in addition to the labeling of image and sound recordings. It makes a difference whether a text was written by people you know and can classify, or by a machine, he told the newspapers of the Bavarian media group.