The DeepFake epidemic is looming and Adobe is preparing for the worst
The maker of PhotoShop and Premier Pro has given the world AI-powered tools to create compelling knockoffs. Now CEO Shantanu Narayen wants to clean up the mess.
By Aayushi Pratap
IImagine a deep fake video of House Speaker Nancy Pelosi in which her speech is intentionally garbled and the words she uses are altered to deliver an offensive message to a large number of voters. Now imagine that the technology used to create the video was so sophisticated that it seemed completely real, making the manipulation undetectable, unlike the clumsy Pelosi fakes that circulated – and were quickly debunked – in 2020 and 2021. What would be the impact of such a video on hotly contested House races in a midterm election?
That’s the dilemma facing Adobe, maker of the world’s most popular photo and video editing tools, as it undergoes a top-to-bottom review and redesign of its product line using artificial intelligence and deep learning techniques. This includes upgrades to Photoshop software and the company’s Premier Pro video editing tool. But it’s also true that “Photoshop” something is now a verb with a negative connotation – a reality that Adobe CEO Shantanu Narayen is all too familiar with.
“You can argue that the most important thing on the internet now is content authentication,” Narayen told Forbes. “When we create the content of the world, [we have to] assist in the authenticity of such content or the provenance of such content. »
So three years ago, Adobe launched an initiative called the Content Authenticity Initiative, starting with a handful of media and technology industry partners. It’s a company that’s grown to encompass more than 700 companies, with global events to publicize the “provenance” push, as 59-year-old Narayen calls it, in which content designers and consumers can, if they wish, create and track a digital trail that shows who is responsible for a given video or image and any edits they have made to it.
Deepfakes are just one of Narayen’s headaches. Adobe posted $15.8 billion in sales in 2021 (fiscal year ending Dec. 3), but the San Jose-based company’s forecast has missed Wall Street estimates for the past two quarters. Blame the usual suspects: rising interest rates, supply chain issues and trade embargoes in Russia and Belarus. Since peaking at $688 per share in November, Adobe shares have fallen 47% to recently hit $366, compared to a 26% decline for the Nasdaq. “The company is still growing, but it’s experiencing a significant downturn,” said Cornelio Ash, analyst at William O’Neil & Co Inc.
“Very soon, because AI can be more powerful than human editing, you won’t be able to tell fact from fiction, fact from artificial reality.”
For years, Adobe has relied on a suite of venerable flagships, which it bundles together as a digital media group, to generate the bulk of its revenue. Products like Photoshop (first released in 1990), Illustrator (1987), Premiere Pro (2003) and Acrobat (1993) generated 73% of its revenue in 2021. But, despite a successful transition to the cloud, these businesses are slowing down. The group’s annual recurring revenue has historically averaged growth of around 20%, but in Abode’s last quarter it slowed to 15.5%, Ash says.
Then there’s Digital Experience, launched in 2012. These are services that Adobe provides to businesses by analyzing their customers’ “digital footprints” – i.e. tracking people’s behavior online – say how long they spend on a specific web page and what products they view.
“If…you’ve engaged or interacted with a customer, and you still don’t act like they know the customer, there’s nothing more frustrating than that experience,” Narayen says. This is what the company wants to change with its Digital Experience activity. Despite being Adobe’s fastest growing segment, generating $4 billion in sales in 2021, up 24% year-over-year, it faces intense competition from Salesforce and Google.
Adobe also faces increasing competition from smaller rivals such as Australian graphic design platform Canva and US company Docusign. “Adobe was a bit late in responding to the market space that Canva was targeting…a market space that wasn’t actually professional designers,” says Gartner analyst Chris Ross.
Much cheaper (starting at just $120/year) and easier to use than Adobe’s offerings, Canva has quickly become a real threat. After all, who wants to pay $600/year for a Creative Suite subscription — and spend hours learning Illustrator — just to design a menu or a wedding invitation? In September, Canva, which is less than ten years old, was valued at $40 billion. Melanie Perkins, its 35-year-old CEO, is currently worth around $6.5 billion.
Adobe responds. In December, it launched Creative Cloud Express, a new app, even cheaper than Canva, aimed at first-time users, from college students to social media influencers. More broadly, the company’s attitude is anything but complacent. Adobe has embarked on an effort to reinvent all of its products using artificial intelligence and deep learning techniques – an initiative known internally as “AI First” – which can make time-consuming edits in Photoshop possible. in minutes rather than hours. “AI provides features like sky replacement, which is at the click of a button, you go from blue to gray,” says Dana Rao, general counsel and chief trustee at Adobe. “Then you can go in and make fine adjustments, but that’s a much quicker task.”
In 2016, the company launched Adobe Sensei, a suite of features that, among other things, allows users to get rid of unwanted objects from video footage, smooth skin tone, alter facial expression, and change the tone of the voice with great ease.
Of course, the same tools can turn fiction into reality. Rao recalls sitting down with the company’s chief product officer, Scott Belsky, three years ago: “We were looking at all the innovations we had in AI editing, and we had all the two the same thought at the same time: very soon, because AI may be more powerful than human editing, you won’t be able to tell fact from fiction, fact from artificial reality.
The content authenticity push is one answer to this problem. Another is an extensive internal process, over which Rao presides, for subjecting new Adobe features and products to rigorous ethical reviews — a process that sometimes prevents the work of Adobe’s engineering teams from being made public. For example, says Rao: “On the imaging side, when we think about integrating AI into our technology, we have to ask ourselves if it was trained on a diverse set of images, datasets, so that the output is respectful of all types. people who use it.
Abhishek Gupta, founder of the Montreal AI Ethics Institute, believes that adding “truth labels” to content won’t make much difference. Public supporters of altered content often desperately want to believe that it is genuine, despite all indications to the contrary. “Fake content is suitable for people with quick interactions with very little time spent judging whether something is genuine or not,” Gupta says.
Ultimately, says Adobe’s Narayen, consumers themselves have an obligation to seek the truth – to use the tools available to verify that the images and videos they view, share, post and retweet are real. “They also have this responsibility, which is no different than any other responsibility a consumer has – to make sure they are protected.”