December 2, 2022
Our world is moving ever faster thanks to the changes in our technology that happen daily. One rising technology which is gaining heavy impact on how society is run is Artificial intelligence, machine learning in particular. Neural Networks are begin used in various applications across the planet today, and GANs are incredibly efficient in handling images.
GANs are generational adversarial Networks. At their core they are not just a singular Neural Network, but a system of two separate and competing Neural Networks operating simultaneously (4). One of which is trained in order to generate images similar to that of which it is fed (generator). The other half of the model is trained to detect
wether an image is real or generated (discriminator). Both halves of a GAN are used to train each other, and as one half gets better the other does as well. In the end you have a system of Networks, both of which fine-tuned to excel at their respective functions. A GAN model is said to be completely trained once the discriminator can no longer depict the fakes generated by the generator. GANs aren’t just limited to images but can be used for any type of digitized data.
While GAN models take a long time to train, the resulting programs are extremely powerful. Some of the current applications of GANs are (3):
Generate artificial photos
Add animation to still photos
Identify elements within a photo
Generate Datasets to train other models
Increasing resolution of photos
Generating fake elements in videos (coin in Tom Hanks video)
Because the method of creating a GAN is generalized, it’s able to boast an impressive catalog of applications. One of which that I find distinct in not only application but also ethicality is the generation of deep fakes.
A deep fake is “a digitally forged image or video of a person that makes them appear to be someone else” as described by TECHSLANG (5). Here's where I admit I lied earlier. Scroll back up to the video of Tom Hanks, could you tell something was off before? If you did then you were right because the person you see is a façade. The face you see and even the voice of the actor was generated my GAN models. The coin is real though...
Thinking about the applications of GAN models in this regard raises many questions about the ethics of said programs. Having
the ability to generate representations of data can be cool like generating digital instruments, or helpful like for predicting which protein sequences would combine most successfully (6), or harmful like for exploiting people with facetious representations of them. Is there a line that can’t be crossed? Is there really a good vs bad in regard to applications, if so, how do we draw the line? Are potential positives greater than potential negatives? Should advancements in AI such as GAN be regulated, if so by whom?
Another important question that many may not think about is what rights to these models deserve if any? Attention has been brought to this question by the rise of creations by AI models, like generated art. One example in particular is by DABUS. DABUS is an AI
developed by physicist Stephen Thaler, which created a geometric design for a food container. When Thaler went to get a patent for his design, some courts approved the patent while others denied it. This disjunct made way for a conversation of whether or not an AI machine can be an inventor or even has capacity (7).
Before giving my answer to these questions I think it’s crucial to take one step first. Whenever these conversations are held, “AI” and most machines more robust than simple arithmetic are surrounded aby a cloud of mystery. Let’s lay the foundation, all AI models including GANs are mathematical formulas. There is no hidden brain or magic. They can become very complicated, but there is nothing truly mystical about them. With this settled, I believe it’s best to see AI machines in the same light as other tools we use such as pencils, cars, watches, etc. In saying this, it would be ignorant to not weigh the unique power that computational machines can have. I believe that GANs are truly powerful and can bring much benefit an enumerable of fields.
In regard to the rights of AI machines, I believe the creator(s) of the machine and contributors to each product should be credited for valuable creations. For example, if a model is trained to generate pictures based on a given text, I believe the creators of the AI along with who provided the script should all be ascribed for the art produced.
The morality of each application is much trickier. In each light cast there is an accompanying shadow. For example, generating medical data for improving predictive analysis (8) is the same technology that can be used to blackmail individuals with AI generated images of t
hem. Unfortunately, I believe this potential for dangers is an evil of AI machines we’ll have to live with. I don’t see any effective way to truly regulate the use of any type of models. Implementing general solutions would lead to impeding technologies that have too much potential to stifle.
The use of GANs can be implemented in various ways in our current society. With the strong amount of applications of this technology, an equally strong question of ethicality arises. What power do your belief people should be trusted with? Do you believe the advancement of AI machines should continue to grow freely?
Comments