This project wasn’t just using data to locate or delineate a specific target audience. It used data to actually create an artificially intelligent online persona which could interact with a target audience of potential sex traffickers.

Gracie was also designed to collect data, logging every conversation to help Gracie become even more accurate and authentic, and gathering valuable insights on the behaviors of potential online traffickers. This data fueled the largest ever verifiable study on underage sex trafficking, which gives us ammunition to use in fighting for real policy change and action.

Child sex trafficking is a growing problem, with the internet making it easier for predators, harder for law enforcement, and nearly impossible for anyone to collect verifiable data on it. And with so little data available, securing funding and policy changes to fight it is extremely difficult at best. And since it’s such a difficult subject to talk about, awareness of the full extent of the problem is very low.

Gracie is an artificially intelligent online presence, a learning bot designed to mimic real underage kids being sold for sex. She intercepts predators who are searching for sex online, collecting valuable data on their behavior.

First, predators click on ads that we placed, leading them to a text chat with Gracie. Then, when they solicit underage sex from her via text, Gracie sends a warning to them and sends their information to law enforcement. And lastly, she directs them to resources where they can find help.

She is the first of her kind with 112 distinct, adolescent personas, all developed using transcripts of predator and victim interactions from the Georgia Bureau of Investigation, text chats from students at local universities, and guidance from the Centers for Disease Control. She’s able to accomplish in just a few minutes what would normally take hundreds of law enforcement officers working around the clock.

Online sting operations, which “out” potential sex traffickers and predators, have been a very successful weapon for law enforcement in fighting sex trafficking. And statistics have shown that 80% of first-time buyers do not repeat once they are outed.

However, each successful intercept of a predator requires an enormous amount of man hours and money, which just isn’t available. And the number of predators stopped by these operations, as well as the amount of data collected, is sadly low.

So, Street Grace set out to find a way to accomplish the same thing but in an automated fashion that could be easily scaled up and could exponentially increase the effectiveness of law enforcement, while collecting valuable data at the same time.

Gracie was released gradually over five months, starting in Georgia and spreading to more states. As she interacted with more and more predators each month, Gracie’s AI continued to learn from each conversation, making her more believable and authentic each day. Currently, she is performing over 1,000 intercepts per month.

Gracie is powered by AI tools including a natural language processing engine, a dialog interface, a conversational logic layer, and a transactional database. Everything from the ads we place to the web analytics is stored in a database made available to our research partners.

In just five months, Gracie has interacted with potential predators across 25 states and 78 cities, achieving undeniable results. She has exchanged over 54,000 messages, performed over 1,000 intercepts per month, and has reported over 6,000 would-be predators. Gracie’s therapy resource has a 9% clickthrough rate. She has also fueled the single biggest verifiable study on sex trafficking ever done, helping inspire real policy change, and becoming perhaps the greatest weapon we have to fight underage sex trafficking.

Most importantly, every conversation Gracie has with a predator is one less conversation a predator has with a real kid.

To understand the typical conversational structure of an escorting inquiry, we reviewed SMS transcripts from Street Grace’s manual conversations from their previous Project Intercept efforts. We also reviewed SMS transcripts from cases prosecuted by the Georgia Bureau of Investigation, and interviewed their special agents to learn about the psychology and tendencies of sex predators with whom Gracie would interact. To be sure that Gracie used the language of a teenager and exhibited the psychology and constraints of a typical teenager, we gathered SMS samples from a writing workshop with college students at the Savannah College of Art and Design. To ensure we collected data in the most impactful way, the Centers for Disease Control and Kennesaw State University consulted on our research design.

All of this together combined to make Gracie a totally convincing online presence, capable of believably impersonating a real teenager.