Project 5 - DeepFakes and Forensics
Drawing on Hany Farid's work, research the current state of AI-generated media manipulation. How does blockchain-based verification potentially counter deepfakes? What are the limitations of using the blockchain as a tool for media authentication?
Project 5 - DeepFakes and Forensics
Project 5 - DeepFakes and Forensics
George Legrady
legrady@mat.ucsb.edu
legrady@mat.ucsb.edu
Re: Project 5 - DeepFakes and Forensics
AI-generated media manipulation has become one of the most serious challenges in contemporary digital culture. Deepfakes refer to manipulated media in the form of photos, videos, and audio clips that can be created or altered with the use of artificial intelligence. In the past, manipulated media often required advanced editing skills and could sometimes be recognized by obvious mistakes. Nowadays, artificial intelligence technology can create realistic faces, voices, and motions, and even enable real-time video communication.
I think the works of Hany Farid provide an example of how this problem can be addressed. For a long time now, Farid has been researching modified photographs and videos. He has been concerned with issues related to the authentication of visual evidence and the detection of digital trickery. It is possible to argue that this problem involves more than just technical aspects; it also has social and political implications. Deepfakes can be used for misinformation, harassment, fraud, politics, and the destruction of the public perception of reality.
The current state of AI media manipulation has several issues attached to it because the technology keeps improving in terms of speed and affordability. Using consumer-grade AI software, an individual can create fake portrait images, fake voice messages, edited video clips, or full video sequences. Even though some deepfake content serves an artistic purpose, for instance, when it is used in films or art pieces, such technology may also facilitate deception and fraud because it allows one to create fake proof and spread false information. In addition, audio deepfakes are particularly dangerous because one may manipulate even a few seconds of a person’s voice and create a credible speech message. Moreover, video deepfakes keep advancing too.
One way that blockchain technology can be used to address deepfake media is through its verification capabilities. By documenting the owner of the content, the time and process involved in making the media, and changes made thereafter, blockchain can act as an unchangeable timestamp for any photo, video, or audio that has been entered into it at the time of its creation. Thereafter, by comparing the media to its entry in the blockchain, one can detect any changes that have been made.
In this way, Blockchain verification for media authenticity could work in practice. Provenance is essentially an historical account of the origins of a media object and any occurrences that happened to it subsequently. To cite an example, a photograph captured could be tagged with a digital signature straightaway. The signature can be uploaded to a blockchain or a verification system. Any changes made to the image later on can also be logged in. It should be clear that this process doesn’t necessarily make the image authentic, but at least it can show where the image was created from.
Such an approach fits within the general standards for content authenticity, such as the C2PA and Content Credentials. This method uses cryptographic signatures and metadata to show that digital media is from its source and its editing history. Blockchain can help with this by making the data harder to delete or tamper with secretly. This will be beneficial for artists, journalists, and other organizations, since it makes the chain of trust stronger. In addition, users no longer have to rely only on visual cues when assessing whether the content is authentic.
However, the use of blockchain technology is not a complete solution to combating deepfakes. First, it should be noted that blockchain can only validate the data that is stored on it. In other words, if an artificial video is made, but there is no original proof on the blockchain itself, it will not automatically recognize that the video is fake. Thus, while blockchain can validate the existence of some data, it does not guarantee the authenticity of the information stored.
The other limitation involves adoption. For the blockchain-based verification system to work effectively, there needs to be adoption on the part of the cameras, software firms, media firms, social media platforms, and even consumers. In case a relatively small number of producers adopt the verification system, the large majority of the online media will still be impossible to verify. Moreover, social media sites tend to remove metadata in uploaded videos, making it difficult for a content origin verification system to work.
I think the works of Hany Farid provide an example of how this problem can be addressed. For a long time now, Farid has been researching modified photographs and videos. He has been concerned with issues related to the authentication of visual evidence and the detection of digital trickery. It is possible to argue that this problem involves more than just technical aspects; it also has social and political implications. Deepfakes can be used for misinformation, harassment, fraud, politics, and the destruction of the public perception of reality.
The current state of AI media manipulation has several issues attached to it because the technology keeps improving in terms of speed and affordability. Using consumer-grade AI software, an individual can create fake portrait images, fake voice messages, edited video clips, or full video sequences. Even though some deepfake content serves an artistic purpose, for instance, when it is used in films or art pieces, such technology may also facilitate deception and fraud because it allows one to create fake proof and spread false information. In addition, audio deepfakes are particularly dangerous because one may manipulate even a few seconds of a person’s voice and create a credible speech message. Moreover, video deepfakes keep advancing too.
One way that blockchain technology can be used to address deepfake media is through its verification capabilities. By documenting the owner of the content, the time and process involved in making the media, and changes made thereafter, blockchain can act as an unchangeable timestamp for any photo, video, or audio that has been entered into it at the time of its creation. Thereafter, by comparing the media to its entry in the blockchain, one can detect any changes that have been made.
In this way, Blockchain verification for media authenticity could work in practice. Provenance is essentially an historical account of the origins of a media object and any occurrences that happened to it subsequently. To cite an example, a photograph captured could be tagged with a digital signature straightaway. The signature can be uploaded to a blockchain or a verification system. Any changes made to the image later on can also be logged in. It should be clear that this process doesn’t necessarily make the image authentic, but at least it can show where the image was created from.
Such an approach fits within the general standards for content authenticity, such as the C2PA and Content Credentials. This method uses cryptographic signatures and metadata to show that digital media is from its source and its editing history. Blockchain can help with this by making the data harder to delete or tamper with secretly. This will be beneficial for artists, journalists, and other organizations, since it makes the chain of trust stronger. In addition, users no longer have to rely only on visual cues when assessing whether the content is authentic.
However, the use of blockchain technology is not a complete solution to combating deepfakes. First, it should be noted that blockchain can only validate the data that is stored on it. In other words, if an artificial video is made, but there is no original proof on the blockchain itself, it will not automatically recognize that the video is fake. Thus, while blockchain can validate the existence of some data, it does not guarantee the authenticity of the information stored.
The other limitation involves adoption. For the blockchain-based verification system to work effectively, there needs to be adoption on the part of the cameras, software firms, media firms, social media platforms, and even consumers. In case a relatively small number of producers adopt the verification system, the large majority of the online media will still be impossible to verify. Moreover, social media sites tend to remove metadata in uploaded videos, making it difficult for a content origin verification system to work.