It was not senseless that Stephen Hawking warned: machines could do away with mankind. And it's not a joke that Elon Musk does promise free artificial intelligence (AI) for everybody. Everything is serious enough. Though AI has not been created in full yet, and the perspectives of such a creation seem rather vague for the nearest future, machine learning is one of the hottest trends to correspond to these warnings and promises.
So, what is artificial intelligence? Is it just a term to denote the increasing influence of technology on our lives? Nowadays nobody can state that AI has been created at full, and there is a lot of uncertainty about the possibilities of using even some parts of it. And all this is not for free, of course. So, it is still very difficult to answer the question: how much does artificial intelligence cost? How much cost will it take to develop, maintain, use, and improve it? Unfortunately, we do not know for sure. So, having machine learning as a certain element of Artificial Intelligence is much helpful in the attempt to estimate this cost and see how efficient it may be. Moreover, it is already improving our web and mobile services, so it seems to be quite promising and beneficial.
Introduction to Machine Learning
Therefore, what exactly machine learning is. It differs from the general concept of Artificial Intelligence. The latter is represented as a creation of some machine that can mimic a human's mind. The main idea of machine learning is building computer algorithms that will be able to improve themselves in time by detecting patterns in the information which already exist to move on without further control. So, such use reminds something like statistical analysis or data mining more than the classic AI.
All the artificial intelligence developers agree that the characteristic feature of machine learning is that the more accurate data in big amounts the algorithm combines, the more exact it becomes. The mechanism of machine learning is getting more and more efficient nowadays because the quantity of data is growing and our technological advance goes on ensuring further growth.
Therefore, this article is neither about the role and cost of Artificial Intelligence as a whole nor about machine learning and its uses. It is about the pros and cons of such a model as deepfake technology, its possible advantages, and dangers it can produce.
GAN: The Idea That Can Change the World
This idea was introduced in 2014 by Ian Goodfellow and co-authors. It stands for the overall conceptual progress of machine learning. In short, Generative Adversarial Networks (GANs) can produce or generate new content. As a good illustration of this model, its base, and its development, the process of generating random variables can be put forward. The Artificial Intelligence companies have proven that it is possible to generate some sequences of numbers whose characteristics will be very close to the characteristics of some theoretical sequences of random numbers.
A GAN consists of two sub-models. They are a generator model that helps to generate new samples and a discriminator model lending a hand in classifying and distinguishing between real generated examples from the domain and non-real (fake) examples that were made out by the generator model. These two models compete against each other like in a two-player game. This program typically works with image data, and it uses Convolutional Neural Networks, or CNNs, for its two models.
Using GANs as a deep learning method in research and data generation is called data augmentation. It is a computer vision that helps to perform certain patterns and reduce generalization errors. For example, GANs can generate photos of people, objects, or scenes that do not exist/have never existed in real life.
Opportunities Deepfake Technology Provides
Let us answer the question: what is deepfake? The base for this technology is Artificial Intelligence, of course. It is used when you want to produce or change some video content so that it could feature something that has never existed. The term 'deepfake technology' appeared in 2017 after Reddit users who applied this sort of deep learning technology to change the face of some famous people featured in pornographic clips.
The deepfake technology uses the two components of the GAN AI system mentioned above - a generator and a discriminator. First, a generator makes up a false clip and then sends it to a discriminator to make out whether the clip is true or fake. The generator and discriminator work together as a GAN in the constant fight with each other. The task of a discriminator is to distinguish the fake video, and the task of a generator is to create the video which will be impossible for the discriminator to distinguish as a fake one.
The good news is that to create a fake video, a person does not need any considerable skill or knowledge. It means that anyone with a computer utilizing some deepfake software can make it for some specific purpose. The bad news is that it can lead to the gap in trusting the validity of all the video clips available.
Most people are worried about the impact of deepfake images on society. However, many fears about technological advances are overstated. There is no reason for techno-panic about violent computer games or pulp novels, so they seem to have remained in the past.
Deepfakes, in their turn, can be used in many interesting ways, for scientific, educational, social, or media purposes. For example, GANs can create photorealistic imaginary personalities. The technology can also do many amazing things with pictures. For example, it can sharpen fuzzy images and colorize black-and-white photos.
This technological approach can also be used by scientists to create virtual molecules for chemical research. If done so, scientific and medical discoveries will be speeded up. While generating new molecules, scientists will be able to see what way these molecules can behave.
The Dangers of Deepfake Technology
Nevertheless, deepfake technology is rather dangerous, and it can affect politics, education, media, show-business, and other fields of human activity in the most negative way we can imagine.
The first thing to be worried about is using deepfakes in porn. It happens when a celebrity's face is artificially combined with porn actors' bodies. People are concerned about the possible mess such photos can make in political debate.
Another example is when a public figure in the video is presented as drunk or incapable. The technique for this is quite simple. The real video is slowed down to almost 75% of its natural speed, and the pitch of the voice is raised to keep it sounding authentic. This way, the slow speed of the image creates a weird effect of being drunk.
Most often, such videos with celebrities can produce dramatic effects. For instance, Mark Zuckerberg's deepfake, where he has announced his total control over all people's stolen information, became viral in a short time.
Moreover, spreading fake news, fake scientific research results, and survey statistics might make a disastrous influence on people's mental health and overall well-being. Many scholars and researchers, show-business personalities and politicians will be completely ruined, broken, and forgotten if they once get a powerful enemy or competitor, able to use this technology and determined enough to take revenge.
To conclude, deepfake technology has more cons than benefits. The governments of Great Britain and the USA have already made some attempts to monitor everything connected with using GANs.
Another way to prevent and combat the negative consequences of this model is making improvements in education. Digital literacy and critical thinking should be the focus of any school curriculum teaching children to spot fake news.
Of course, apart from official steps taken and explanations made, the core accent should be put on the morality of using deepfake technology. The social advertisement and public education representatives, religious and non-governmental organizations need to provide some time, cost, and effort on explaining the negative effects of deepfakes on the society, the ways of spotting fake information, avoiding such influences, and blocking them out from personal access if it is possible.
The straightforward, honest, and earnest communication among the representatives of authorities, business, media, education, and IT development will be able to overcome the negative effects of deepfakes and exploit their technical capability for the sake of progress, enjoyment, and mutual understanding.
