In April, a New York-based startup, Runway AI, introduced a groundbreaking technology that generates short videos by typing a sentence into a computer interface. Although blurry and choppy, these four-second videos signaled a significant advancement in artificial intelligence capabilities, hinting at the potential for increasingly convincing videos in the future.
Ten months later, OpenAI, based in San Francisco, has introduced a comparable system that produces videos resembling scenes from Hollywood movies. The demo showcased short videos of wolly mammoths traversing a snowy meadow, a monster gazing at a melting candle, and a captivating Tokyo Street scene. This development highlights the intense competition of OpenAI among giant and startup companies, including Google, Meta, Facebook, Instagram, Runway, etc. The enhanced technology promises to speed up the work of seasoned filmmakers and possibly replace less experienced digital artists while posing potential challenges in combating online disinformation.
Oren Etzioni, a professor at the University of Washington and the founder of True Media, expressed deep concern about the potential misuse of such technology, especially influencing tightly contested elections. OpenAI has named its new system “Sora” symbolizing limitless creative potential. Despite its remarkable capabilities, the company has chosen not to release Sora to the public. Instead, they are prioritizing understanding the potential risks associated with its use. OpenAI is collaborating with a select group of academics and external researchers to thoroughly evaluate the technology and identify potential misuse scenarios.
To address concerns about authenticity, OpenAI has implemented a watermark to label AI-generated videos. However, the company acknowledges the challenges of spotting and removing these watermarks. The system uses generative AI principles and learns by analyzing digital data, including publicly available videos and licensed content. It responds to short descriptions to generate compelling videos. Although the results are impressive, occasional imperfections and illogical imagery are observed.
The rapid evolution of generative AI technologies, including DALL-E, Midjourney, and other still-image generators, has raised concerns among digital artists, as these systems can now produce images that closely resemble photographs. This has not only complicated the identification of disinformation online but also impacted employment opportunities for digital artists. As AI technology advances, it continues to push the boundaries, unlocking new possibilities and reshaping what we consider achievable. We must explore these developments thoughtfully and responsibly, ensuring the benefits are utilized while safeguarding against potential misuse.