The Promise (And Pitfalls!) of Digital Avatars

True story: I’ve wanted an AI avatar for a long time.?

So often, I want to do more than I’m able to do. I want to:?

???Speak at conferences, but for one reason or another—scheduling conflicts, I don’t know the language—I’m unable to do so.?

?? Extend the content I have, making it more accessible for specific groups of people with different learning styles.?

?? Create training courses that are customizable and tailored to each company’s values and objectives.

?? Generate personalized messaging that understands what my relationship is like with a client—and then provide tailored proposals (and maybe even hold a Q&A!)

So??

I built an avatar. ??

What Are AI Avatars??

Avatars are a digital likeness of a person. Here, a “person” can be broadly interpreted:?

… It could look like me.?

… It could look like a completely fabricated person.?

… It could be the mascot of your team or company.?

You can use all sorts of different images, audio, and video references. When I started playing with avatars about a year ago, I was able to take a headshot and some audio from a livestream and create an early version—it looked like me. It moved like me, sounded like me, and could speak different languages. Recently, it’s gotten even better.?

Avatars offer some pretty significant features and benefits:

• There’s no need to go into a studio to create professional videos. This is a huge time saver, and a balm if you’re camera shy. The quality is great, it’s low cost, and it’s great for highly repeated content (think training, sales, customer service inquiries, or FAQs—anywhere you want consistency).?

• Scale. The scale is incredible. All you need to have is some text, and you feed it to the video. You can customize each script and each video for each person. And it can speak in multiple languages!

• Interactivity. Provided your avatar has a knowledge base, you can actually interact with it.

How I Created My Avatar

I chose to work with Simulus.ai for their expertise in video production. With three decades of video production experience, they know how to capture video well. I had a video shoot at their studios in Atlanta, where, against a green screen, I read from a script, changing outfits, and capturing video from three different angles.

Simulus.ai rendered my avatar in HeyGen, applying what felt like magic to enhance its appearance.

Next, I generated a complete script (with some AI assistance). Simulus.ai then created the video featuring my avatar, adding graphics as needed. I now have an avatar "in the can" ready for video creation. While it's not directly accessible to me (I'll continue working with Simulus for professional-quality results), it's a powerful tool for creating professional content. Here's a screenshot of my avatar in action.

In my recent livestream , I showcased a video of my digital double and even conducted a Q&A session. Here's a brief excerpt from my conversation... with myself.

AI Charlene: Hi, I'm Charlene Li's digital human avatar. How is your day going? How can Charlene Li's digital human assist you today??

Me: My day is going. Great. I have a question. What's the best way for organizations to create a generative AI strategy?

AI Charlene: That's great to hear. For organizations to create a generative AI strategy, they should start by educating themselves on AI capabilities. Upskill top leaders and involve HR early in the planning process. Collaborating with other departments and communicating transparently about AI's impact is also key. Would you like to know more about any specific aspect in particular??

If you watch the livestream , you’ll see my avatar is clearly not ready for prime time—but it is my voice that's being used to render those answers, and those answers are coming right from my knowledge base that I’ve uploaded to it!

AI Avatar Limitations

When I ask audiences how many find all this fascinating, many raise their hands. But far more hands go up when I ask, "How many think this is really freaky?"

I get it—there's something about my avatar that isn't quite right. It lacks the nuanced speech and tone of a real person. It's also rather stiff—exaggeratedly polished. The emotions aren't quite authentic, and it definitely can’t yet tell a joke. The first time I saw my digital doppleg?nger, it was a surreal, out-of-body experience to see me “doing” something I had never done before.?

AI avatars are bringing us closer to the Uncanny Valley, a concept Nir Zicherman wrote about beautifully in a recent newsletter —and it makes some people really uncomfortable. This theory, first proposed by Japanese scientist Masahiro Mori, suggests that our affinity for these tools increases until they become too human-like but not quite there. At this point, we hit the Uncanny Valley, and our trust plummets. Tiny glitches in intonation, movement, or even hand positioning can trigger us to think, "That's not real. I don't trust it."?

Avatars in particular are bringing us closer and closer to that valley, and it's something we want to be very aware of, especially as GenAI continues to evolve.?

Significant ethical considerations also exist: While I own and have access to my avatar, it's now quite easy to create avatars with just a headshot and an audio recording. This opens up possibilities for misrepresentation and deep fakes.

I work with various groups to create safeguards around this. Tools that create these avatars always have agreements stating you won't create representations of people without permission. However, this doesn't prevent someone from doing so for nefarious reasons. It's a significant issue we must address.

I come back to what I know to be true: Technology itself is neutral—it can be used for good or evil. It's incumbent upon those of us who want to use it for positive purposes to demonstrate its beneficial applications, while also educating people to recognize misinformation and deep fakes. I'm a big believer that the negative social effects of technology should not be a reason to halt our progress. We can keep moving forward, but we do need to put safeguards in place.?

Final Thoughts

As with most things, there are some essential “do’s,” and some very essential “don’ts.”?

When it comes to AI Avatars:?

? Do use them for repetitive or standardized content (think training, onboarding, FAWs, product demos for consistent delivery)

? Do personalize the experience. Leverage multilingual capabilities

? Do keep humans in the loop

? Do test for user engagement

? Do use as supplemental content

? Do find applications that take advantage of scaling customization?

? Don’t use for high-stakes communications

? Don’t ignore the uncanny valley risk

? Don’t skip on script optimization

? Don’t misrepresent or deceive

? Don’t overload with visual effects??

I get excited when I think about what the future can look like with the support of AI avatars, especially when it comes to training and communication. There’s so much potential here!?

At the end of the day, I think of avatars as the ultimate way to build relationships and connections.?

… Isn’t that what it’s all about??