You may remember a video by director Jordan Peele showing former President Barack Obama warning about fake videos – which turned out, itself, to be such a fake video. Known as “deepfakes,” these videos can make anyone, with just a snippet of video and audio data, say anything they want. And Peele’s video was done in 2018 – they’re a lot better now.
Consequently, there’s been a great deal of effort since then in the hopes of detecting such deepfakes before they cause serious problems. But a Boise-based entrepreneur says that’s the wrong approach.
“It is a perpetual game of cat-and-mouse and whack-a-mole,” said Jason Crawforth, founder and CEO of Swear Inc, and former CEO of TreeTop Technologies, one of the Boise startup community’s OGs. “Defense doesn’t win this game.”
The reason detection doesn’t work is that deepfake generation technology is advancing a lot faster than deepfake detection technology, Crawforth said.
“High-tech manipulation of video/audio images is already past what humans can detect and will very shortly be past what computers can detect,” Crawforth said. As an example, five years ago, software needed 30 minutes of recorded speech in a person’s native language to create an audio of that person saying something else, by piecing together those sounds, he said. “Now, they need seven seconds.”
Similarly, generating deepfake videos used to be more challenging because deepfake videos didn’t blink. “Blinking is a complex thing, because rarely are someone’s eyes closed,” Crawforth said. Now, deepfake generation software is way past blinking. “They can put microheartbeats on the face that we can’t see, and reflections on the eyes,” he said.
And such deepfake videos are more than just a parlor trick, Crawforth warned. “They’re very dangerous,” he said. “What can be done with these types of manipulations can be catastrophic. They can destroy relationships and marriages and bury businesses.”
The problem is that more and more business organizations rely on audio and video digital assets to substantiate business processes – not just the media, but also insurance companies, real estate companies, and surveillance and intelligence organizations, Crawforth said.
Scared of deepfakes yet?
Crawforth started thinking about how to defeat deepfakes about six years ago. “What we knew was going to happen in the world of technology was it would rely on forensic defense methods to try to identify and discredit fake content,” he said. “We knew that was a failing proposition. Average citizens will not have access to those,” and anyone could be a victim of fraudulent, threatening, or extorting videos, he said.
So the idea of Swear was to go on the offensive instead, and Crawforth has spent several years on research and development. “We’ve created a solution that records video and audio and makes them so they’re completely unchangeable,” he said.
Well, not exactly. They can be changed. The point is, using Swear’s real-time blockchain technology, you’ll know that they’ve been changed. “We are creating unbreakable digital content in real time using Web3 distributed ledger technology,” Crawforth said. “If something does change – a single pixel, a soundbyte – we can detect it. We use a color code. Green means that portion is original and unmodified; the section that’s red is modified.” The software can even detect removing sound artifacts such as static and hiss, he added.
Swear does this by capturing metadata while the audio, image, or video are being recorded, embedding that content into the audio, image, or video, and then blockchaining it, which all happens in less than a second, Crawforth said. “We don’t wait until the end,” he said. “We do it instantly,” so the digital content can’t be manipulated along the way.
The metadata can be anything handy that’s going on while the digital content is being recorded – Global Positioning System data, the tilt of the phone, speed, altitude, acceleration, the Wi-Fi signal, satellites overhead, the position of the sun and moon, global clock world time, Crawforth said. The video never leaves the recording device, and only the hashed data goes to the blockchain, meaning people wouldn’t be able to reverse-engineer the hashes to recreate the message, he said.
How exactly the product will be marketed and sold, Crawforth isn’t sure yet. “We’re still working on it and talking with various organizations,” he said. More than likely, it will come out in two versions to address two major markets.
The first would be consumer, addressing influencers and the media who are producing content on their own. “Every minute there are over 800 hours of content published on YouTube,” Crawforth said. That component might come out as soon as Q4 this year or Q1 next year, he said. In addition, consumer customers will be able to publish their content on their own channel on the Swear.com website, he added. “The ultimate goal is to have it on every iPhone and Android as an option,” he said.
The second component is commercial, working with organizations that could use the Swear technology to protect content collected by its devices, such as surveillance systems, police bodycams, and voting machines, Crawforth said.
“Our technology is pretty agnostic,” Crawforth said, and it can work with a variety of blockchain systems, public or private. For example, if he were to partner with the military, chances are it would have its own blockchain technology it would want to use, he explained.
Building a company
Exactly how many employees Swear has is a difficult question, Crawforth said. Swear is working with a local software company, Ventive, to help develop the technology. But regardless of exactly who the people are working for at the moment, the total number of people working on the project right now is about 12, he said.
Needless to say, protecting Swear’s intellectual property is key. The company already has six patents that have been approved and published, and more are pending, Crawforth said.
Another important aspect of the company was the name, and that was predicated on being able to get a URL to reflect it. “We don’t want a domain like WeCreateVideosThatCantBeModified.com,” he said. “We were looking for a word. This is world-changing. You’d better have a name that is equally powerful.”
While Crawforth looked at URLs such as Truth.com, that didn’t feel right, so he settled on Swear, which has a double meaning in this context. “When you swear, you put your hand up and swear to tell the truth,” he said. “You can swear now, and protect the truth, or swear later, if you don’t.” He wouldn’t say how much he had to spend to acquire the Swear.com URL, but admitted it was six figures. “It was not cheap.”
Because of the technology’s broad application, Swear’s exit strategy is…complicated. “If a massive industry player wants to acquire and utilize us, that’s great,” Crawforth said.
The problem is, there probably isn’t a single industry player, no matter how massive, that spans the breadth of the technology’s potential. “Let’s say Apple wants to acquire it,” Crawforth said. “There’s a lot of industries that they will never be in,” such as police bodycams, voting machines, security cameras, traffic cameras, and so on. So what will likely happen is that Swear will see multiple or cascading exits – license it to Apple and then license it to Google a year later – or a major player acquires the technology and then licenses it back to Swear to in turn license it to companies making bodycams, voting machines, and so on.
But what’s really important to Crawforth is that the technology gets out there and gets used, he said. “I didn’t start Swear to make a bunch of money,” he said, noting that he’s 52 and had been happily retired for 15 years. “I did it to save the world from something terrible coming. It has to be the right exit. I don’t want someone to put this on the shelf.”
Sharon Fisher is a digital nomad who writes about entrepreneurship.