Bots Against Slavery
See how Seattle Against Slavery is fighting sex trafficking with chatbots that pose as sex trafficking victims and that interact directly with the buyers - By Liz Rush
September 4, 2018
Here at Seattle Against Slavery, a nonprofit dedicated to combating labor and sex trafficking, we’ve been working on a new strategy to reduce sex trafficking—launching chatbots, or bots, that pose as sex trafficking victims and that interact directly with the buyers of trafficked sex.
Project synopsis: Deterrence messaging for prevention
Our bot platform has two main purposes: to help gather information about the prevalence of sex buyers in a given area and to send out deterrence messaging. These “bots for good” connect with potential buyers by posing as sex trafficking victims and engaging with the buyers via text message. The bots are programmed to understand and respond to questions about its age, body type, location, and what sexual services it can provide.
When the bot determines intent to buy sex, it sends out a deterrence message with a link to a website that helps explain the realities of sex trafficking. The site also gives buyers resources for help changing their behavior, such as links to counseling services or sex addicts anonymous groups.
The bot platform, in conjunction with our victim outreach platform, provides a dual approach to tackling the issue of sex trafficking. It helps those victimized by the trade, while simultaneously making an effort to reduce the demand for trafficked sex.
Where our bots began
Our bots were developed as part of the Hack For Good Hackathons, where Microsoft employees brainstormed ways to use technology to disrupt those who seek to buy sex from trafficking victims. By using the Azure stack and Microsoft’s LUIS.ai bot framework, engineers were able to rapidly prototype several bot personalities that would interact with potential buyers. The hackathon work has since grown into a platform with active customers, such as nonprofits that offer services to victims and survivors, community NGOs, and law enforcement agencies.
In January 2018, as part of the National Johns Suppression Initiative led by the Cook County Sheriff’s Office (in Illinois), Seattle Against Slavery launched 25 bots in five regions to pose as trafficking victims.
Our bots posed as trafficking victims with different personalities and physical features. With names like Brooke, Emilynn, and Cari, the bots interacted with more than 9,000 unique potential buyers in one month. And over 60 percent of those potential buyers received the deterrence message.
Ethical considerations for bots that pretend to sell sex
When most people consider the average use case for bots, they picture a bot which helps customers. A company’s bot might answer simple questions when a user seeks customer support, or it might help a shopper buy something on a website. Your favorite bot might be one that helps you find the perfect GIF for your Slack channel. These bots tend to be straightforward.
But the recent boom of conversational bots has led many to explore the ethics of bots. Most folks start with Isaac Asimov’s “Three Laws of Robotics” and extrapolate from there. In March 2016, Amir Shevat, the VP of Developer Experience at Twitch, wrote “The bot rulebook,” which is a wonderful and simple introduction to some ethical bot design principles.
One of the most common ethical guidelines for bot-making is bot transparency—the design principle that a user should know when it is interacting with a bot and not a real human.
We realized early on that our bot would not be able to disclose its true nature. Although a customer service bot can surely disclose its botness and continue to serve the user, a bot posing as a trafficking victim would be rendered useless if it were to tell the user that it is a bot during their first interaction. By definition, the buyers who interact with our bot are seeking to connect with real people from whom they can buy sex, unaware of whether they’re interacting with a bot, a law enforcement officer, or a trafficking victim.
We often get questions about how potential buyers begin talking with our bots. The bots themselves never initiate contact. You’ll never get an unsolicited text from one of our bots.
When one of our bot numbers is posted online, potential buyers choose to text the number and begin a conversation. The bot won’t text them unless they text it first. The bot is a responsive conversationalist and follows the lead of the buyer. The buyer can walk away from the conversation at any time, and many do.
We also get questions about whether this type of bot is a form of entrapment. Although it’s true that many people will interact with our bot and indicate their intent to break the law, our bot never entraps potential buyers. In most cases, entrapment requires two components: government inducement of the crime and the lack of predisposition to engage in the criminal conduct. Of course, we are not government agents and there is no real crime in texting with our bot about purchasing sex.
Our bot design is based entirely on the predisposition of buyers to engage in criminal conduct—if there were no predisposition to buy sex from trafficking victims, we’d have no need for the bot and the world would be a much better place! By making our bots something that buyers must seek out—buyers must find the websites on which to buy sex, choose the ad with our bot’s phone number, enter that number into their phone, and then text it—we ensure that the buyers interacting with the bot are truly those who intend to buy sex from our decoy trafficking victim.
When tackling the issue of sex trafficking and exploitation, reaching out to trafficking survivors and local service providers was key to our design. By listening to survivors’ experiences and integrating their input and knowledge, we were able to create bots with the language and understanding to represent the lived reality of those who have experienced sex trafficking.
Although there are many tools and services available to integrate natural language processing–powered (NLP-powered) AI into new tech projects, one thing that these services cannot provide is the insight to ensure that the created designs reflect reality. We partnered with survivors and local service providers to develop the language and mannerisms of the bots. With a deep understanding of the vocabulary and techniques used by victims of trafficking, our bots are able to accurately reflect how the conversations develop. This maintains a believability so that buyers aren’t aware that they are interacting with a bot.
Learning the language
One key design element in our bots is fallibility and human-like conversation skills. We avoid over-engineering or over-optimizing the bot, since that’s just not the way real people text. Our bots have variable response times that are never immediate, and the texts include typos, punctuation errors, and slang that give them a lifelike feel.
Although the bots use NLP to understand and respond to potential buyers with whom they text, we chose to not have our bots learn language to use from the conversations in which they engage. We want to avoid the possibility that people interacting with the bots can influence the behavior or language. This means that a buyer can send texts with terminology or slang that the bot doesn’t understand but instead of attempting to learn that new term, we let our bot gracefully fail. The bot states that it doesn’t understand and then redirects the conversation back to topics that it does comprehend.
By restricting the bot so that it is only learning new language from our input and direction, we ensure that our bots can never become hijacked by malicious interactions intended to teach it to say inappropriate things.
What’s next for the Seattle Against Slavery bots
Thus far, our bots have been developed around the profile of female sex trafficking victims, with multiple profiles, such as underaged, inexperienced, or experienced. With the expertise of our partner organizations, we’re now looking at developing new features for our platform to reach other victims. To expand the impact that we can make with this technology, we’re looking at models for reaching male and trans victims of sex trafficking.
We’re also working on developing more robust analytics reporting features. By understanding the behavior patterns of buyers, we can better interrupt their sex-buying habits. We are also working on making the deterrence messaging and support content more effective by tracking the interaction and follow-through responses to our bots.
Unlike most bots that aim to keep customers engaged, the ultimate goal for our organization is to help create a world in which we get no response to one of our bots posing online as an underage sex trafficking victim. Until then, the interactions of our bots with potential sex buyers can help us track the changes in online sex buying habits, gather data about the prevalence of online sex buying, and help shift the cultural attitude around buying sex from trafficking victims.
About the Author
Liz Rush is the Technology Director at the non-profit Seattle Against Slavery, where she oversees the implementation, development and adoption of anti-trafficking software to reach victims and reduce the demand for trafficked sex. She previously founded a web & mobile software agency and has several years experience in Seattle's start up and small business community.
About Seattle Against Slavery
Seattle Against Slavery is a grassroots coalition working to mobilize our community in the fight against sex and labor trafficking. Driven by the belief that the rights and dignity of every person should be respected, we build partnerships that unite passionate volunteers and advocates with local and national non-governmental organizations and government agencies. Together with our partners, we work to address the root causes of sex and labor trafficking, while also advocating for the rights of survivors. SAS seeks to raise awareness and create comprehensive solutions to prevent human trafficking, better advocate for victims, enhance survivor care, and build capacity for direct service providers.