Over 90% couldn't distinguish AI-generated videos. We need to consider the implications of this tech evolution.
Study Results on Identifying AI-Generated Video
Original: Runway Research | The Turing Reel
Importance: AI生成動画の識別能力が低下していることは多くの人に影響を与えるため。
Summary
A study by Runway revealed that over 90% of 1,043 participants could not distinguish between AI-generated videos and real ones. This highlights the advancements in video generation technology with the latest Gen-4.5 model, suggesting that discerning video authenticity is becoming increasingly difficult. Runway has launched a site where anyone can test this ability.
Key Points
- Over 90% couldn't identify AI videos
- Indicates evolution of Gen-4.5 model
- Shifts perception of video authenticity
- Launch of site for public testing
- Human-related videos easier to identify
View developer notes (APIs, breaking changes, migration)
Runway conducted a study using the Gen-4.5 model to evaluate the ability to distinguish between generated and real videos, recruiting 1,043 participants who viewed 20 videos each (10 real, 10 generated). Only 9.5% (99 participants) achieved statistically significant accuracy, with overall detection accuracy at 57.1%. Human-related videos were detected at 58-65%, while animals and architecture fared worse at 45-47%, indicating a potential fundamental shift in how we perceive video authenticity.
Source: https://runwayml.com/research/theturingreel
Outlet: Runway
This article is an AI-generated summary (OpenAI GPT-4o-mini) of publicly available information from Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Sakana, and other vendors. The original source URL is always provided in accordance with fair-use citation requirements. Summaries are AI-generated and may contain mistranslations or misinterpretations. Always verify details with the original source.