Artificial Intelligence (AI) has improved remarkably since its invention. It has a variety of uses from data analysis, robotics, and applications in the medical field. All these advancements have made impressive societal impacts, however are recent progressions in AI contributing more harm than good to society?
Generative adversarial networks (GAN’s) are a dual AI system which can simply be defined as AI’s which work against each other to improve each other. One AI system considered the generator, develops content and the other AI system determines whether it is fake or not. Every time the secondary AI correctly establishes that a generated piece of content is fake, it provides feedback to the generating AI so that it can improve at producing more believable content. This is an example of machine learning, as the AI systems are continuously bettering themselves from experience without the need to be specifically programmed. This AI technology is considered a neural network as it was created to resemble a human’s neural network, so that the computers will learn like humans.
This leads us to the question as to why GAN’s can be considered harmful? This is because GAN’s are used to create Deepfakes. The term deepfake is used to describe video and audio recordings which have been generated through machine learning. These videos are not real but can be very convincing.
The above video shows a side by side comparison of Robert De Niro and Al Pacino in the film Taxi Driver. In the original film Robert De Niro is the actual actor, however in this video through the use of deepfake technology the actor Al Pacino looks just as authentic as the lead.
Source https://en.wikipedia.org/wiki/File:Woman_1.jpg#filelinks
Another example of the power of GAN’s is the image above. The picture is a generated image from a GAN and despite how real or familiar this person looks the image is not of a real person. Sites such as “this person doesn’t exist” are an entertaining way to witness the abilities of these GAN’s, as on this site every time you refresh the page a new generated image of a fake person is displayed.
For purposes like film production, the use of deepfake technology could actually be quite helpful, however it could also be used for much more damaging motives. Technology like this could be used for propaganda in political campaigns. For example videos could be produced of political candidates saying damaging words which would impact their campaign, however these videos could all be fake and a product of deepfake technology. This causes the overall manipulation of information. Additionally this causes a serious impact in the way information is communicated to the public. It also could potentially decrease the public’s trust in true information as it will become harder to decipher what is real from what is fake.
In a study earlier this year researchers had found that although currently it still takes technical skill and a team effort to produce realistic simulations of people’s faces, it is slowly developing to becoming a more automated and accessible technology. If a wider range of people have the ability to use this there are countless ways this technology can be applied; good and bad.
To prepare for this potential increase of deepfakes making its way online, we should be wearier and more critical of videos and information we are presented in the media and online. It is important for us to be aware of the presence of deepfakes so that we can be less susceptible to believing in false information.
-Shania Mander
One response to “Deepfake Technology: Can you Tell what’s Real and Fake?”