Rapid advancements in Artificial Intelligence (AI) and the increasingly gendered nature of deepfakes have brought the spotlight back on Big Tech's policies and the Indian legal system's ability to tackle the problem.
"Technology-facilitated online gendered violence is a reality, and deepfakes are just another form of perpetuating such violence," Radhika Roy, advocate and associate legal counsel at Internet Freedom Foundation (IFF), told The Quint.
The need for stronger regulations around deepfakes resurfaced after an AI-generated deepfake video of actor Rashmika Mandanna recently emerged online. The video was created by morphing Mandanna's face onto the body of Zara Patel, a social media influencer.
The Quint spoke with experts in the field of technology, policy, and law to understand what safeguards are in place to protect celebrities as well as other people in case there is a deepfake video of them.
Rashmika Mandanna Deepfake Row: Why India Needs an AI Law Before It's Too Late
1. What Are Deepfakes & Why Are They Dangerous?
According to University of Virginia, deepfake is an artificial image or a series of images (i.e a video), that is generated by a specific kind of machine learning termed as "deep learning".
Deep learning is a machine learning technique that uses hidden layers in neural networks to replicate human brain learning. These layers perform mathematical transformations to convert input signals to output signals, such as creating really good fake images. Recursive neural networks are known for their effectiveness in image recognition tasks.
To understand why deepfakes have become common, we spoke with software engineer Smit Shah and he said,
"Deepfakes are very dangerous as a lot of data is available in the bandwidth. If someone has images to make videos, it becomes very easy. Earlier, in the television era, one required 24 frames per second. Now it's mostly 30. So you just need to render 30 images in motion to create a model that can be easily used to generate a fake video."
"People who are at risk are the ones easily available in the public domain; for instance, influencers and celebrities, whose videos, mannerisms, etc. are easily available on the internet. The reason why it's only going to increase is because now a lot of people are investing significantly in CI technologies," Shah added.
Expand2. The Gendered Nature of Deepfakes
Deepfakes have been used for a few years now to harass and intimidate women as well as drive traffic to websites containing malicious content. Deepfake pornographic content of celebrities are sold online and have become a million dollar industry. At the same time, deepfakes are also used to blackmail and exploit individuals who are not in the public eye.
"Deepfakes have been an increasing area of concern with the developments in AI. They are being used to spread misinformation, disinformation, harass, intimidate, create pornographic images, and in several other ways to undermine people," Mishi Choudhary, founder of the Software Freedom Law Centre (SFLC), said.
Research done on pornographic deepfake content by Amsterdam-based identity verification company Sensity AI showed that 96 percent of deepfakes available online between 2017-2020 were sexually explicit and were created without the consent of the women.
Speaking on the gendered nature of deepfakes, Roy told The Quint, "Deepfakes are becoming increasingly gendered, but that’s how it’s been in the country. Technology-facilitated online gendered violence is a reality, and deepfakes are just another form of perpetuating such violence."
Choudhary added that the research on deepfakes has inadvertently ended up making deepfake technology better.
She also pointed out the crucial role that fact-checkers play at such a time.
Expand3. Can Celebrities Safeguard Themselves?
According to Roy, although the term "deepfake" hasn't been defined explicitly in any statute, there are legal provisions to tackle it, such as Section 67 of the IT Act, "which can be used for publishing obscene material in electronic form".
"More specifically, we have Rule 3(1)(b)(ii) which prohibits defamatory, obscene, pornographic, paedophilic, invasive of privacy, including bodily privacy, insulting or harassing on the basis of gender, etc. As per the Rule, intermediaries are required to expend reasonable efforts to ensure that users do not “modify” any information that is prohibited," Roy added.
Roy said that while people can also resort to provisions under the IPC, they are not enough.
"There needs to be clearer explanations with clearer consequences when it comes to deepfakes, given their insidious nature and how it’s sometimes impossible to tell what’s real from what’s fake."
Radhika RoyChoudhary concurred with Roy's view and said, "Some resources are being worked upon to give help to individuals, as the law has not kept up with it. Our police forces are not trained, nor are our judges or courts".
According to Union Minister of State for Electronics and IT Rajeev Chandrasekhar, social media platforms are legally obligated to remove misinformation within 36 hours once reported by a user or the government under the IT rules notified in April this year.
“If platforms do not comply wth this, rule 7 will apply and platforms can be taken to court by aggrieved person under provisions of IPC,” he said, adding that “deepfakes are latest and even more dangerous and damaging form of misinformation and needs to be dealt with by platforms”.
Expand4. How To Identify Deepfakes?
The first step is to pay attention. Even in actor Rashmika Mandanna's viral video, upon closer inspection, the face of the original person can be seen in the video's initial few frames.
Secondly, verify the information, just like we verify WhatsApp forwards.
Have a healthy skepticism. Don't believe what you see. One can refer to fact-checking platforms that verify the information for users.
Choudhary talked about the Detect Fakes website created by Massachusetts Institute of Technology (MIT) that helps people identify deepfakes.
Lastly, Shah spoke about how advanced deepfake-detection systems look for biometrics. They try to understand the gestures, heartbeat, how often the eyes are blinking, the subtle movements, and the sounds that might have been picked up in the microphone, etc.
(At The Quint, we question everything. Play an active role in shaping our journalism by becoming a member today.)
Expand
What Are Deepfakes & Why Are They Dangerous?
According to University of Virginia, deepfake is an artificial image or a series of images (i.e a video), that is generated by a specific kind of machine learning termed as "deep learning".
Deep learning is a machine learning technique that uses hidden layers in neural networks to replicate human brain learning. These layers perform mathematical transformations to convert input signals to output signals, such as creating really good fake images. Recursive neural networks are known for their effectiveness in image recognition tasks.
To understand why deepfakes have become common, we spoke with software engineer Smit Shah and he said,
"Deepfakes are very dangerous as a lot of data is available in the bandwidth. If someone has images to make videos, it becomes very easy. Earlier, in the television era, one required 24 frames per second. Now it's mostly 30. So you just need to render 30 images in motion to create a model that can be easily used to generate a fake video."
"People who are at risk are the ones easily available in the public domain; for instance, influencers and celebrities, whose videos, mannerisms, etc. are easily available on the internet. The reason why it's only going to increase is because now a lot of people are investing significantly in CI technologies," Shah added.
The Gendered Nature of Deepfakes
Deepfakes have been used for a few years now to harass and intimidate women as well as drive traffic to websites containing malicious content. Deepfake pornographic content of celebrities are sold online and have become a million dollar industry. At the same time, deepfakes are also used to blackmail and exploit individuals who are not in the public eye.
"Deepfakes have been an increasing area of concern with the developments in AI. They are being used to spread misinformation, disinformation, harass, intimidate, create pornographic images, and in several other ways to undermine people," Mishi Choudhary, founder of the Software Freedom Law Centre (SFLC), said.
Research done on pornographic deepfake content by Amsterdam-based identity verification company Sensity AI showed that 96 percent of deepfakes available online between 2017-2020 were sexually explicit and were created without the consent of the women.
Speaking on the gendered nature of deepfakes, Roy told The Quint, "Deepfakes are becoming increasingly gendered, but that’s how it’s been in the country. Technology-facilitated online gendered violence is a reality, and deepfakes are just another form of perpetuating such violence."
Choudhary added that the research on deepfakes has inadvertently ended up making deepfake technology better.
She also pointed out the crucial role that fact-checkers play at such a time.
Can Celebrities Safeguard Themselves?
According to Roy, although the term "deepfake" hasn't been defined explicitly in any statute, there are legal provisions to tackle it, such as Section 67 of the IT Act, "which can be used for publishing obscene material in electronic form".
"More specifically, we have Rule 3(1)(b)(ii) which prohibits defamatory, obscene, pornographic, paedophilic, invasive of privacy, including bodily privacy, insulting or harassing on the basis of gender, etc. As per the Rule, intermediaries are required to expend reasonable efforts to ensure that users do not “modify” any information that is prohibited," Roy added.
Roy said that while people can also resort to provisions under the IPC, they are not enough.
"There needs to be clearer explanations with clearer consequences when it comes to deepfakes, given their insidious nature and how it’s sometimes impossible to tell what’s real from what’s fake."Radhika Roy
Choudhary concurred with Roy's view and said, "Some resources are being worked upon to give help to individuals, as the law has not kept up with it. Our police forces are not trained, nor are our judges or courts".
According to Union Minister of State for Electronics and IT Rajeev Chandrasekhar, social media platforms are legally obligated to remove misinformation within 36 hours once reported by a user or the government under the IT rules notified in April this year.
“If platforms do not comply wth this, rule 7 will apply and platforms can be taken to court by aggrieved person under provisions of IPC,” he said, adding that “deepfakes are latest and even more dangerous and damaging form of misinformation and needs to be dealt with by platforms”.
How To Identify Deepfakes?
The first step is to pay attention. Even in actor Rashmika Mandanna's viral video, upon closer inspection, the face of the original person can be seen in the video's initial few frames.
Secondly, verify the information, just like we verify WhatsApp forwards.
Have a healthy skepticism. Don't believe what you see. One can refer to fact-checking platforms that verify the information for users.
Choudhary talked about the Detect Fakes website created by Massachusetts Institute of Technology (MIT) that helps people identify deepfakes.
Lastly, Shah spoke about how advanced deepfake-detection systems look for biometrics. They try to understand the gestures, heartbeat, how often the eyes are blinking, the subtle movements, and the sounds that might have been picked up in the microphone, etc.
(At The Quint, we question everything. Play an active role in shaping our journalism by becoming a member today.)