How social workers use AI to help unhoused teens | Anamika Barman-Adhikari | TEDxMileHigh
URL: https://www.youtube.com/watch?v=1E7BoJQf050 Video ID: 1E7BoJQf050 ============================================================ Transcriber: Gefi George Reviewer: Pari Szi Even when you might not know it, we have all experienced how artificial intelligence can make our lives easier and more fun. Machine learning algorithms are the reason why a horror movie will never cross my Netflix screen. It Schitt’s Creek, The Office or nothing else. These sneaky algorithms have a way of figuring out what I want even before I know it. Then they help me find those products and then those exact products, follow me from website to website, day and night. I’m still in a moment of weakness. I click the ad and purchase the product. But shopping and scrolling are just the tip of the iceberg. As a social work scholar, I have seen how artificial intelligence can help provide solutions to some of the biggest problems facing our society. Problems like homelessness, poverty, addiction. For the last seven years, I have been working with a computer scientist and technologist to develop algorithms that can help some of the most vulnerable members of our society. Young people experiencing homelessness. Among the countless issues that they experience, Substance use is almost an epidemic among this group of young people. Studies show that almost 50% of all young people experiencing homelessness use some kind of illicit substance. In comparison, less than 5 to 10% of young people, who do not experience homelessness do the same. Statistics can, however, be numbing. So let me explain to you why substance use is so pervasive among this group of young people in a manner that perhaps can be more relatable. Imagine you’re 18. You have nowhere to stay. You cannot call your mom or your dad. Everything that you own, you have to carry on you at all times, either in a backpack or a shopping cart. And you have to find a new place to sleep in every night. Take a look at this picture taken by a 19 year old girl experiencing homelessness right here in Denver. This is what it’s like. Sleeping on a dirty stained mattress. Against a coal cinder block wall. Trash littered everywhere. Use condoms and heroin needles on the floor. I would be terrified in a place like this, wouldn't you? You would do anything to escape this reality and nump your pain. And that’s what a lot of these young people do. They often use substances to cope with their trauma, to feel comfortable in an uncomfortable world. Now, researchers have tried to tackle this problem for decades using a method that sounds great on the surface: pure led groups. For 6 to 8 weeks, these young people come together to talk about their problems, get advice and positive support from their peers, and work on developing better coping skills. These groups are relatively inexpensive to operate, which is a big deal for cash strapped nonprofits who are often the ones running them. And the reason these groups work is because they’re designed to be developmentally appropriate. Adolescents are almost biologically hardwired to question, resist and doubt what adults tell them to do. If you're a parent or a teacher, you know what I'm talking about. But let’s be clear. Our friends can sometimes also be toxic. I mean, I literally smoked a banana peel once. (laughter) Because my friends told me that it would get me high and it was the cool thing to do. Any kids out there, including my own, please do not repeat the same mistakes I did. But you get the point. Let’s put a group of young people who might mimic each other’s negative behaviors in a group for 6 to 8 weeks. What could go wrong? A lot of you might have guessed it. Instead of going down. Substance use actually increases along with other negative behaviors, Until recently, researchers didn’t have a reliable way of predicting which groups would work and which groups had spiraled out of control. Social workers are usually the ones putting these group groups together, either assigning young people to these groups randomly or allowing them to self-select into those groups. Unfortunately, neither scenario is ideal. Let's take a look at this graphic from my research to better understand why. In case you’re wondering, this is a social network map of 160 young people experiencing homelessness in Los Angeles, California. Every dot on this picture is one young person experiencing homelessness. And the links between them are shown by the lines. The youth who are very popular and well-connected, find themselves at the centre of this network. And the youth who are not as popular or well connected are pushed to the periphery of this network. Now notice the red dots on this picture. These are the young people who use meth. And more importantly, notice how tightly connected they are. You know what that tells me? It tells me if we put these young people in a group together, Meth use will probably go up instead of down. Because they’ll just reinforce each other’s meth use behaviors, and also, the grey dots will catch up to what everyone else has been doing. Randomly assigning them to groups. It’s less harmful, but no better, because it ignores natural social influence. Who are you more likely to listen to? Your friends or some random stranger? So I just I started wondering, what if you could use artificial intelligence and machine learning to figure out what’s the best way to configure these groups? In order to do that, I partnered with researchers from USC’s Center for Artificial Intelligence and Society and got permission from a group of young people experiencing homelessness to use their social network data and basic personal information. Then we fed the machines this data about their social relationships, their substance use behaviors and other relevant behaviors. Then the machines would run iteratively testing 30 to 50 groups at a time. Finding out the best configuration. How did the algorithm do this? The algorithm found a way of mathematically maximizing the number of existing friendships in each group while minimizing the number of substance users. Not only that, the algorithm also found a way of accounting for how people’s relationships would change in these groups in response to the intervention, that positive relationships would develop and deepen, and negative relationships with subside and influence over the course of the intervention. You might ask, why can’t a person do this? Well, to start with, social services are notoriously understaffed and overworked. It would be extremely time consuming and mind numbing for a person to do all these gas calculations, especially for a large group of people. Second, a person, unlike a machine, cannot predict in advance which groups would actually reduce substance use. The results of the algorithm were astounding. compared to randomly allocated groups, these AI generated groups were able to reduce substance use by almost 40 to 70%. That’s right. 40 to 70%. That’s huge. Results like these could mean that social workers can use their time and resources more efficiently to help more youth faster. And for the youth themselves, getting sober can be life changing. Staying sober can be one of their pathways out of homelessness. Now, This is just example of many of how technology can improve social services. I’ll provide a few others. In Allegheny County, social workers are using a predictive analytics algorithm to screen for and triaged the most serious child abuse cases. Researchers in New York, are studying young people’s tweets to develop an algorithm to screen for signs of loss, aggression and substance use as a way of reducing youth violence. In one of my own studies, I’ve been using young people’s Facebook post to develop an algorithm that could predict their likelihood of developing substance use disorders. Compared to traditional methods that can predict substance use accurately, maybe 30% of the time, this algorithm can predict substance use almost 80% of the time. This level of accuracy could mean that social workers are better able to prioritize their caseload instead of expanding their limited time and resources on people who might not need it at that moment, they can instead help people who need it the most. And that’s the goal. This is not about machines replacing social workers. Algorithms are not perfect in the wrong hands. They can actually exacerbate bias and systemic racism. However done right, they can be a powerful tool in the hands of social workers to empower them to do what they do best, to use their intuition, their empathy, and their experience to help people overcome life's most difficult challenges. And that’s the kind of future I wish to see. And I now I hope you do as well. Thank you so much for your time. (Aplouses)