Frigid temperatures this weekend prompt cold weather advisories

MPR News helps you turn down the noise and build shared understanding. Turn up your support for this public resource and keep trusted journalism accessible to all. 

Turn Up Your Support

 after a staff member was charged with using digital tools — including AI — to manipulate student photos in sexually explicit ways. 

William Haslach worked for Adventure Club, a before-and-after-school care program with Stillwater Area Public Schools, and is facing 

 for possessing child sexual abuse material and using AI to create it. 

The U.S. Attorney’s Office has identified 100 children victimized in the case. Haslach has pleaded not guilty. But the case has alarmed parents and raised broader questions about how easily AI can be misused.  

, an AI and analytics expert at the University of Minnesota’s Carlson School of Management, discussed potential misuses of AI software and the need for safeguards on Morning Edition. 

The following transcript has been lightly edited for clarity. Listen to the conversation by clicking the player button.

Law enforcement says AI generated images depicting child sexual exploitation and abuse is a growing issue, not only among adults, but some kids are creating sexually explicit images of other kids, too. What does AI photo manipulation look like? 

We’re kind of in this pre-seat-belt moment, with respect to cars. When cars were first invented, people drove them without seat belts, without traffic lights, without guardrails — we didn’t say that cars were crazy, but we built those guardrails over time. 

We can think of two types of AI. Traditional AI is something that I might use to build a model to detect whether there is cancer in some kind of radiology image. 

Generative AI is another type that involves conditional generative modeling. Imagine taking an image, a photo or maybe a 10-second sound clip of me speaking. There is technology now that can represent that, that can create a digital signature of my profile, of my sound, in those 10 seconds, or the profile of that image. They call this an embedding. It’s basically a representation of what that image looks like. 

Then there are these existing models that are trained on thousands, and tens of thousands, of people’s images or voice. And they can plug in your voice signature, or my voice signature, or my imagery, into that large model and then make it do anything. So that’s what’s going on.

How easy is it for people to use these tools?

The technology is advancing really fast. This capability is out there. It’s baked into many of the popular AI platforms that are out there. 

There are a lot of positive uses of this as well. If I wanted to clone my voice to create an audio version of my book, I could do that and I don’t have to spend three days reading my book out there. Technology always has its two sides. 

How would a school even regulate something like this? 

There are lots of things we can do even today, and the industry recognizes this problem. They’re working on a variety of safeguards. There are technologies evolving around watermarking content. 

As an individual user, you may want to turn off downloads on your TikTok or your Instagram presence, and maybe even resist loading high-resolution images. Because high-res images are the perfect raw material for somebody wanting to misuse this. If you’re using ChatGPT, you might want to go into the settings and tell it, “Do not train on my data.”

Then, of course, schools as well have to have policies around mobile phone usage. It should not even be allowed, actually, for employees to use their phones in this manner. 

How do you balance protection from malicious misuse with the preservation of a creator’s rights?

I think this is, again, uncharted territory in many ways. We have to not stifle innovation and creativity, while at the same time making sure that this dark side is taken care of. So this is where consent filters, water marking and legal frameworks are evolving in different jurisdictions that will give, for example, my voice, property rights. 

And I think that's there's going to be innovation on that front. I know there are startups working on that area, as well as established players like Adobe, Microsoft — all of them are creating some of this technology. 

Is there a potential with synthetic media, where society might lose its grip on what is visually real? And if so, then what?

My view is that we are in this pre-seat-belt phase with cars, where we were liking this technology, but we are building the guardrails as we go along. I think a lot more innovation is going to happen.

I'll give you an example: Meta now has a paid feature — which is a problem for me; I think this should be rolled out to everybody — where if you’re creating content, if you’re Meta verified, you can say that this content should not be used for impersonation.

Why not roll that out to everybody? And therefore, when creators go out and create content, they have the rights to how it can be used. They have the rights to how it can be remixed. I wouldn’t mind maybe my content getting remixed, but I should get compensated for it, right? 

That’s the future I see coming in the next two to three years. 

Do you have a question about artificial intelligence? The Morning Edition team is digging into the rise of AI. What questions do you have? What topics interest you? E-mail us at tell@mpr.org.

Shows

Morning Edition

Cathy Wurzer

Host

Heidi Raschke

Author

Matthew Alvarez

AI analytics expert Ravi Bapna discussed harmful uses of artificial intelligence and the need for safeguards with Cathy Wurzer. 

AI-generated images raise questions about safeguards

Go Deeper.

Like this?