What are the different types of deepfakes and how you can spot them?
The deepfake controversy involving Indian celebrities has sparked a discussion around the urgent need for AI regulations and safeguards.
- Deepfakes come in various types from visual to vocal and even text-based ones
- Various types of manipulated media content have raised concerns about misinformation and its societal impacts
- Identifying a deepfake can be challenging but there are some key indicators to look out for
The realm of deepfakes is an ever-evolving landscape that introduces various types of manipulated media content, raising concerns about misinformation and its societal impacts.
Earlier in the month, a deepfake video emerged featuring Rashmika Mandanna’s face digitally altered onto British-Indian social media figure Zara Patel. Recognisable to those versed in deepfake technology due to its uncanny nature, the video gained significant traction due to Rashmika’s widespread popularity across India, her vocal stance against deepfake manipulation, and even drew attention from Prime Minister Narendra Modi, who expressed concerns. Amidst the controversy, one positive outcome was the demand by many to engage in global discussions about AI and the regulation of its human application.
“One of the primary concerns surrounding deepfakes is their potential to impersonate authorised personnel and manipulate financial transactions. By creating deepfake videos or audio recordings of executives, fraudsters can trick employees into approving fraudulent transactions or releasing sensitive financial information,” Founder of Zeron Sanket Sarkar told Business Today.
Here’s an extensive look at different types of deepfakes:
1. Face-swapping deepfakes: This is the most common form, where the face of one person is superimposed onto another’s using deep learning algorithms. It can be done convincingly, making it appear as if the target person is saying or doing things they haven’t.
2. Voice synthesis: Generating synthetic voice recordings that mimic someone’s speech patterns and intonations. These can be used to create fake audio messages or mimic someone’s voice to an eerily accurate degree.
3. Gesture and body movement manipulation: Deep learning techniques can also alter body movements, gestures, and expressions in videos, making it seem like a person is doing or saying something they didn’t.
4. Text-based deepfakes: AI-generated text, such as articles, social media posts, or even emails, that imitate the writing style of a specific individual, potentially leading to misleading content creation.
5. Object manipulation: Beyond faces and bodies, deep learning can manipulate objects within videos, changing their appearance or behaviour. This could lead to deceptive presentations of events or situations.
6. Hybrid deepfakes: Combining multiple techniques, such as face-swapping with altered voice or body movements, to create more sophisticated and convincing deepfakes.
7. Malicious use cases: Deepfakes have been employed in various malicious activities, including revenge porn, political manipulation, spreading hoaxes, and financial fraud.
Understanding the different types of deepfakes is crucial for recognising and addressing their potential negative impacts. The proliferation of this technology calls for robust countermeasures, including:
– Detection tools: Developing algorithms to identify and flag manipulated content. AI and machine learning models are being created to detect discrepancies in videos, audio, and texts.
– Regulatory measures: Policies and laws aimed at curbing the misuse of deepfake technology. Governments and tech companies are working to establish guidelines and regulations to govern its creation and dissemination.
– Awareness and education: Educating the public about the existence and potential risks of deepfakes can help people become more discerning consumers of online content.
– Technological advancements: Advancing technologies to not only detect deepfakes but also to create digital watermarks or authentication systems that verify the authenticity of media content.
Identifying a deepfake can be challenging, but there are some key indicators to look out for:
1. Inconsistencies or Oddities: Watch for inconsistencies in facial features, unnatural movements, or irregularities in the background.
2. Unnatural Facial Expressions: Deepfakes might display strange facial expressions or unnatural eye movements that don’t align with the audio or context.
3. Blurry Edges or Artefacts: Pay attention to blurry edges around the face or body, or noticeable artefacts that appear out of place.
4. Abnormal Audio Sync: Sometimes, the audio might not sync perfectly with the lip movements, or there might be glitches in the voice.
5. Contextual Anomalies: Assess the context of the video. Does it seem plausible? Is there anything in the background or actions that seem out of place?
6. Source Verification: Always try to verify the source of the video or image. Check multiple reliable sources or look for corroborating evidence.
7. Use Technology: Some software or online tools are designed to detect deepfakes by analysing various elements in the video.
8. Consult Experts: If in doubt, consulting experts in digital forensics or AI-generated content can help determine the authenticity of the media.
