Most people couldn't tell difference between real videos and those generated by Sora

A few weeks ago, Open AI introduced the Sora neural network, capable of generating realistic videos up to one minute in duration with a resolution of 1920 × 1080 pixels based on textual descriptions. Now, HarrisX conducted a survey, asking American adults to distinguish AI-generated videos from real ones. It turned out that the majority of respondents made mistakes in 5 out of 8 videos presented in the survey.

In the survey conducted from March 1 to March 4 in the USA, more than 1000 Americans participated. The researchers generated four high-quality videos using the Sora neural network and selected four short videos filmed with a camera in the real world. Respondents were shown these videos randomly, with the aim of determining whether the video was shot by a person or generated by AI. The opinions of the survey participants varied, but in 5 out of 8 cases, the majority of respondents gave incorrect answers.

According to Variety, this study indicates that content created using generative neural networks is becoming increasingly realistic, making it more challenging to distinguish from real content. This is why calls for legislative regulation in this segment are becoming more frequent in different countries. Among other suggestions is the proposal to mandate users of neural networks to label generated content appropriately to avoid misleading others and prevent it from becoming a source of misinformation.

The Sora algorithm is not yet available to the general public but is already causing significant concern in society, especially in the entertainment industry, where the development of video generation technologies brings about a multitude of negative consequences, such as for film studios. Moreover, there is a growing concern that algorithms like Sora could be used to create fake videos involving politicians and celebrities, leading to unpredictable consequences.