TOKYO - Hitachi has announced a groundbreaking advancement in technology to detect whether a document was created by generative AI, employing a world-first "triple electronic watermark" system aimed at preventing the spread of fake information.
As the internet continues to be flooded with "fake news," a negative byproduct of the widespread use of generative AI, Hitachi's new technology offers a significant step forward in combating this issue.
The system works by embedding specific words that function as "electronic watermarks" within the text generated by AI. The presence and frequency of these words help determine if the document was AI-generated.
A reporter demonstrated, "This text contains 'watermarks' that can identify if it was created by generative AI."
The innovation addresses a critical vulnerability: previously, a single electronic watermark could be easily forged. Hitachi's latest technology uses multiple, layered electronic watermarks, enhancing security.
Koichi Nagatsuka, from Hitachi's R&D Group, explained, "'Electronic watermark' technology essentially distinguishes between AI-generated and human-created text."
In an era where the proliferation of fake information has become a significant social issue, this technology has garnered attention for its potential to authenticate the creators of videos and documents.
Although electronic watermarks have been effectively used to detect fake videos and images, applying them to text has been challenging. Nagatsuka noted, "With text, changing even a single word can alter the entire meaning, making it difficult to implement."
Hitachi plans to further develop this technology to prevent the spread of misinformation and combat fraud using generative AI.
Source: TBS
