Exploration of the Deep Fake technology, its implications and policies, laws in other countries

Technology is an ever evolving field, every day we see something new and innovative in the field. In particular, deep fakes are a topic of heavy discussion. In this blog let’s explore this idea. What is deepfake? Where did it come from? Is it good or bad? How do we combat it? These are some of the questions we shall look into in this blog. 

Definition

First of all, it is crucial to understand what deepfakes actually are. Deepfakes are media content created using AI tools. It can be in the form of videos, audio recordings and photos. They are generally meant to be deceptive and can lead to impersonation and misinformation. Often, videos on Instagram and other social platforms portray political leaders, celebrities and other public figures as talking about something or taking some actions they never would. That's exactly what a deepfake is.

 It typically represents a human subject altered using deep neural networks (DNN). This alters the person's true identity. The underlying technology can change the face, manipulate facial expressions, synthesize faces, and speech. 

Origin

This concept emerged in the year 2017 when a user on Reddit started posting AI generated videos under the name of ‘deepfakes’. It is interesting to know that deep fakes are connected to the evolution of AI and Machine Learning (ML). Even though the name came into light in 2017, the technology of using ML in the area of computer-vision research was known in the media and gaming industries as well as in research related studies. 

Evolution

In 1997, researchers created a Video Rewrite program using lip syncing technology which could portray a person as saying something they did not. These videos were not deepfake given that they used computer vision technology and not DNN. Later on in the early 2000s, face swapping and Computer Generated Imagery (CGI) were created. It could completely reconstruct the way a person looked in an image or a video and created realistic human expressions. 

Further on, in the year 2012, there was a major breakthrough in deep learning models. Deep learning is a way in which computers are taught using huge amounts of data. It uses neural networks and has many layers. In the year 2014, Generative Adversarial Networks (GANs) were born. It was a deep learning model created by Ian Goodfellow. It used fake data to make it look realistic. It is the element that makes deepfakes realistic. Finally, in 2017 and the following years deepfake came into being. All the elements mentioned prior to this led to the creation and evolution of deepfake. 

Types

It is crucial to also know the different types of deepfake that exist. They mainly come in five forms: textual, visual, audio, real time or live and social media deepfake. 

In today's ever evolving world, real time or live deepfake is a relatively new concept. It includes allowing firms to generate advertising clones, governments to imitate political adversaries, and hackers to recreate user voices to pass voice-based authentication. Youtubers are already changing their faces in real-time using a novel deepfake program. For example, DeepFaceLive is an open-source artificial intelligence software that can convert your visage into someone else’s via videoconference and streaming networks. Streamers already have started utilizing the feature on platforms like Twitch, and both broadcasters and developers of any other media output can use this program.

Technology behind deepfakes

The main question that arises now is how does deep fake truly work? Creating a convincing deepfake is truly a challenging task. Sources such as DeepFaceLab and FaceSwap make this task simpler, however, creating high quality deepfakes require powerful GPUs, a lot of training time which can range from weeks to months and skilled post creation editing. The process is divided in many stages - gathering matching source and target footings, choosing the right frames, training deep learning models and conversion and refinement of the videos using professional tools. While deepfakes are being increasingly created and difficult to identify, there are often cues in images and videos suggesting it is fake. 

One very popular example of this is the Deepfake of Former American President Obama, the deepfake clearly showed the ex president saying things he never would. The way I identified that it was fake was through the overlapping images in the video, the background and tone. 

Threats

From all of this we can see that deepfakes pose a great threat to our life. They can often be used for exploitation. Deepfake content pornography disproportionately victimize women. It creates widespread misinformation, especially influencing elections, creating public distress, and can even be used as a method for psychological warfare. 

Deepfakes are at various times used for committing frauds. It is a method for cyber criminals to completely manipulate situations and put themselves in convincing situations. It created opportunities for illegal activity. For example, if a criminal gets access to an individual's sensitive information and personal information they can easily generate a deepfake, which will be hard to determine as counterfeit. Criminals can also sound like authoritative figures or trusted individuals and get money or sensitive information from victims. This actually happened to a British energy company’s CEO in 2019. It cost the business over $243,000. 

Identification

By now we can very well see just how dangerous deepfakes can be, however, there are ways in which deepfakes can be identified and combated. 

Some ways to identify deepfake include noticing: 

• Inconsistent eye blinking

• Mismatched facial features

• Lack of definition in the image 

Combating deep fakes

Combating deepfake is still a blurry topic. Companies like Microsoft and Intel have experimented with methods to combat it using AI tools to check for mismatched blinking and other identification techniques. While all this is being done, there are still numerous policy questions about deepfake that need to be answered. 

Laws and Policies 

Many countries and supranational organisations have created laws and policies in this field:

•  One of the most recognised laws includes that by the European Union. Under the Digital Services Act, platforms must label content as manipulated and take down harmful deepfake content. It also protects individuals from unauthorized use of personal data(including face and voice). 

• China was one of the first countries to regulate deepfake. Their laws state that explicit permission is required by the individual before generation of content. It also states that deepfakes must be clearly labelled or the companies and individuals involved will be severely penalized. 

• Brazil is another country that has taken major steps in this area. Under the electoral reform law and the superior electoral court, deepfake content that is generated is taken down. Also, AI generated misinformation in times of election are accounted for.  

• Nigeria has the cybercrime act of 2015 which penalizes identity theft, impersonation and image manipulation. It can also be applied to deepfake content. 

• France has the Fake News Law of 2018, which has application similar to laws in Brazil. 

• While the USA does not have direct laws towards deepfake on a federal level, few states have created laws to protect the public. This includes states like California, Texas, New york, Illinois, and more. 

Conclusion

Coming from India, I see the widespread use of deepfake all around me and how it affects our daily lives. Currently, there are no specific laws created to settle this problem. While the IT Act 2000 and Indian Penal Code do highlight significant laws related to AI there are no direct methods or ways to solve the issue of deepfake and ever evolving technology. 

The use of deepfake is increasing in Indian society, in the past we have seen deepfakes of famous personalities such as Ratan Tata, Rashmika Mandanna, Ravish Kumar and others. Two PILs have been done to tackle this topic previously. Despite these efforts, a concrete law or guideline is still awaited. I believe a law that can evolve with time is crucial to protect the image of society at large.