After the rollout of Chat GPT in 2022, Iftikhar became alarmed by the potential for humans to develop unhealthy emotional attachments to robots. Amid increasing numbers of young people who treat chatbots like therapists (with sometimes disastrous results), she broadened her research to study how AI can harm mental health. 

“My research has evolved into looking at the things computer scientists working in the mental health space are not aware of or not understanding as a community,” she said. “How can we communicate these issues from therapists to computer scientists?”

In a study currently under peer review, Iftikhar and Huang explored how AI may violate codes of conduct set by psychologists and the ethical issues posed by using standalone AI interventions to address mental health. 

In another project, Iftikhar tested whether deep learning models could predict whether participants felt empathy in peer support sessions. When she and Sara Syed (who graduated from Brown in 2022) applied a state-of-the-art empathy detection model to data from hundreds of real-life sessions in which participants said they experienced high levels of empathy, the AI tool failed — it concluded that there was no empathy present. 

“There is clearly some disconnect happening between what humans think therapy is and what AI thinks therapy is,” Iftikhar concluded.

Iftikhar and Huang are now collaborating with Nicole Nugent, a research scholar in the Department of Psychiatry and Human Behavior at the Warren Alpert Medical School at Brown, and other mental health professionals. The team conducted a study to learn how large language models can be aligned with the practice of cognitive behavioral therapy, then analyzed differences between human and AI-provided support. Their findings are informing an experiment exploring how human practitioners and AI can “collaborate” to deliver mental health support. 

According to Iftikhar’s research, AI can’t replace human therapists. Chatbots tend to over-agree with people and reinforce unhealthy beliefs about themselves and the world, she’s found. 

“AI is really bad at understanding what’s going on with a person and how to address their core issues,” she said. “That makes sense, because these models are trained on vast amounts of data which can dismiss people’s lived experiences.”

However, she said that AI can educate people about their mental health and offer coping strategies, like breathing exercises or journaling. And she acknowledged that one of AI’s main advantages is that it can reach many more people than the current health system.

“My goal is to clearly outline what an AI can do and what human mental health professionals can do, and how they both can support each other in scalable mental health treatment,” Iftikhar said.

Finding the essence of empathy at Brown

Iftikhar has experienced firsthand the power of human-to-human collaboration and empathy with faculty and peers in the Department of Computer Science at Brown and in the graduate student community.

“The computer science department is extremely collaborative,” she said. “I can go to any professor’s office hours and they’ll point me in the right direction. And I really appreciate my colleagues’ willingness to answer a Slack call for help — even at 1 a.m.”

Because even the most well-supported scholars need breaks and benefit from connections to humans and nature, Iftikhar teamed up in 2023 with Andrew Hollis, associate director of graduate student activities and leadership development at Brown, on a plan to get graduate students outside. 

Iftikhar completed a wilderness first aid training course and served as an inaugural trip leader for Brown’s Graduate Outdoor Adventure Leader Training program. She and other organizers expected around five students for the first trip and were surprised when more than 30 showed up. They’ve organized hiking trips around New England and plan to expand to camping and backpacking trips this year. There are now about 20 graduate students trained to lead treks. 

On campus, Iftikhar mentors students in Huang’s lab, and to her delight, three of them have decided to pursue post-graduate studies.

“Showing students that research can be a fun learning experience has been very motivating and fulfilling for me,” she said. 

She’s still interested in pursuing training in social work or psychology, but Iftikhar would eventually like to teach at the university level. This way, she said, she can conduct computer science research and, importantly, continue to advise and mentor other curious human minds.