Tellmi Artificial Intelligence Project
Artificial Intelligence gets a lot of bad press, yet new technologies have the potential to help us all do more, better, faster. We began introducing AI into Tellmi in 2023 as part of an Innovate UK funded project which was designed to help us provide better, faster, support to young people experiencing suicidal ideation. Most young people who experience suicidal thoughts will never act on them, but some do, and for those young people, the time between thinking about suicide and acting on those thoughts can be as little as ten minutes (Deisenhammer et al., 2009).
Suicide-specific help and support needs to be delivered as rapidly as possible, and it also needs to be easy to access because young people in distress don’t have the emotional and practical capacity to browse through webpages to find such information (Cohen et al., 2021).
“At Tellmi we are constantly testing new ways to make our service faster, safer and more effective and AI is part of that process.”
In collaboration with data scientists from the University of Sussex, we began to explore how machine learning and natural language processing could help us to improve the support that Tellmi provides to young people experiencing suicidal ideation. We were particularly interested in finding ways to speed up moderation times.
Prior to the project, a post from a suicidal young person would sit in a queue with all the other posts until a moderator got to it. Once a moderator had seen it they had to risk assess and topic tag it, which meant selecting multiple tags from a list of over 200. It was a time consuming and not always 100% accurate process. Once a high risk post was risk assessed and tagged, it was passed to a counsellor who would respond privately, but the time between the post being made and the counsellor responding could take up to an hour, which was obviously far too long.
Through the course of this project, machine learning and natural language processing were trained to identify the topics that relate to an individual post so that they could be served up as suggested topics to the moderator. The moderator can then accept, or edit the topics, in a fraction of the time that it used to take them to tag a post. We estimate that the new AI topic tagging system has doubled our moderation capacity and increasing the speed of moderation means that high risk posts get to counsellors much more quickly.
“As part of the this project Tellmi built AI powered smart signposting that makes it much easier for all Tellmi users to access local, national and web-based services.”
In collaboration with the university of Bristol, clinicians and young people with lived experience of suicidal ideation, we explored what else we could do to deliver support. Previous research has also shown that individuals with lived experiences of suicide want signposting to include a variety of local, national and web services, including 24-hour services (Carrote et al., 2021) and the Tellmi workshop confirmed this. When someone is very distressed, the last thing they need is to have to trawl through a huge list of services to find something relevant. As part of the this project Tellmi built AI powered smart signposting that makes it much easier for all Tellmi users to access local, national and web-based services. Once Tellmi AI has identified the topics that relate to a post, it can link the post to topic related services that are housed in the Tellmi directory. When the post is published in the feed, all users have to do is click on the words ‘see related resources’ that appears at the bottom of their post. This takes them to a directory of curated services and support that are personalised to them and to the issues that they need help with.
Tellmi is now working on a number of other AI enhanced features. We have begun building models to interrogate passive, as well as high risk data to identify risk patterns, analyse semantic content and predict risk. The model that we have developed is now being trained to predict the risk levels of posts that have been made by young people, but that have not yet been seen by a moderator. Once our model can determine the risk level of a post with sufficient accuracy, it will be able to intercept high risk posts and pass them straight to a counsellor, bypassing the moderation process entirely and exponentially speeding up access to support to young people in distress.
At Tellmi we are constantly testing new ways to make our service faster, safer and more effective and AI is part of that process. We will never remove our human moderators and counsellors, but improving our technology to make their jobs easier helps us to be more efficient. That enhances scalability and ultimately makes Tellmi a more cost-effective and therefore a more sustainable mental health solution.
References
Deisenhammer, E. A., Ing, C. M., Strauss, R., Kemmler, G., Hinterhuber, H., & Weiss, E. M. (2009). The duration of the suicidal process: how much time is left for intervention between consideration and accomplishment of a suicide attempt?. The Journal of clinical psychiatry, 70(1), 19–24.
Cohen, Z. P., Cosgrove, K. T., DeVille, D. C., Akeman, E., Singh, M. K., White, E., Stewart, J. L., Aupperle, R. L., Paulus, M. P., & Kirlic, N. (2021). The Impact of COVID-19 on Adolescent Mental Health: Preliminary Findings From a Longitudinal Sample of Healthy and At-Risk Adolescents. Frontiers in pediatrics, 9, 622608. https://doi.org/10.3389/fped.2021.622608
Carrotte, E., Webb, M., Flego, A., Vincent, B., Heath, J., & Blanchard, M. (2021). Acceptability, Safety, and Resonance of the Pilot Digital Suicide Prevention Campaign “Better Off With You”: Qualitative Study. JMIR Form Res, 5(3):e23892. https://doi.org/10.2196/23892