Deepfakes and digital signatures

Our concept of trust is about to change.

The recent rise of deepfakes, highly convincing fake audio or video files that are manipulated using machine learning techniques, is deeply disturbing and will have a high impact on what is meant by trust in the near future. How do we decide what is real and what is fake? Will deepfakes be misused by parties to affect local or global politics? And how will deepfakes impact us in the next decade or so?

Digital signatures may provide a partial solution to these problems. However, the solution is not perfect and cannot prevent that deepfakes will have a profound impact on how we decide what is true and false in the digital era. We might find ourselves in a difficult situation in only a matter of years.

It is easy to see how powerful deepfakes can be: the ability to let Biden or Putin say anything you want him to say in a video fragment could severely impact political discourse. It’s not too hard to believe that in a few years, the technology may have improved as much as to make deepfakes that are indistinguishable from reality. With that assumption in mind, how will we know what videos are real, and which ones are not?

One solution might be to look at digital signatures. In a few words, digital signatures are used to provide a digital method to ensure that a message or a file was sent by a claimed sender, and anyone reading the signature can verify that the message or file was indeed sent by that sender. One example that you may know of are the TLS certificates used by a website: when you visit for example http://www.google.com, you can click the little lock in the address bar and see the certificate, which ensures you that you are indeed dealing with Google LLC.

But digital signatures can be used for any type of data, not just messages or websites. For example, it can also be used to sign audio or video footage! This means that if anybody tampers with the footage, that the digital signature becomes invalid. Let’s say you record a video and sign it with your digital signature, then anyone can verify that this video is from you. Any deepfake based on such a video will not have a valid signature from you. 

Now, a naive implementation of this would only worsen the problem, as was noted in this blog on Forbes. However, digital signatures do have the potential to authenticate real footage. A real video released by the Biden administration would be signed by the White House. A real video of a celebrity would be signed by that celebrity. Deepfakes based on such footage would not, and would therefore not be trusted. This is similar to the analogy of websites: a real website will have the certificate of the company behind the website, while a phishing website will not.

This raises two questions. First, how do we know what the real signature of an entity is? And second, what is the impact on trusting audio or video footage? 

The first question is easy to answer: Similarly to certificates for websites, we will most likely have to rely on PKI technology, where we have to trust some third party with verifying the validity of a signature.

The second question is much harder: A video file may be real, but unsigned by anyone. Or perhaps it is signed by a shady news organization that has tampered with the video but signed it anyway. In the end, we will no longer be able to trust audio or video footage just because it exists, as we were able to do until now. We will only be able to trust the entity signing the authenticity of that audio or video footage. 

One of the implications of this is that the concept of fake news is expanded much further than what we consider today: not only will a polarized society read different news items and dispute basic facts about reality, they will have a visually different experience of truth based on which entities they trust. If what you see is based on what you trust, you may end up in a completely different reality to your neighbor.

And so, we will reach an era where audio and video evidence becomes truly disputable. It will be easy to verify the authenticity of footage signed by persons, organizations and entities you trust, but your truth world will be entirely decided by your personal trust. And so, we may find ourselves in completely different truth worlds from one another in a truly post-truth society.