Annika Schoene

I am a Research Scientist at the Institute of Experiential AI (EAI) in the AI + Health Team and Responsible AI Practice and incoming Assistant Professor at the Bouvé College of Health Science at Northeastern University.

My current research focuses on developing Responsible AI methods and evaluation metrics to enhance the safety, security, and ethical integrity of AI systems, particularly large language models (LLMs). I investigate the robustness of LLMs in sensitive contexts, identifying and mitigating risks such as jailbreak vulnerabilities, harmful outputs, and unsafe behaviors. This work includes conducting comprehensive safety assessments, designing guardrails, and building evaluation frameworks that prioritize value alignment and the secure deployment of AI. My work contributes to broader efforts to ensure that AI  is not only effective but also safe, trustworthy, and beneficial – especially when deployed in complex domains such as mental health. I also gather evidence to support non-technical stakeholders and policymakers in making informed decisions that reduce algorithmic harm healthcare settings.

To this end, I collaborate with interdisciplinary research teams and have been appointed a Faculty Fellow position at the Institute for Health Equity and Social Justice Research as well as a Visiting Scientist at MaineHealth and the University of Southampton (UK). My research has been published in various peer reviewed journals and was featured in the press discussing the use cases of AI in the mental health sector with a focus on raising awareness of potential ethical, bias and fairness issues when deploying such systems in the real world.

Before joining EAI I was a postdoctoral fellow at the University of Manchester in the National Centre for Text Mining (NaCTeM), where I worked on NLP for mental health and Named Entity Recognition for both textual social media data as well as health science. I hold a PhD in Computer Science from the University of Hull (UK) and as a graduate student I worked on (i) accurate classification of sentiment / emotion in social media data and (ii) detecting suicide notes / suicidal ideation using her deep learning, machine learning and linguistic knowledge.