This Nonprofit Is Using AI to Power 24/7 Counseling at a Moment’s Notice

Since the introduction of the National Suicide Prevention Lifeline in 2005, the number one way in which Americans communicate has gone from calling to texting, creating tension for those who may feel more natural disclosing sensitive information through a message rather than a voice call. In this episode of Mastering Innovation on SiriusXM Channel 132, Business Radio Powered by The Wharton School, Nancy Lublin, Founder and CEO of Crisis Text Line, discusses the use of artificial intelligence to help triage millions of texts, the platform’s integration into popular chat services such as Snapchat and Facebook Messenger, and their extensive team of devoted volunteers across the country.

Today, people are more connected than ever before, yet many people find that they don’t have a comfortable place to talk about their mental health or tough situations. After running the nonprofit Dosomething.org for 12 years, Nancy noticed that the organization was receiving texts for campaign suggestions that had nothing to do with the campaign, instead touching on issues of drug addiction, anxiety, or problems at home. Realizing that some young people had no place to safely divulge their personal struggles, she began the Crisis Text Line. With a network of more than 4700 volunteers, anyone can quickly receive counseling and defuse a potentially harmful situation. With the implementation of artificial intelligence, the Crisis Text Line is able to analyze users’ language and flag those who are likely to be in immediate danger.

To reach the Crisis Text Line, text “HOME” to 741-741 for free, confidential help, available 24/7.

An excerpt of the interview is transcribed below. Listen to more episodes here.

Transcript

Nancy Lublin (Founder and CEO, Crisis Text Line)

Nicolaj Siggelkow: I can imagine this is an amazingly rewarding but also stressful job to be sitting on the receiving end of these texts. How do you train your volunteers? How do you support them in training and responding?

Nancy Lublin: One of the best innovations about Crisis Text Line is this incredible volunteer corps. In the last 28 days, 4,712 crisis counselors have been on our platform. That is a massive volunteer corps. It’s the second largest distributed network of volunteers after Wikipedia; it’s next-generation volunteering. Otherwise, if you want to volunteer, you have to be based in a city and available during business hours.

For us, these are people at home in their jammies on their couch. Our volunteer corps is totally distributed. You apply online and go through a background check and a 30-hour training, but not everybody passes. It’s only about a 30% passage rate. This is hard. It would be easier to get most degrees in America. You can become a crisis counselor with us if you meet those high standards. These counselors are veterans. We have two dozen people over the age of 70, we have stay-at-home moms, and we have a lot of people in college and graduate school who were thinking about careers in counseling and find that this is a great way to test it. It’s a phenomenal way to volunteer. Our crisis counselors (I’m just going to put this out there, Nicolaj) are the best people in America. They are the most incredible people, and we love them.

Siggelkow: You also have some professionally trained counselors. Is that right?

Lublin: We have supervisors, full-time staff, who have a master’s degree in a related field like social work or psych, and we have one minister. The supervisors are watching all the conversations in real time. As a texter, you’re being viewed first by an algorithm and then by the crisis counselor and supervisor. As a crisis counselor you’re not alone either. You’re set with other crisis counselors and the supervisors are watching all the conversations in real time. When something is imminent, when someone has met the ladder-up risk assessment (that means they have the ideation, the plan, the means, and the timing to commit harm to themselves or someone else) those supervisors are the ones who call in the active rescue to 911. They’re the ones who make that decision and call for help. We’re doing that on average 32 times a day.

Siggelkow: Give us a couple more numbers, because they are staggering, of the number of texts you’ve been dealing with, to give us an idea of the scale.

Lublin: 32 times a day is still well under 1% of our volume every day. It’s tiny. Those are edge cases. 99.5% of the time, we successfully help you feel better in that conversation and point you in the direction of self-care and a plan to keep you safe. It’s only about 0.5%, a half a percentage point at the time, when we have to call in an active rescue. The volume is staggering. We’ve processed over 87 million messages. We were in all 295 area codes in the United States very quickly, all through organic growth. By the fourth month, we were everywhere. I have lots of numbers; what numbers do you want to talk about? We love data here.

“99.5% of the time, we successfully help you feel better in that conversation and point you in the direction of self-care and a plan to keep you safe.” – Nancy Lublin

Siggelkow: I want to talk about the data a little bit later, but I want to first take you up on the word “algorithm.” What allowed you to scale, triage, and help train is the use of technology, right? This is not, “We’re sitting in a room and waiting for the text to come in.” Tell us about the technology side of Crisis Text Line.

Lublin: We built this on top of Twilio. We love the folks at Twilio, but we built all this technology. About a third of our team here writes code or is in product and data. You used some magic words there. You talked about the algorithm and triaging. Let me explain how that works. We function more like a hospital emergency room. In the hospital emergency room, if you come in with a gunshot wound or a heart attack, they’re going to wheel you right in first. The kid with a sprained ankle from the soccer game might wait an hour. We think that’s how a hotline should work also.

When we first built this, we put in some words into the algorithm, “die,” “suicide,” “overdose,” “gun,” and said if those words come in, we should take those conversations first. We did, and it worked pretty well. We took those high-risk people in under two minutes. That was pretty good. Then when our corpus grew to about 20 million messages, we layered on machine learning and said, “What really is high-risk? When do we have to call 911?” and we found some false positives.

For example, the word suicide led to some false positives because people were texting and saying, “It’s not like I’m suicidal or anything.” The algorithm would pick up the word “suicide” and make them number one in the queue. It also found thousands of n-grams, bigrams, and trigrams. That’s where the word combinations are more high-risk than the word “suicide.” For example, if you text in, “I’m in the military,” it’s twice as likely that that conversation will end up with us having to call 911 than if you text in, “I wanna kill myself.” The word “military” is a more lethal word, sadly, in America than the word “suicide.” The unhappy face crying emoji makes it four times as likely that we’re going to have to call 911 than the word “suicide.” The word “gun” makes it four times as likely that we’ll have to call 911 than the word “suicide.” You want to guess the most lethal words in America that we see?

Siggelkow: I will not dare to guess.

Lublin: What’s the most common drug in your house, Nicolaj?

Siggelkow: It’s the Ibuprofen.

Lublin: Exactly. Ibuprofen, Advil, Tylenol, Flanax, Percocet, any named pill, any named drugs, are the most lethal words that we see. Someone not only has the ideation, but they clearly have a plan. They likely have the means, because these pills are all around us, and the timing, because it’s imminent, they are thinking about this, or someone’s already swallowed a bottle of something. Those are some of the most lethal words that we see, and we put all of that information into the algorithm. Now, we take in our high-risk people, on average, in under 38 seconds.

About Our Guest

Nancy Lublin does not sleep very much. She is currently the Founder & CEO of Crisis Text Line, which has processed more than 82 million messages since 2013 and is one of the first “big data for good” orgs. She was CEO of DoSomething.org for 12 years, taking it from bankruptcy to the largest organization for teens and social change in the world. Her first venture was Dress for Success, which helps women transition from welfare to work in almost 150 cities in 22 countries. She founded this organization with a $5,000 inheritance from her great-grandfather. Before leading three of the most popular charity brands in America, she was a bookworm.

She studied politics at Brown University, political theory at Oxford University (as a Marshall Scholar), and has a law degree from New York University. She is the author of 4 books and is a board member of McGraw Hill Education and The Truth Initiative, and board chair for Change.org. Nancy is a Young Global Leader of the World Economic Forum (attending Davos multiple times), was named Schwab Social Entrepreneur of the Year in 2014, and has been named in the NonProfit Times Power and Influence Top 50 list 3 times.

Leave a Reply

Your email address will not be published. Required fields are marked *