The police urge scam victims involving fake videos, images, or deepfakes created with artificial intelligence (AI) technology to come forward and lodge a report to facilitate further action as only victims can verify if their faces and voices were used in fraudulent activities.
Bukit Aman Commercial Crime Investigation Department Director Datuk Seri Ramli Mohamed Yoosuf noted that such scams are becoming more frequent and are likely to continue.
Bernama reported him saying that deepfakes are particularly challenging to tackle because they require verification from the victims.
For instance, when it involves political leaders, only the individual concerned can verify or deny their involvement. He added that the police have not received any reports about such cases so far.
Based on his observations, Ramli said the public is often deceived by AI-generated deepfakes, especially when influential persons like politicians, businessmen, and celebrities are involved.
Creating videos using the images and voices of these individuals can result in commercial crimes.
For example, there may be videos featuring influential individuals urging people to invest in dubious schemes. The spread of deepfake videos can cause unrest and disharmony if not addressed early.
Ramli urged the public to exercise caution and verify the authenticity of any information received to avoid falling victim to such scams.
Communications Minister Fahmi Fadzil also urged the public to be wary of AI-generated deepfakes after popular songstress Datuk Seri Siti Nurhaliza fell victim to these tactics, where scammers are using AI technology to imitate her voice and face for profit.
He acknowledged that while AI technology aids the country’s development, some parties are misusing it.
Fahmi also mentioned that he will meet with representatives from all social media platforms in Singapore later this month to discuss several issues, including the dangers of AI on their platforms.
How AI deepfakes work?
Deepfakes are created using advanced artificial intelligence techniques, primarily involving deep learning algorithms.
They can be used for various purposes, from entertainment and art to more malicious uses like misinformation or harassment.
Here’s a simplified breakdown of how they work:
- Data Collection: Large datasets of images, videos, or audio clips of the target person are collected. This data is used to train the AI model.
- Training the Model: The AI uses a type of neural network called a Generative Adversarial Network (GAN). GANs consist of two parts:
- Generator: This part creates fake images or videos.
- Discriminator: This part evaluates the generated content and provides feedback on its realism.
- Face Swapping: For video deepfakes, the AI maps the facial features of the target person onto another person’s face. This involves:
- Facial Recognition: Identifying and tracking facial features in the source video.
- Face Replacement: Superimposing the target person’s face onto the source video, adjusting for expressions and movements.
- Voice Mimicking: If audio is involved, the AI analyses the target person’s voice patterns and generates new audio that mimics their speech.
- Refinement: The generated content is refined through multiple iterations, with the discriminator providing feedback to improve realism.
