Generative adversarial networks (GANs): The age of synthetic media

• Author:
• Author name

  Quantumrun Foresight
• December 5, 2023

Insight summary

Generative Adversarial Networks (GANs), known for creating deepfakes, generate synthetic data that mimic real-life faces, voices, and mannerisms. Their use ranges from enhancing Adobe Photoshop to generating realistic filters on Snapchat. However, GANs pose ethical concerns, as they're often used to create misleading deepfake videos and propagate misinformation. In healthcare, there's anxiety over patient data privacy in GAN training. Despite these issues, GANs have beneficial applications, such as aiding criminal investigations. Their widespread use across various sectors, including filmmaking and marketing, has led to calls for more stringent data privacy measures and government regulation of GAN technology.

Generative adversarial networks (GANs) context

GAN is a type of deep neural network that can generate new data similar to the data it is trained on. The two main blocks that compete against each other to produce visionary creations are called the generator and discriminator. The generator is responsible for creating new data, while the discriminator tries to differentiate between the generated data and the training data. The generator is constantly trying to fool the discriminator by creating information that looks as real as possible. To do this, the generator needs to learn the underlying distribution of the data, allowing GANs to create new information without actually memorizing it.

When GANs were first developed in 2014 by Google research scientist Ian Goodfellow and his teammates, the algorithm showed great promise for machine learning. Since then, GANs have seen a lot of real-world applications across different industries. For example, Adobe makes use of GANs for next-generation Photoshop. Google utilizes the power of GANs for both generation of text and images. IBM effectively uses GANs for data augmentation. Snapchat utilizes them for efficient image filters and Disney for super resolutions. 

Disruptive impact

While GAN was initially created to improve machine learning, its applications have crossed questionable territories. For example, deepfake videos are constantly created to mimic real people and make it look like they’re doing or saying something they didn’t. For example, there was a video of former US President Barack Obama calling fellow-former US President Donald Trump a derogatory term and Facebook CEO Mark Zuckerburg bragging about being able to control billions of stolen data. None of these happened in real life. In addition, most deepfake videos target women celebrities and place them in pornographic content. GANs are also able to create fictional photos from scratch. For example, several deepfake journalist accounts on LinkedIn and Twitter turned out to be AI-generated. These synthetic profiles can be used to create realistic-sounding articles and thought leadership pieces that propagandists can use. 

Meanwhile, in the healthcare sector, there are growing concerns over data that can be leaked by using an actual patient database as training data for the algorithms. Some researchers argue that there must be an additional security or masking layer to protect personal information. However, although GAN is mostly known for its ability to deceive people, it has positive benefits. For example, in May 2022, police from the Netherlands recreated a video of a 13-year-old boy who was murdered in 2003. By using realistic footage of the victim, the police hope to encourage people to remember the victim and come forward with new information regarding the cold case. The police claim that they had already received several tips but will have to perform background checks to verify them.

Applications of generative adversarial networks (GANs)

Some applications of generative adversarial networks (GANs) may include: 

• The filmmaking industry creating deepfake content to place synthetic actors and re-shoot scenes in post-produced movies. This strategy can translate to long-term cost savings as they won’t need to pay actors and crew additional compensation.
• The increasing use of deepfake texts and videos to promote ideologies and propaganda across the different political spectrum.
• Companies using synthetic videos to create elaborate branding and marketing campaigns without hiring actual people aside from programmers.
• Groups lobbying for increased data privacy protection for healthcare and other personal information. This pushback may pressure companies to develop training data that are not based on actual databases. However, the results may not be as accurate.
• Governments regulating and monitoring firms that produce GAN technology to ensure the technology is not being used for misinformation and fraud.

Questions to comment on

• Have you experienced using GAN technology? What was the experience like?
• How can companies and governments ensure that GAN is being used ethically?