Deepfakes: A Prime Example of AI’s Creative Potential and Ethical Risks

Because October is Cybersecurity awareness month, I’d like to look at one of the greatest threats posed by artificial intelligence (AI): deepfakes. As you’ll know, this tech has been used to generate fake content apparently featuring politicians such as Donald Trump and Joe Biden or other prominent figures like Taylor Swift and Elon Musk. This material is so realistic and convincing that it can undermine the credibility of any Internet content – particularly audiovisual content.

But loss of credibility isn’t the only risk associated with deepfakes. They also figure in criminal activities, data theft, or scams that can cost companies significant sums . In fact, their threat is so great that Gartner predicts that by 2026, AI-generated deepfakes will lead 30% of enterprises to consider identity verification and authentication solutions to be unreliable in isolation. Let’s take a closer look at the risks of deepfakes – and at the opportunities they offer.

What Exactly Are Deepfakes?

Deepfakes have been a hot topic for some time. But it’s worth recapping what they are before looking at them in more detail. Essentially, deepfakes are synthetic media that accurately mimic a person’s appearance, voice, or behavior. A deepfake can consist of audio, visual images, or a combination of both (video).

The term deepfake comes from the deep learning algorithms used to create this bogus content. Deep learning is a machine learning (ML) method that leverages neural networks to analyze and interpret vast amounts of data to generate fake content that’s so realistic as to be indistinguishable from authentic sources.

Building a Deepfake: The Underlying Tech

Deep learning is underpinned by one of the following language models:?

• Generative Adversarial Networks (GANs) or
• Autoencoders

As discussed in an earlier blog, GANs are neural networks consisting of two main components – a generator and a discriminator. Autoencoders are another kind of neural network. They learn to compress and decompress data and are pivotal in learning how to imitate every aspect of an individual’s appearance – for example, their characteristic facial expressions and movements.

The realism of deepfakes is further enhanced through the use of facial mapping and tracking. Here, the movements and expressions of the real person’s face are analyzed in real-time and applied to the face of the fake person.

Deepfake Technology: Use Cases and Opportunities…

As mentioned, deepfakes are often used to create misinformation about politicians and other prominent figures, as well as in illegal activities. But that doesn’t mean the tech is inherently malicious or dangerous. It also offers considerable potential for innovation in a wide variety of areas.

In film and entertainment, for instance, it can be used to create special effects, de-age actors, and conceal the identities of vulnerable people in news reports or documentaries. In marketing and advertising, it can be deployed to create personalized advertisements. However, the use of deepfakes in the advertising sector can also raise concerns about scamming and fraud.

…Challenges and Risks

Scams and fraudulent activity aren’t the only issues associated with deepfakes. Privacy violations are another potential hazard. Not only can deepfakes spread misinformation and enable data theft; the individuals involved are often deepfaked without their consent.

But even where people have consented, deepfakes can misrepresent them and statements they’ve made. What’s more, AI-generated documents, photos, and biometric images can be illegally used for identity verification services. And in the finance sector, deepfake scamming can result in damages that run into the millions.

Regulatory Problems and Easier Access to the Tech

Deepfakes present many legal and regulatory difficulties. As in other fields of AI, the technology has rapidly outpaced existing legal frameworks, opening up regulatory gaps. To counter the associated risks, providers such as OpenAI and YouTube have already modified their regulations relating to AI-generated content.

In addition, the technology is now easier than ever to access. Not all that long ago, creating a deepfake called for a firm grasp of the technology and came with a hefty price tag. Today, almost anyone can quickly and easily create a “second self”: For as little as USD?15 you can now buy a fake AI-generated ID; USD?5 and one-minute-long voice sample enable criminals to realistically impersonate influential individuals, such as senior executives.

Safeguarding Yourself and Your Business Against Deepfakes

What’s particularly worrying is that some 65% of people don’t recognize deepfakes as such. This isn’t just because we don’t usually anticipate attacks of this kind, but also because we tend to trust familiar faces.

However, help is at hand. AI-powered cybersecurity systems can now detect shifts in network behavior, suspicious activity, and suspicious image, audio, and video material in the run-up to a potential attack.

To raise awareness of and provide education about the potential dangers of deepfakes, Accenture has launched First AI-iD Kit – a digital platform that offers practical tips on identifying, safeguarding against, and responding to deepfakes. In addition, it delivers accessible information designed to promote vigilance and responsible handling of AI technologies.

Seize the Opportunities – but Don’t Underestimate the Risks

Although deepfakes have legitimate uses, they must be considered a major cybersecurity risk. While the cost of creating a deepfake may be low (on average just USD 1.33), the tech’s potential to cause significant financial losses is very high.

As Generative AI gains traction, deepfakes are also on the rise: 2024 has seen a staggering 3,000% year-on-year increase in deepfake attacks. In light of these developments, every individual and company should take effective action to protect themselves against deepfakes. Fortunately, there are mandatory regulations and laws – for example, the EU AI Act –, which force providers and platforms to review and evaluate content more rigorously.

Meanwhile, in the Real World…

Do you have any deepfake stories – positive or negative – that you’d like to share? Then, leave a comment below.