Generative Adversarial Networks (GANs) are a subset of Machine Learning algorithms with the ability to generate synthetic content that resembles real-world data.

For instance, GANs can create a fake Picasso painting or images of a mouse that does not exist. Observers, beware about this proliferation of online scams that can leverage automated fake image generation by using GANs or similar means. In a business world, such techniques give rogue actors an advantage of creating realistic fake websites at scale.

In the near future, it wouldn't be uncommon to see websites with fake logos or other artifacts infringing on a brand's trademarks. So what does the Maltese entrepreneur need to know to protect its business copyrights and other valuable customer data? Since the inception of GANs, the technology has seen rapid development.

Researchers have proposed various GAN architectures and techniques to improve stability, training efficiency, and the quality of generated content. Some notable advancements include deep convolutional GANs (DCGANs), conditional GANs (cGANs) and progressive GANs (PGANs), among others. Put simply, at the heart of GANs is a game-like framework involving two neural networks - the generator and the discriminator - that work in tandem to create and evaluate content.

Now let us delve deeper into how this process works. To start with there comes the Generator in scene one. The generator network starts with random noise as input and creates data. For example, in the case of generating images, the generator produces pixel data. Initially, the generated data is likely to be of poor quality and far from resembling the real data. Enter his nemesis - the Discriminator: this network is responsible for evaluating data, determining whether it is real or fake. It takes both real and generated data as input and assigns a probability that a given data point is real.

During training, the generator and discriminator networks engage in a competitive process. The generator strives to create data that the discriminator cannot distinguish from real data, while the discriminator tries to improve its accuracy in distinguishing between real and generated data. As training progresses, the generator gets better at creating realistic data, and the discriminator becomes more adept at differentiating between real and generated data.

This competitive dynamic results in a generator capable of producing high-quality data that is often indistinguishable from human-created content. Business leaders in Malta need to acclaim's themselves with GANs. These have revolutionised the field of computer vision, but their use comes with several potential drawbacks.

Firstly, GANs require substantial computational resources, making them less accessible for researchers at university or small business organisations with limited hardware capabilities. Combined with a high computational cost this also translates to longer training times, which can be a significant hindrance in rapid development cycles.

Secondly, GANs can suffer from mode collapse, where the model fails to capture the diversity of the training data and instead generates very unmatched outputs. This limits the utility of GANs in applications requiring a broad range of distinct outputs.

Thirdly, training GANs is often a delicate process; it requires careful balance between the generator and discriminator, which can be challenging to achieve and maintain. This sensitivity can lead to unstable training processes and unpredictable results.

Additionally, there are ethical concerns, especially regarding the generation of realistic but fake images or videos, which like fire in a baby's hands can be used maliciously for misinformation or deception.

Determining responsibility for the misuse of GANs can be challenging. Lawyers ask: is it the responsibility of the developer, the user or the platform hosting the technology?

Addressing these ethical concerns is crucial as GANs become more integrated into our daily lives. Striking a balance between creative freedom and responsible use is an ongoing challenge. Just like embarking on a magical mystery tour, one uses GANs to create new data that is indistinguishable from real data.

This makes them a good choice for applications where creativity is important, such as image generation and text generation. GANs can be used to generate data of any kind. This makes them a good choice for applications where the data is not well-defined, such as creative writing. So, what are the extensive attributes of GANs?

This technology makes them a good choice for applications where a lot of data is needed, such as training machine learning models.

Another attraction is the novelty of GANs. These can be used to generate novel data that has never been seen before. This makes them a good choice for applications where new ideas are needed, such as product design and exclusive marketing. But there are pitfalls such as the generation of deepfakes and misinformation.

One of the most pressing issues is the use of GANs to create deepfake content, where individuals' voices and images can be manipulated to create convincing fake videos. This technology has the potential for malicious use, including spreading misinformation, impersonation and fraud. Political deepfakes prior to major public elections have been a particular focus of concern, where synthetic media could be used to influence elections, sow discord or spread propaganda.

The fear is that deepfakes could become powerful tools for disinformation campaigns, undermining trust in democratic processes and institutions. The misuse by rogue states and their lawmakers of GANs is well documented. They can be used extensively by malicious politicians for purposes like creating deepfakes, conducting financial fraud or destabilising economies.

It is essential that public and commercial sentiment in Malta could become wary of these technologies. High-profile cases of misuse could lead to headlines that emphasise the risks over the benefits, shaping a negative and surreal public image. So, what can regulators such as MFSA, FIAU, MDIA and MCA and others do to regulate misuse? In extreme cases, regulators may be tempted to impose strict caveats on the liberal use of GANs, potentially stifling innovation and creating hurdles for legitimate uses in industries ranging from entertainment to gaming, manufacturing and healthcare.

Such abuses could further contribute to a climate of distrust and aversion towards GANs, affecting their adoption in commercial applications and in mainstream political arenas. Yet, not everything that touches GANs turns into ruin. They can be very useful, if properly handled by the scientific and commercial communities to highlight and focus their positive, transformative potential. Wake up Malta and do not hesitate to join the revolution. 

 

George M. Mangion is a senior partner at PKF Malta

gmm@pkfmalta.com