What Is The Difference Between A Deepfake And Shallowfake?


What Is The Difference Between A Deepfake And Shallowfake?

While deepfakes are commonly perceived as a major technological development of concern, less is known about so-called ‘shallowfakes’. An interesting term that differentiates itself from the machine learning phenomenon of deepfake technology.

Understanding the difference between deepfakes and shallowfakes is an important part of understanding the problems in Artificial Intelligence (AI) with regards to manipulated media content as a whole. This article aims to shed some light on those distinctions and will further explore this new phenomenon known as ‘shallowfakes’. First, let us explore what this term actually means.

What Is A Shallowfake?

A shallowfake, as opposed to a deepfake, is a method of manipulating media content without the use of machine learning technology and algorithmic systems. Instead, shallowfakes utilize simple video editing software to alter existing media content.

Shallowfakes do not require the use of deep-learning systems, which make them fundamentally different from the better-known deepfake phenomenon.

Note the removal of the word ‘deep’ in the term ‘shallowfake’. With the exclusion of deep-learning systems from the process of media manipulation, the term ‘shallow’ would implicate an opposite (yet similar) method of creation.

The fact that shallowfakes aren’t fabricated with the use of AI does not make them notably different in terms of quality or quantity, compared to the more commonly known ‘deepfakes’. It merely indicates how the content was produced and which types of technologies were avoided in the production process (i.e. the avoidance of deep-learning AI).

Differences Between Deepfakes And Shallowfakes

Having a good overview of the terminology helps us humans understand them better, but most people will actually start understanding the concepts best when you directly place the two concepts against each other.

So when we look at the main differences between shallowfakes and deepfakes, the following factors of interest emerge to the surface immediately (pun intended):

  • Method of production: The clearest differentiation between the two concepts can be made within the production method. While shallowfakes apply general editing software and manual adaptation of pre-existing media content, deepfakes will apply algorithmic changes to pre-existing media content by feeding data to a computer program. So while the former is essentially based on manual human labor, the latter is based on automated iterations of a machine learning tool (i.e. a computer is doing all the work for us humans). This obviously also impacts the efficiency and speed of producing ‘faked’ media content pieces.
  • Involvement of deep machine learning: The very term ‘shallow’ implies the absence of deep machine learning efforts in the methods applied. So while shallowfakes do not require the use of deep learning tech, their deepfake counterparts absolutely will use these types of tools in the process of creating the falsified media content.
  • Involvement of artificial intelligence: While the creation of a shallowfake does require a tremendous amount of skill and effort, as well as a piece of software that enables a human to actually create the falsified media content, this process is arguably still manual in nature. It requires a human being to be completed, thus it is not artificial. On the other hand, deepfakes are automatically generated by a subtype of artificial intelligence, which would be the deep learning algorithms that are needed to develop the falsified media content.

Understanding the key differences between the concepts is the first step to combating them. The method of production of falsified media content is the most important aspect in this regard. Regardless of this, the similarities that can be found are striking and should also be addressed to gain an understanding of what we are dealing with. The next section will expand upon the most important similarities between the two closely related concepts.

Similarities Between Deepfakes And Shallowfakes

When you have two closely related concepts, the similarities are often easy to spot and don’t need much extensive explanation. So without further ado, here are what we believe to be the main similar features of the shallowfake and deepfake concepts:

  • The goal is always to twist reality: Take a real piece of media content, put it in a blender, and spit out something new. In its core, this is what deepfakes and shallowfakes do. They will alter reality in some way, shape or form. This intent does not always need to be malicious, it can also be used for entertainment purposes (e.g. letting celebrities say or do anything you desire).
  • Both concepts utilize pre-existing media content: Regardless of the method used, the general rule of thumb is to use some type of image, video, or audio content and take that as the basis for the changes you’d like to implement. Shallowfakes place this in certain types of editing softwares, while deepfakes use algorithmic iterations to forge new types of media content creations.
  • Both can be a tool in a misinformation campaign: With the emphasis on ‘can’, because they can also be used for other intents and purposes. Misinformation campaigns will benefit equally from a shallowfake as they would from a deepfake. The main difference being the means by which they are created.
  • Both should be utilized with caution: When placing powerful manipulation tools like these software classes into the wrong hands, both could be dangerous and a major source of misinformation. They should be handled with care and not fall into the hands of criminals.

The above list of similarities is not exhaustive, but covers some essential aspects of the concepts discussed. That said, the question remains which one of the two concepts poses a greater threat to global societies right now, and further into the future.

Is A Shallowfake More Worrysome Than A Deepfake?

Both shallowfakes and deepfakes pose a certain danger when the tools fall into the wrong hands. Given the fact that we are dealing with often widely available software pieces that are distributed through the internet, it is safe to say that both concepts pose an immediate and severe threat to societies at large.

Comparing the methods of production, which is the main difference between shallowfakes and deepfakes, logic would dictate that identification of algorithmic changes are easier to create. Algorithms work with a certain pattern, and patterns can be detected by deepfake detection software.

On the other hand, the creation of shallowfakes requires some form of manual labor or manor intervention at some point during the creation process. This would logically mean that the time it takes to create a single piece of falsified media using this conceptual method would take considerably more time than if an AI would auto-generate the content for you.

Nevertheless, manual labor isn’t based on patterns as much, which would make this creation method much harder to detect by a detection software. At least, when they are created professionally and with care. Otherwise it might even be possible to spot incoherencies with the naked eye.

Combating Falsified Media Content

Regardless of the production method, faked media content will remain a serious worry for the global societal system as a whole. When a single person with access to the internet and a computer can impact many thousands of lives with a few software tricks, it is clear people should combat that situation with every tool possible.

While the fight against shallowfakes and deepfakes is showing some promising progress, some experts fear that it will never be enough to catch up to the latest piece of technology that might fool the public into thinking something is real.

And perhaps that’s not even what our main weapon should be against these falsified media pieces. Educating the public and learning young children to remain critical of what they are told, telling kids to keep asking questions wherever possible, and simply verifying claims with multiple sources will make the world of difference. Whether someone is faced with a deepfake or a shallowfake, this will be the key to sticking to the truth as much as humanly possible.

Post a Comment