With the easy availability of techniques to manipulate, copy, and share digital files, nothing is sacrosanct. What looks like a genuine photo may be a deepfake
Our recent skirmishes with deepfake videos have left us all wary of trusting our eyes. These incredibly convincing videos use advanced AI technology to morph the faces of prominent figures on top videos originally featuring other people. These videos are easy to disseminate over social media. And boy, do they seem real!There has been a radical shift in recent times over how we receive and process information. Gone are the days when we had to wait to get our news from newspapers and TV channels. Today, we feed on a constant stream of unverified information from our social networking websites and our WhatsApp forwards. This ‘news’ is available at our fingertips, served in palatable formats that take seconds to grasp.
There is no background, no finer minutiae to discern, and definitely no verification of authenticity. One clicks to see and one click to forward.As a result, these spread like wildfire. There is global concern about deepfakes. Prime Minister Narendra Modi has called upon the media to spread awareness about the misuse of artificial intelligence (AI) to create deepfakes. He has emphasised the need of tagging digital products at source, creating accountability at the level of generation of any videos. British Prime Minister Rishi Sunak has called for a joint global alliance on AI safety.
The US President has issued an executive order on AI, a first-of-its-kind action from the government to tackle some challenges posed by AI, including plans on how to identify if an image is real or fake. Joe Biden's order calls for a new set of government-led standards on watermarking AI-generated content, to check deepfakes.Though deepfakes have been around for several years, the problem has become grave of late because sophisticated techniques for altering photos and videos have become so accessible that deepfakes have become commonplace. They can be used by criminals and scamsters to the detriment of citizens, to spread fake news and incite hatred, even violence.AI is ubiquitous and it is changing the world very fast. It has limitless potential but also poses some unique threats. Biden said, “AI is all around us. To realise the promise of AI and avoid the risk, we need to govern this technology.” His order is a first step that seeks to ensure that AI is trustworthy, not deceptive and helpful, not destructive. The order is meant to guide how AI is developed so that developers can profit without jeopardising user safety.
Using the Defense Production Act, the order directs AI companies to share safety test results and other information with the government. The National Institute of Standards and Technology has been asked to create standards to ensure that AI tools are safe and secure before they are released for public use. The Commerce Department is to issue guidelines to digitally watermark AI-generated content to help differentiate between authentic creations and those generated by software.Stamp papers, documents, currency notes and some fancy photographs and paintings carry watermarks to establish genuineness.
These are created at the wet stage of making paper. Similarly, a digital watermark is embedded at the time of creating digital content. A digital watermark is like a digital signature that is embedded into a digital file to identify its originator or owner. This mark can be a code, logo, text or a unique sign that identifies the content’s owner and also helps to prove its authenticity.A digital watermark has to be very creative, a way of altering a file that is not noticeable to the user. It might be difficult to embed a watermark within a text file. However, it is quite easy to do it for audio, video and image formats by changing a few bits, without making a noticeable difference in the music, movie or photo.Like watermarks on paper currency or documents, digital watermark helps distinguish between a real object and a fake one and also ascertain who owns it. It can track the online use of the file and warn the creator against unauthorised access and use. Digital watermarks can check the spread of fake images to some extent but cannot control it.Though digital watermark may not be a solution to deepfakes, it is an important tool to fight against piracy and unauthorised access and distribution of digital files.
With the invention of easy techniques to copy and share digital files, companies and businesses are facing a huge challenge to safeguard their valuable content and content creators. But if a watermark is embedded into the master copy of a DVD movie, then all copies of that disc carry a unique identification. If a licence holder distributes them outside his authorised jurisdiction, the watermark provides a trace.Digital watermarking also helps companies in content tracking and monitoring its distribution and usage across various consumers and platforms. This can help them make more informed decisions about their marketing, advertising and distribution strategies. By ensuring that their content is protected from unauthorised use, businesses can maintain greater control over their brand identity.Watermarking technology is a simple solution that holds promise. But it may not be enough on its own. It is naive to accept this as a foolproof protection against piracy and deepfakes. It is very easy to remove a watermark that is toward the edge of an image; it can simply be cropped out of the picture. Further, there are apps available that can remove most watermarks in just a few clicks. This is why a digital watermark cannot offer complete protection.The Distill scientific journal defines AI Safety as “The goal of long-term artificial intelligence (AI) safety is to ensure that advanced AI systems are reliably aligned with human values — that they reliably do things that people want them to do.”A sustainable solution to deepfakes should be able to balance the needs of AI companies with consumer rights and that will require a bouquet of practical solutions like a digital watermark.
That will require cooperation from online platforms and search engines to voluntarily decide to allow only those contents that bear a watermark. This is a larger issue and calls for strict adherence to ethics in the business of AI development.
(The writer is Additional Chief Secretary in Govt of UP. Views are personal)