How to Prevent Deepfakes in The Age of Generative AI: Facia’s Expert GuideAuthor: Luke Oliver | 02 Oct 2023
In This Post
What are deepfakes?
Deepfakes replace an original face or voice in a video with a fabricated one, this is the simplest way to put it. It makes use of advanced machine learning and artificial intelligence techniques and has become increasingly common around the world.
The technology has interesting applications like personalised avatars or video dubbing but it also poses risks such as identity theft, misinformation, and fraud. This guide talks about the negative consequences of the widespread use of deepfakes and provides insight into how they can be identified and tackled.
Facia specialises in identifying and mitigating the risks of deepfakes, but we believe that not everyone has that level of tech awareness to understand the overall complexity of this media. To help everyone understand, we have put together an easy-to-understand guide to help you recognise a deepfake. Let’s first discuss what makes them so dangerous.
Deepfakes and Generative AI
Generative AI has transformed how the online world operates. Generative Artificial Intelligence Techniques are becoming more sophisticated day by day, and it has placed generative technology into the hands of any ordinary internet user. The implications are vast, as people are able to generate content with a basic prompt. This paves the way for increasing cybercrime activities, including the widespread use of deepfakes.
What Makes Deepfakes Dangerous?
Deepfakes are powered by artificial intelligence and machine learning algorithms that can generate incredibly lifelike videos. This technology can potentially be misused to spread false narratives, carry out scams, or ruin reputations.
Thus, learning to identify deepfakes becomes critical for individual and collective security. The contemporary world of information and technology entails a significant exchange of information. Hence, it becomes imperative on how this information can be safeguarded.
How to Prevent Deepfakes?
The first thing you notice in any video or live feed is usually the visuals. The visuals catch your attention before audio or any other contextual cues. Deepfakes do a great job of replicating exact surrounding visuals, but there are plenty of ways you can detect differences. The key here is attention to detail and focusing on certain features.
Artificially generated videos often get the blinking wrong. Focus on specific blinking movements to detect whether it’s a live person or a deepfake.
Pay close attention to the synchronisation of mouth and voice. AI does not get the timing right, and unnatural mouth movements can be detected easily.
Lighting and Shadows:
Deepfake algorithms may have a hard time replicating natural lighting. Look for inconsistencies in lighting and shadows on the face or background.
If facial expressions seem too rigid or emotionless, it’s probably a good idea to investigate further and see if it is a deepfake. Where AI can get most emotions to replicate, sometimes it misses out on accurate expressions.
Check for distortions or visual inconsistencies around the face, hair, or background. These may be subtle but are usually present unless it’s a very high-quality deepfake.
Deepfake algorithms can generate physically impossible characteristics, like a third eye or misaligned features. Such anomalies however are pretty evident to the naked eye.
Sound distortion or an unnatural voice can often indicate what’s wrong with a video or live feed. Deepfakes generate very high-quality visuals but their audio quality isn’t up to the mark just yet. Here are a few things you can focus on.
Pay attention to any unnatural pitch or modulation in the voice. It may indicate that the original audio has been manipulated. An inconsistent pitch after regular intervals or an unusually high-pitched sound could be indications of a deepfake.
In many cases, the audio won’t perfectly sync with the video. This works both as a visual and hearing cue to detect a deepfake.
Deciphering Truth: Using Contextual Clues to Unmask Deepfake Videos
To reiterate, deepfakes are getting better with time. If you’re unable to detect any irregularities, you need to conduct a deeper analysis of the video’s premise. If the video does not fit well with the person’s opinions or viewpoints, it could be that it is a deepfake.
A person who clearly opposes the use of cryptocurrency suddenly decides to ask for a $1000 investment, and that too on a time urgency.
The above case is drawn from an actual example where a student had his account hacked, and a deepfake was generated to encourage others to invest. One of his close friends rightly pointed out in a group chat that he never liked cryptocurrency, and it looks fishy how he is suddenly advocating the whole concept.
Pay close attention to the fact that the person in the video would likely be in the given situation, saying or doing what’s presented. This is an ideal way to question the authenticity of the video.
Always consider the reliability of the platform or source from where you found the video. This is true for most news-generated videos as well. Clickbait platforms that focus on quicks sensationalise a lot of news elements to gain traction. However, such platforms could go a step further by releasing deepfakes and inciting people to fake news.
Advanced Methods: Metadata and Reverse Image Search
Metadata can reveal clues about the origin and modification history of a video file. There are free tools available for this or you can contact experts and get a paid service.
Reverse Image Search
Extract frames from the video and use them in a reverse image search to see if they appear elsewhere on the internet. If they do, check out the source and notice any unusual links.
When in Doubt, Seek Expert Help
For critical content like legal, political, or highly sensitive videos, it may be best to consult experts for deepfake detection. If you’re worried about your business being affected, implement a liveness detection system.
Facia is currently the world’s fastest liveness detection platform with a response time of less than a second.
With the evolution of technology, deepfakes are also becoming harder to detect. However, there are a significant number of organisations and regulatory bodies that focus on deepfake detection to protect individual privacy.
You can also stay vigilant by following this guide and noticing each element of footage closely in case you suspect it’s a deepfake. Everyone has a role to play in combating misinformation and the potential abuse of deepfake technology.
*Disclaimer: The above guide serves as a starting point for recognizing deepfakes and should not be considered as foolproof. Advanced deepfakes may require professional analysis for accurate detection.*
Frequently Asked Questions
AI deepfake technology utilizes advanced deep learning methods, specifically generative adversarial networks, to create convincing simulations of real individuals. The increasing sophistication of this technology has led to a rise in cases, where deepfakes are used for malicious purposes, making them harder to detect and easier to generate.
Liveness detection in biometrics refers to the ability of a biometric system to distinguish between a live genuine sample and a fake or spoofed sample. It ensures that the biometric data being presented during a verification or identification process is from a living person rather than from an artificial source.
Deepfake examples include the video by Jordan Peele, manipulating Barack Obama's footage to issue a warning about deepfakes. Another is the fabricated video of Facebook's CEO, Mark Zuckerberg, falsely boasting about data control. Such instances underline deepfake capabilities and the necessity for discernment in today's digital age.
Liveness detection analyzes facial features and movements in real-time to determine if the subject is a live person or a digital representation. Deepfakes, being pre-recorded or computer-generated visuals, fail to exhibit genuine human micro-movements or respond to liveness prompts, making them detectable and distinguishable from authentic human presence.
Deepfakes have become increasingly sophisticated, often deceiving the naked eye. Liveness detection offers an added layer of security by ensuring that the person in front of the camera is genuine and currently present. This technology effectively combats identity fraud attempts that utilize deepfake videos or images.