Fill out the brief form below for access to the free report.
Deepfakes are a Different Kind of Threat to Democracy
Glenn Gaffney, a former CIA officer who headed the agency's science and technology division, discusses the rise of deepfake videos and the need for tools to discern the truth in what we are seeing.
Glenn Gaffney spent more than 30 years working in the intelligence world as a CIA officer, including as the Director of Science and Technology. To him, the heart of intelligence work is being able to discern truth, even though others may use techniques to make it hard to discover the truth. Now the Executive Vice President of InQTel Labs, a nonprofit organization created to ensure national security agencies have access to innovative technologies, he continues his work as an expert in emerging technologies.
Gaffney spoke with Lindsay Lloyd, the Bradford M. Freeman Director of the Human Freedom Initiative at the George W. Bush Institute; Chris Walsh, Senior Program Manager in the Human Freedom Initiative; Andrew Kaufmann, Host of the Bush Institute podcast, The Strategerist; and William McKenzie, Senior Editorial Advisor at the Bush Institute, about the intersection of technology and democracy. He particularly focused on the rise of deepfakes, which are synthetically-altered videos that nations or individuals create and distribute with ill intent. He also discussed technologies that companies are developing to authenticate videos. Creating those tools will help us discern the truth, but they need to be built fast enough to keep deepfakes from undermining democracies.
Mankind invented the internet, then not long after that invented the ability to transmit images on the internet, and after that came up with the ability to transmit fake images on the internet. What's different now with deepfakes? How are they different than what's been happening?
It’s useful to think about fakes in three ways. One is fakes, which is you know it's fake and you can see that it's fake. You can discern that. Then there are deepfakes. You are aware that there are deepfakes, but they are hard to otherwise detect or discern them as fakes. It takes a different level of effort to uncover them. Then, there's another class that I would call very deepfakes. You’re not even aware that something is fake.
It’s also useful to think about this in three phases. If you look back at Hollywood creating visual images that were pretty close to being real, you could look at them and say, "That’s really good, but it's clearly not the person." Think of some of the brilliant editing that was done with Forrest Gump. Then look at creative expression today. It is exceptionally hard to discern whether an actor or background is real.
When you apply machine learning to that highly-skilled craft, you are applying new techniques and democratizing them by making them free and available to folks over the internet. Different from creative expression, you can apply these things to defraud and deceive. And the spread and ease of use of the technology makes it available to many who would use it for nefarious purposes. That trend continues to grow.
Now, some great technology advances in recording and then reproducing voice have helped people who were going to lose their voice because of a medical condition or surgical procedure. But you apply that same ability to fake your way into a voice-authenticated interaction on the internet, you're using that for a nefarious purpose. The more these things become accessible and easy, the more we see them employed for nefarious purposes.
What would a nefarious purpose look like?
One example is using voice to defraud or deceive to gain access to something that's not yours. Same thing with facial recognition. Can we use a version of you that is good enough to defraud or deceive?
And the nefarious purpose could be as simple as eroding the basic trust in what you see. Casting doubt on what you see and hear begins to erode the basis of truth, which is what we put so much of our faith in in a democracy. Trust underpins our whole democracy.
Casting doubt on what you see and hear begins to erode the basis of truth, which is what we put so much of our faith in in a democracy. Trust underpins our whole democracy.
We’ve seen fraudulent activity against elections in Central Europe. We've seen fraudulent activity incite mob violence in India. We see it exacerbate extremist positions in the United States. The ability to have the internet and social media amplify and spread them quickly can have real and lasting impact through undermining trust, questioning what we perceive, and how we act on truth.
Were those examples you gave about Central Europe, America and India done by individuals or by state actors?
Increasingly, individuals have the capability as these technologies become easier to use and more ubiquitous. It used to be nations using propaganda or disinformation. Now, the ability to do mass distribution to specific audiences allows for a different form of disinformation or propaganda than before.
What impact is this having on governments, commerce, or elections? What are some examples?
The inciting of mob violence in India led to a whole group taking action against another group. Or Iran challenging whether their ships were in international waters. The ability to cast doubt or change the perception of the way that people are seeing an event has a profound impact, begins to separate people, and exacerbate existing divisions.
The sheer volume of these things and how they are happening erode the confidence of people in what they see and believe. Along with our existing tensions, this tends to drive people further into corners.
There’s already an erosion of trust at scale. Our social division is easy for a nation to exploit. Russia looks into those seams to drive people further apart and to further distrust each other. That makes it harder for us to nurture the kind of compromise that democracy is built on.
Our social division is easy for a nation to exploit. Russia looks into those seams to drive people further apart and to further distrust each other. That makes it harder for us to nurture the kind of compromise that democracy's built on.
You recently said at a Bush Center event that “truth discernment” has always been a part of how nation-states compete. But you also asserted that deepfakes are a different kind of threat. How and why do they differ from the more traditional aspects of international competition?
The “why” of intelligence for me has always been the relentless pursuit of truth. We are looking to uncover the truth and present it in a way that policymakers can make quality decisions supported by real evidence.
The opposite of that is deception, hiding, covering. Folks want to hide what they’re up to. They don't want it discovered, so you're into this cat-and-mouse game of discovering truth and presenting that unadulterated truth to policymakers. It may or may not agree with their policy. You’re just producing the truth and then let them make the decision.
Meanwhile, the other nation is working to throw you off the trail, to plant deceiving things. They may have an entire disinformation campaign. The ability to discover, uncover, and discern gets met with new capabilities to cover, hide, and obfuscate.
When it was just a nation-state competition, or part of great power competition or warfare between countries, you're expecting things to happen in certain ways. The sheer volume and speed outmatch you when suddenly the ability to present things as truth that aren't truth spreads through a wide range of people. The level of technical effort at speed and scale becomes a real challenge.
I don't think I ever served in an administration that didn't ask the CIA or NSA [National Security Agency] at some point to verify a video. For example: Is this video that we received of this terrorist group or this leading terrorist figure authentic? Is it what it purports to be?
Highly-skilled analysts and technicians would look at these issues. Nonetheless, it gets harder and harder, so you have to apply and develop new techniques.
In the Arab Spring, an event was captured, presented, and had this amazing effect across a population and then spread. It couldn't have been predicted, it just happened. Of course, that was an authentic event, but what if it were not? What is a realistic timeframe to determine the authenticity of something forensically when it spreads so far and action is taking place? Now, you're talking about this at a whole different scale and requires a different approach. We have to equip the populace to understand what they're seeing, test it, and validate it.
There’s a level of human psychology here. If you’re presented with something that you're predisposed to want to believe, that will spread much faster.
This is particularly an acute problem if we're predisposed to believe this newspaper or this network. To question the video that you see there becomes even more of a challenge.
How many of us have been brought up with seeing is believing? Our brains are such efficient engines that reward themselves for completing a story. Once you see an image, your brain records that image and begins to connect other things to it to complete the story, whether true or not.
You have to be trained to challenge your own thinking. And this deals with how we teach adults today and in the next generation to process what they're seeing via the internet and different sources. How do we equip them to pause and ask if this information is truthful? We need to challenge one another and present evidence. It takes a different level of discipline to slow down. And we need open source technology to have the kind of automated forensic investigation that needs to take place.
Thinking about the private sector, how do you rate the ability of our modern technologies to stay ahead of the spread of deepfakes? Is it a fair race? Is it a race at all?
It is a race. A lot of the technology wasn't invented for this purpose. It's been misappropriated. That's the nature of tools in technology, they can be used for good and evil.
But having been grim, let me be positive. Our innovation culture in this country is second to none. The speed at which we identify problems and come up with new solutions is incredible. The growing ability to open source technology and solutions and make them ubiquitous and available is a strong point. Developing the techniques and technologies, making them available to people, and still capitalizing on them commercially is a uniquely American approach to innovation. That is a real strength.
Our innovation culture in this country is second to none. The speed at which we identify problems and come up with new solutions is incredible. The growing ability to open source technology and solutions and make them ubiquitous and available is a strong point.
Startups are incredible at identifying a problem and getting after it and developing unique and exquisite technology. This is a problem that they have begun to address in a serious way. The market will grow much the way the cybersecurity market grew. What was once the purview of PhDs became the purview of anybody with a laptop and a modem. There's an analogous growth and spread to the kind of issues that we're talking about. We’re already seeing the beginning of this trend.
But the venture capital community can demand more. They're putting money on the table. They can push a startup looking at a particular technology to think about how it might be misused and how to counter that. The ones with the money making the investment can have a real impact.
And there is a role for industry to serve as channel partners for startups. Partnerships could allow startups to experiment, grow, and be an accelerator. Government also can support the open source community and share how advanced forensics might be done.
Schools can develop the next generation of researchers, too. A lot of the research that kids do for papers is done on the internet. They can share information: Did you see this? What about this? Watching my kids at the dinner table was a fascinating exercise. Some of my younger kids would say, "Hey, did you see such and such on the internet?" An older sibling would respond, "Check your sources." The dining room table is one of the best laboratory environments in the world for technical issues and the social sciences.
Following up on that analogy, what would you say to someone sitting around their dinner table in Topeka and wondering, "How can I ever discern whether these deepfakes are real or not?" What might the tools be like in the future?
I'm excited about a couple of things. One, companies are developing new methods for authentication and looking images at the point of origin. They could provide a Good Housekeeping seal of approval by capturing whether a picture or video was actually taken and not synthesized. They could authenticate it upfront. A number of companies are looking at doing this at scale.
University research also is looking at different forensic techniques that could determine if something was completely synthetic or synthetically manipulated. In that research, and in our organization’s work with startups, we're seeing a suite of methods being applied to a whole different set of signatures. They could begin to effectively answer the problem.
University research also is looking at different forensic techniques that could determine if something was completely synthetic or synthetically manipulated.
You have mentioned startups and the open source community. But what about the big social media and big search companies? Do they have a responsibility also? Or is it more on the creation and authentication methods?
Those companies do have a responsibility. I see some of them trying to be more proactive. They have put some challenges out to try to draw open source competition and begin to build solutions. Some big media companies are trying to address some of these pieces. And partnerships are being formed.
I never feel like the action is fast enough. But I give them credit and think there's opportunity to do more.