Generative AI is becoming more and more common. The outputs are increasingly realistic, and it is becoming far more difficult to differentiate human-created content from AI-generated content than it was even a year ago. In this article, I will share a bit about my experience creating a deepfake video of myself.

My Experience

I created a deepfake video of myself for a board report. It was pretty simple to do, and it was inexpensive. The platform I used, HeyGen, streamlines the process. Additionally, they have you record a disclaimer when you upload your “dataset” (a 2-minute video of you talking about anything in the setting you choose) that gives them permission to create an avatar of you. This is an important step for responsible companies like HeyGen, striving to do as much as they can to protect personal information.

The Responses

Personally, I believe that AI use should be disclosed. For this particular board report, it was not disclosed until the end because we wanted to see the response. Some of the people in the room knew me, and others did not. They had varying levels of AI familiarity. Most of them felt that something seemed a little strange (for one thing, video board reports are not standard practice from my department), but they couldn’t put a finger on what. A few knew it was not actually me.

The more amusing response came when I showed it to my 30-year-old daughter. After about 20 seconds, she said, “Turn it off! I can’t, I just can’t watch that anymore!” She was very perturbed. Later, she explained that she saw it as me but that she knew it wasn’t me and that it just freaked her out. I think I read something about a company that can create digital avatars of deceased loved ones for future conversations … or maybe that was a movie … either way, I don’t think my daughter will be using that technology in the future.

Not much creeps me out, but I must admit that creating this deepfake video of myself left me feeling a little unsettled. Now that I know how easy it was to create, I have a greater sense of discomfort or empathy for those individuals who are likely to have many deepfakes created without their permission or consent—politicians, actors, journalists, etc. And I worry about whether I will be able to tell the difference between human-created videos and AI-generated videos.

“Real” Versus “Fake” Content

Note that I don’t call AI-generated content “fake” and human-created content “real.” Whether created by AI or a human, the output is actual content. Real is a term often used to designate human-created content, but that can be interpreted in different ways and is becoming increasingly blurry as AI is integrated in more and more tools and processes.

Is it real content if AI is asked to edit a final draft or create some title options for an article? If a human has a novel idea for content and asks AI to draft a poem using that idea, is the output real? Or is it only real content if it’s created by a human hand? What if it’s typed on a computer that has spell-check? If it’s an avatar that really shows a version of you that you voluntarily created, and if you wrote the script for the avatar to say in the video, is the output real?

Potential Use Cases

While we know that bad actors are using this technology, legitimate uses are plenty. Some examples might include instructional videos featuring the same instructor people saw in “real life” classes to help connect learning; personal anecdotes or stories told by a realistic avatar of the person who experienced them to help engage the viewer; a sales demo video of your salesperson showing how a product can be used in a real setting, which can be impactful and drive revenue.

Conclusion

The possibility of generative AI creating content that is indistinguishable from human-created content is already happening today, and I believe that indistinguishable content will explode over the next few years. This poses significant societal risks as well as rewards. Many are concerned about the impact on the upcoming election season in the United States. Others are excited that they can now create realistic, quality education much faster and on larger scales than before.

Whether this creeps you out or not, it’s a good idea to at least understand the potential. Even if you never use it yourself, you should know how easy it is to create this kind of content—and learn about some of the telltale signs that might be seen in some early AI-generated video content (like too many fingers on a hand or the temples of eyeglasses missing). I’d encourage you to explore and consider how, or if, this kind of AI-generated content might be useful for your role.

Author's note: AI was used in the development of this article by generating title ideas and the video from which the screenshot was taken.