Wednesday 22 November 2023

DEEPFAKE CLEVERLY REBRANDED AS GENERATIVE AI

 



Only a few weeks ago I wrote a blog on the possible misadventures of Artificial Intelligence (AI). I told you how celebrities like movie stars, sportspersons and politicians, who have iconic attributes like looks, style, voice, singing abilities and mannerisms are at risk of being both misused and monetized by AI.  To read it you can click: https://surajitbrainwaves.blogspot.com/2023/10/now-no-more-fake-jhakkaas-and-dhai-kilo.html


Generative AI's power lies in its ability to create content including text, images, code etc, as well as offering services as seamless translation. The algorithms are trained on massive data to help them decode patterns, which they deploy to create content. So, a text prompt can an image of say 'purple parrot eating yellow chilies while driving a Ferrari.'


This can look fantastic, but it also has multiple dangers. Generative AI can be used for scams by creating realistic fake video and cloning voices. Now imagine Joe Biden or Donald Trump giving the speech of Hamas leader Yahya Sinwar in their original but cloned voice where they are criticizing Israel and the U.S and promising free Palestine 'from river to the sea!' What do you think will be the effect of this video on the next Presidential election? This is the power of Deepfakes and it will be impossible to separate them from genuine particularly by the electorally charged common masses. And all this mischief can be done from a nondescript AI platform in Xinjiang!


Deepfakes are not new, they have been around since 2016. Though initially it was just a joke, the technology has improved tremendously, and is doing so almost every week. It started with photoshopped face swaps but now we have lip sync deepfakes. Snapchat, FaceApp, and similar apps have popularized face-swapping, which is the basis for deepfake technology. The very term ‘deepfake’ comes from a Reddit user, who in early days of this technology, created porn. Whoever changed this name to Generative AI not only successfully rebranded it but gave it the respectability it never had.


Deepfakes are manipulated images or videos created using machine learning (ML) and artificial intelligence (AI) algorithms to replace someone’s face or voice with another person’s, creating a new identity. Basically, deepfake technology utilizes deep learning algorithms that can learn how to solve problems by analyzing large amounts of data to make alterations. The algorithms enable the creation of synthetic media that can mimic a person’s appearance and voice with remarkable accuracy.


Thus, an audio can be created by synthesizing the voice of Donald Trump and then the mouth of Donald, standing on the podium or sitting behind his desk can be changed and you have a lip synced image of Donald saying whatever rubbish you want him to say! Ok, I agree it was not the best example, because Donald has uttered the craziest possible things in the past, but you certainly would be surprised if instead of ‘America first’ he is made to say ‘China first’ or ‘Palestine first’!

 

Detecting Deepfakes

Detecting deepfakes is getting more difficult as the technology that creates deepfakes is getting more sophisticated. In 2018, researchers in the United States demonstrated that deepfake faces didn’t blink like humans do, which was considered a great way to detect if images and videos were fake or not. However, as soon as the study was published, deepfake creators started fixing this, making it even more difficult to detect deepfakes. Oftentimes, the research that is designed to help detect deepfakes just ends up helping make deepfake technology better. Just like every anti-virus programme for computers is outsmarted by a new set of viruses, same is the situation with deepfakes.


However, not all deepfakes are products of sophisticated technology. Poor-quality material is usually easier to detect, as the lip syncs may not match well or the skin tone may seem odd. Additionally, details like hair strands are often harder for deepfake technicians to create. Studies have also shown that jewelry, teeth and skin that create erratic reflections can also reveal deepfakes. Here are some of the top methods to identify a deepfake:

  1. Look for inconsistencies: Deepfakes often have inconsistencies in lighting, shadows, and reflections that are not present in real media.
  2. Analyze the audio: Pay attention to the audio in the video. Deepfakes may have unnatural or inconsistent audio quality, such as variations in background noise or voice pitch.
  3. Check for unnatural movements: Deepfakes may have artificial movements that are not typical of real people. Look for glitches or distortions in the video that do not match natural movement patterns.
  4. Compare with known sources: Deepfakes are often created by merging two or more existing videos or images. By comparing the deepfake with known sources, you can identify discrepancies.
  5. Use deepfake detection tools: Several tools have been developed to help identify deepfakes. These tools analyze various aspects of the media, such as facial expressions, eye movements, and skin textures, to detect if the media is manipulated. Microsoft’s Video Authenticator detects blending boundaries and grayscale elements invisible to the human eye, while Facebook’s Reverse Engineering detects fingerprints left by an AI model. 

 

Why are celebrities and politicians the usual targets?

To produce a convincing deepfake video, two machine learning models are utilized: one generates fake videos from a dataset of sample videos, and the other identifies whether the video is real or fake.  Deepfakes are often enhanced by Generative Adversarial Networks (GANs), another type of machine learning. The GAN technique trains these two models to compete against each other until the second model can no longer distinguish between real and fake videos. The outcome is a deepfake that appears realistic to human viewers. Now, GAN is most effective in generating realistic deepfakes when a vast dataset is accessible for it to learn from. This is why celebrities and politicians are popular targets for deepfakes. They have a significant public presence and provide abundant material for AI to capitalize on. No wonder Barak Obama, Rashmika Mandhans, Katrina Kaif and Kajol are the recent targets!

 

Generative AI and Copyright

Generative AI pays no heed to copyright while creating content and there are well founded rumors that ChatGPT could soon be entangled in copyright lawsuits. There is a genuine concern about the data used for training in Generative AI and writers have sued it for copyright violation alleging that their training is on pirated text. The U.S Federal Trade Commission is also investigating if OpenAI violated consumer protection laws by scraping public data and publishing false information. Generative AI can in fact state false information and cite fake citations as fact, which in their lingo is called 'hallucinations'.

 

Primary concerns surrounding deepfakes

These include:

  • Political implications,
  • Social implications and gendered use
  • Economic implications.

1. Political Implications of Deepfakes

The potential of deepfakes in a political context could harm democratic processes or otherwise disrupt government policy and sow discord among citizens. This is particularly true in a multi cultural, multi religious and multi ethnic democracy like India and the U.S. The ability to create convincing fake videos of political leaders can lead to confusion, disinformation, and a loss of trust in public institutions. In addition, such forms of disinformation could manipulate and distort the complicated media ecosystem. Deepfakes can potentially sway public opinion and change voting habits, but so can online misinformation. However, all agree that deepfakes contribute to the “liar’s dividend” problem and sow uncertainty, reducing trust in online news. Deepfakes can also cause short-term and long-term social harm and accelerate the already declining trust in traditional media. Such erosion can contribute to a culture of factual relativism, fraying the increasingly strained civil society fabric.


An ongoing war or agitation becomes a fertile field for deepfakes. We have seen this in Ukraine war and more so in the Palestine conflict. Twelve months apart we can see the technology used in the latter is twice as good, twice as fast, double the uptick and half the price! Indian social media has been bombarded by deepfakes during the Farmer’s agitation and the CAA and NRC agitations demonizing the government.


While legislation to ban political deepfakes would be challenged in the courts by our champions of freedom of expression, social media platforms can still limit or prohibit their use if the government compels them to do so. It will still be necessary to remain vigilant about verifying the authenticity of online media. Technology companies such as Facebook and Google are combating the spread of deepfakes by detecting and removing manipulated content using advanced technology. 

 

2. Social Implications and gendered use of Deepfakes

Social media platforms’ domination of our online lives has created an ever-present danger of deepfake content spreading unchecked. Cybercriminals use deepfakes to commit identity theft and online fraud, while individuals fall victim to deepfake-enabled scams.


Women often fall prey to deepfakes. We live in a misogynistic society and the tech sector is usually comprising of young men – white, brown and black. They are weaponising this technology against women to satiate their fantasies and every female photograph on the web can be their object of fantasy.  Historically, the first case of malicious use of deepfake was detected in pornography. According to a sensity.ai, 96% of deepfakes are pornographic videos, with over 135 million views on pornographic websites alone. Deepfake pornography exclusively targets women. Pornographic deepfakes can threaten, intimidate, and inflict psychological harm. It reduces women to sexual objects causing emotional distress, and in some cases, lead to financial loss, loss of reputation, and collateral consequences like job loss.

 

3. Economic Implications of Deepfakes

In today’s information-based economy, deepfakes can cause severe damage to businesses and economies. One of the most significant economic impacts is the potential for market manipulation. Deepfakes can create false or misleading information about a company, leading to changes in stock prices and decisions that benefit the creators of the deepfakes. They can also manipulate financial data, making incorrect predictions and investment decisions. This can ripple effect throughout the economy, as inaccurate financial data can lead to wrong assessments of market trends and risks.


Additionally, deepfakes can harm the reputation of businesses and individuals, resulting in lost revenue and opportunities. The spread of deepfake videos and images can cause negative publicity, mistrust, and loss of credibility, which can be challenging to recover from. The allegations concerning related-party transactions and MPS norms were key to the January 24 Hindenburg report that wiped billions of dollars from the market capitalization of Adani Group's listed companies. This may very well turn out to be the next deepfake!

 

Benefits of Deepfake Technology

Generative AI or Deepfake technology, despite its controversial nature, has various benefits for businesses. This technology is now being used to reshape various industries, from marketing to education and entertainment. Here are some of the positive applications of deepfakes technology:

1. Low-Cost Video Campaigns

With deepfake technology, marketers can create video campaigns without needing an in-person actor. Instead, they can purchase an actor’s identity license and use previous digital recordings of the actor to create a new video. This can save time and money and also allow for easy edits to be made without the need for reshooting. 

2. Hyper-Personalization

Deepfake technology allows brands to provide customers with more personalized messaging and experiences based on their preferences. For instance, a brand can alter a model’s skin tone in their marketing to better suit a customer’s ethnicity or skin color, thus increasing inclusivity and reaching a broader market with their campaigns.

3. Bringing the Deceased Back to Life

Deepfake technology has also created interactive images of deceased artists and celebrities, enabling audiences to engage with them long after passing. Examples include the Florida Dali Museum’s collaboration with advertising company Goodby, Silverstein & Partners to recreate a digital reanimation of Salvador Dali and Snoop Dogg’s music video featuring Tupac.

4. Saving Time and Labor in the Film Industry

Deepfake technology has the potential to save time and labor in the film industry by automating the face-swapping process currently done manually by VFX artists. This allows for more efficient and cost-effective production of films and other media. 

5. Educating People in a More Interactive Way

Deepfake technology also makes online learning more interactive by generating lecture videos from text-based content or audio narration. Additionally, it can be used to construct artificial voices from historical figures, allowing them to tell their stories/speeches with their own voice. 

6. Engaging with Viewers or Customers

Deepfake technology can also engage viewers or customers by providing personalized recommendations and offers to meet their needs. Fashion companies are using virtual fitting rooms where customers can deepfake their faces onto virtual models to see how clothes would look on them.

 

 

From personal reputation to international relations all can be threatened by deepfakes.  Typically, deepfakes are used to purposefully spread false information or they may have a malicious intent behind their use. They can be designed to harass, intimidate, demean and undermine people. Deepfakes can also create misinformation and confusion about important issues.



1 comment: