A group of fact-checkers, academicians, tech experts and policy experts came together for a panel discussion on ‘Decoding Deepfakes & Elections’ organised by BOOM’s Decode held in New Delhi on 15 April 2024.
The Consensus
- Deepfakes is a worrying trend.
- People generating mis/disinformation are only limited by their imagination since it's extremely easy to create fake stories.
- The technology used to create false information is better than the one used to detect it.
- Investment is more in the market generating deepfakes than the one detecting it.
Mayank Vatsa, professor of computer science, Indian Institute of Technology (IIT), Jodhpur noted that the technology behind deepfakes isn’t the problem but the intent with which it is being used is.
He further said:
“We, as human beings, like to generate new knowledge, new videos, new audios. ChatGPT & DALL-E have enabled the creation of new knowledge to a much larger extent. That is triggering more and more deepfake generations. The technology has evolved to a level where the fact-checkers have to be dependent on technology to detect it.”
Meanwhile, Karen Rebelo, deputy editor, BOOM asked a pertinent question, “How do we watermark reality? The bigger problem is, as humans, we want to take the smart option. The smart option right now is to doubt everything and that is causing dissonance in how people are consuming media online.”
The Problem of Gendered Misinformation
Women are often targeted with false narratives, altered or morphed visuals and now deepfakes too. Decode had earlier reported on how rapid advancements in artificial intelligence (AI) technology has turbo-charged the spread of deepfake pornographic videos on mainstream social media platforms.
Kanksshi Agarwal, founder, Netri Foundation said, “When we look at the nature of deepfakes, people still don’t know that it could be harmful. Politics to begin with is dirty, but women are struggling to get in and deepfakes are a clear deterrent.”
She suggested that one of the ways in which women can tackle this problem is by equipping them with fact-checking tools and techniques.
“Imagine I go to a constituency and there are 15 or 18-year-old boys with these easy apps on their phones and they don’t even need coding etc. They can record me then and there while I am on the stage in the most modest clothing. They can create a very altered or tampered version of it. They can clone my voice and start circulating it. If we change the narrative and we go back to equipping women to counter this effectively which is to say, ‘So what? This is fake and I can prove it to you. I can prove it to you by using a fact-checking app on my phone.’ A fact-checking app on a politician/ campaigner’s phone is a must in the next five years,” she added.
Deepfakes & Elections
The fact-checking team at BOOM is working round the clock to debunk election related mis/disinformation.
During the Madhya Pradesh Assembly elections held last year, several videos of former chief minister and Bharatiya Janata Party leader Shivraj Singh Chouhan and Congress leader Kamal Nath were circulating on social media with false claims.
“Videos themselves were true. So, if you did a reverse image search, you would find that these are old videos of politicians but the audio track in them had been tampered with. So, they were voice clones of Shivraj Singh Chouhan mainly and few of Kamal Nath,” Rebelo said.
She also elaborated on how fact-checkers were split between fact-checking a war on one hand and election-related disinformation on the other as the Israel-Hamas conflict had broken out at the same time.
“So, at that point of time, we didn’t have the tools or expertise to detect whether it was a voice clone and with voice it can be tricky because it could be AI-generated, or a voice clone or it could also be somebody who is mimicking someone else’s voice. What if you take all old speeches and put them together to make a completely new one?,” she added.
BOOM has fact-checked multiple audios of celebrities being circulated on the internet which are actually their voice clones. The voice clones are also being used to create fraud advertisements to scam social media users.
Meanwhile, Agarwal spoke about the challenges faced by politicians and the need for social media platforms to tweak controls.
She said, “Politicians will not close their profile accounts. They want their DMs (direct messages) to be open. So, people will slide into their DMs. How do we then look at: that they want to be accessible to the people, which essentially means that their images are up for tampering. The controls within the platforms need to change.”
Deepfakes: Law & Investments
Vibodh Parthasarathi, associate professor, Centre for Culture, Media and Governance, Jamia Milia Islamia University mentioned that there is more investment in the markets that generate false information and less in the markets that detect it.
“Information markets traditionally can include actors in the news business. Then, there is an emergent market which produces half-baked truth. The second market is actually old. Actors in this market are just as institutionalised as the other and no longer amateurs,” he added.
Speaking about regulations and legal framework for deepfakes, Sanhita Chauriha, tech lawyer, Vidhi Centre for Legal Policy, stated that there isn’t any specific law for it.
“With law, we are not always sure about what problems will crop up in the future. Hence, there will always be a gap. We have new advisories from the Ministry of Electronics and Information Technology and overlapping sections of the IPC. But there is no law specifically for deepfakes,” she said.
The event also saw Prof Mayank Vatsa demonstrating Itisaar, a deepfake analysis tool created by IIT Jodhpur and Amitabh Kumar, co-founder, Social Media Matters showcasing Contrails.ai, a tool focusing on problems fact-checkers face while debunking mis/disinformation. Decode’s Archis Chowdhury took the audience through some AI and deepfake detection tools while Titha Ghosh showcased the work Decode has done in the field of deepfakes and its real life impact.
How Can You Spot AI-Generated Visuals? Quick Tips:
- Observe the visual carefully especially if it has human features.
- Is there an unnatural gaze or lighting?
- Look for blurred or pixelated elements in the background.
- Some publicly available tools that can help you detect if it's AI-generated or not: Hive AI Detector, AI or not, Loccus. AI (paid tool)
Disclaimer: These tools are not always 100% accurate.