Stay Updated

AI and Mental Health: Reflections from the Outside Looking In

By Dr. Andy S. Jagoda, Professor and Chair Emeritus of Emergency Medicine, Icahn School of Medicine at Mount Sinai and Member of DocGo’s Medical Advisory Board

I am not a mental health specialist, but I have a hard-earned perspective on this tough topic. I’ve spent my career in emergency medicine, and I’ve often encountered people at moments of acute mental distress, when the gaps in our broader system become impossible to ignore. Over the past few years, these gaps have become more urgent to address, exacerbated by COVID and the subsequent wave of social, financial, and emotional stress.

Today, access to mental health care is largely determined by geography, insurance status, and socioeconomic resources. As a result, far too many people who need help fall through the cracks in our system. That’s why I find the current interest in AI-enabled mental health tools so significant. It offers real potential to connect people to the care they need, an especially attractive option if the only alternative is silence or delay.

Where AI Shows Promise

To be clear, AI’s greatest promise in mental health lies in serving as a complementary layer, supporting screening, symptom monitoring, and access to care between clinician visits. Traditional self-reporting methods and lengthy assessments depend on a person’s ability to articulate feelings, which is often shaped by social stigma, self-awareness, and the availability and cost of provider visits. 

Chat-based AI tools can offer more consistent, scalable, and 24/7 support, backed by Large Language Models from OpenAI, Anthropic, and Google. Other platforms can detect patterns in speech, behavior, and mood that may be missed in a single session. Providers like Ellipsis Health go further, analyzing vocal biomarkers, both semantics (what is said) and acoustics (how it is said, including tone, pitch, rhythm, and tempo), to reveal cues about a person’s internal state. However, these systems still lack sensitivity to context. They can’t yet interpret body language or shifting life circumstances the way a human therapist can.

Early research in this space is encouraging, particularly for people dealing with mild to moderate symptoms of anxiety or depression. A number of companies are developing web-based CBT tools that have been shown to reduce depressive symptoms in young adults within just a couple of weeks, offering outcomes comparable to short-term, human-delivered therapy.

The Limits of AI in Mental Health

There are clear limits, nevertheless. More complex or high-risk conditions, including PTSD, suicidality, and severe depression, demand something these tools cannot yet provide. Traditional therapy, with its ability to build trust and respond in real time to emotional cues, remains more effective for those cases. In fact, decades of clinical literature show that the therapeutic relationship itself accounts for a significant portion of positive treatment outcomes. That’s not something you can replicate with machine-driven pattern recognition alone.

A Historical Parallel: Evolving to Address Unmet Needs

This evolution reminds me, in some ways, of how emergency medicine became a specialty. In the 1960s and 70s, people with acute medical needs would arrive at general hospital admissions, but hospitals lacked the specialized systems (such as an Emergency Department) and expertise (Emergency Doctors) to effectively triage and treat these critical cases. Over time, emergency medicine developed as a field specifically built around filling that gap. Mental health today faces a similar situation: people need help but the existing system struggles to adequately serve them. Just as emergency medicine evolved new approaches to address unmet care needs, I believe AI can be part of the evolution of behavioral therapies, especially if it helps us guide people more quickly to the right kind of care.

AI in Action

Imagine a patient who doesn’t know where to turn, but who can use an AI system to ask a personal question, something like “I’ve been drinking more than I used to, and I’m worried about it.” AI could engage in a brief, evidence-based screening conversation to assess the severity and context of their concern. Then the AI could provide personalized recommendations, perhaps connecting someone with mild concerns to peer support groups or relevant resources in their community, while directing someone showing signs of more severe withdrawal or addiction to options for clinical evaluation, all filtered by their insurance coverage and preferences. That interaction may not solve the problem, but it’s a good place to start.

There’s also growing interest in more integrated models: hybrid systems that pair AI-driven tracking with clinician oversight. Intake might start with a digital screener that scores risk and suggests levels of care, before a clinician reviews and sets the plan. Between sessions, an AI coach can run quick check-ins, reinforce CBT skills, and track mood, sleep, and triggers, providing the team with concise summaries before each visit. These approaches can ease some of the administrative burden on providers, support earlier detection of problems, and allow clinicians to spend more time on relational care.

Making Care Available to Those Who Need It Most

The integration of AI into mental health is still in its early stages, but the potential is significant. If we focus on improving access while preserving the quality of care and the therapeutic relationship, AI may help us reach people who have gone without mental health support for too long. Just as emergency medicine evolved to meet unaddressed medical needs, mental healthcare may be on the verge of its own transformation, using technology to help make care more available to those who need it most urgently.