AI Deepfake Study Summary
AI Deepfake Study Summary
AI Deepfake Study Summary
Circumstance: There is growing concern about the public's ability to distinguish AI-generated content from real content. The Herbert Institute wanted to quantify how close deepfakes were to being indistinguishable from authentic content.
Circumstance: There is growing concern about the public's ability to distinguish AI-generated content from real content. The Herbert Institute wanted to quantify how close deepfakes were to being indistinguishable from authentic content.
Circumstance: There is growing concern about the public's ability to distinguish AI-generated content from real content. The Herbert Institute wanted to quantify how close deepfakes were to being indistinguishable from authentic content.
Methods: We conducted a study measuring people's ability to detect deepfakes in both video and audio formats, and to quantify the credibility gap between real and AI-generated content. The study included 36 in-lab participants and deployed a remote online survey to 208 national participants. We measured participants' perceptions of the speaker's credibility and the content's trustworthiness. We also analyzed the detection accuracy rates and credibility scores across both mediums to quantify the performance gap between authentic and deepfake content.
Methods: We conducted a study measuring people's ability to detect deepfakes in both video and audio formats, and to quantify the credibility gap between real and AI-generated content. The study included 36 in-lab participants and deployed a remote online survey to 208 national participants. We measured participants' perceptions of the speaker's credibility and the content's trustworthiness. We also analyzed the detection accuracy rates and credibility scores across both mediums to quantify the performance gap between authentic and deepfake content.
Methods: We conducted a study measuring people's ability to detect deepfakes in both video and audio formats, and to quantify the credibility gap between real and AI-generated content. The study included 36 in-lab participants and deployed a remote online survey to 208 national participants. We measured participants' perceptions of the speaker's credibility and the content's trustworthiness. We also analyzed the detection accuracy rates and credibility scores across both mediums to quantify the performance gap between authentic and deepfake content.
End Result: We found 14% fewer participants believed the deepfake was real. This finding held across both video and audio formats. Detection accuracy was approximately 50-56% for deepfakes (essentially a coin flip). Our research concluded that as deepfake technology becomes more accessible and sophisticated, it poses a real threat to the authenticity of information. This study garnered the advertising equivalent of $2 million in free media coverage for Utah Valley University, significantly raising awareness about this emerging technology threat.
End Result: We found 14% fewer participants believed the deepfake was real. This finding held across both video and audio formats. Detection accuracy was approximately 50-56% for deepfakes (essentially a coin flip). Our research concluded that as deepfake technology becomes more accessible and sophisticated, it poses a real threat to the authenticity of information. This study garnered the advertising equivalent of $2 million in free media coverage for Utah Valley University, significantly raising awareness about this emerging technology threat.
End Result: We found 14% fewer participants believed the deepfake was real. This finding held across both video and audio formats. Detection accuracy was approximately 50-56% for deepfakes (essentially a coin flip). Our research concluded that as deepfake technology becomes more accessible and sophisticated, it poses a real threat to the authenticity of information. This study garnered the advertising equivalent of $2 million in free media coverage for Utah Valley University, significantly raising awareness about this emerging technology threat.