Home → Research → AI in Psychiatry → PAWS Digital Therapeutic

PAWS: AI-Powered Digital Therapeutic for Cannabis Use Disorder

By Ryan S. Sultan, MD
Assistant Professor of Clinical Psychiatry, Columbia University Irving Medical Center
March 28, 2026

The Problem: Cannabis Use Disorder in Youth

Cannabis is not the drug it was in the 1990s. The product has changed fundamentally, and the mental health consequences have changed with it.

THC Potency Has Transformed the Risk Landscape

In 1995, the average THC content of cannabis flower was approximately 4%. Today, flower products commonly contain 15-25% THC, and concentrates -- wax, shatter, dabs, vape cartridges -- routinely reach 80-90% THC. This is not a minor change. A young person using cannabis today is consuming a pharmacologically different substance than what was available a generation ago.

The clinical consequences of this potency increase are substantial and well-documented:

• Daily use of high-potency cannabis increases psychosis risk approximately 4-fold, according to the landmark multicenter study by Di Forti and colleagues published in The Lancet Psychiatry
• Cannabis use disorder rates among young people are rising, driven in part by higher-potency products that produce faster tolerance and more severe withdrawal
• The adolescent brain is particularly vulnerable -- the endocannabinoid system plays a critical role in neurodevelopment, and high-dose THC exposure during this window can produce lasting changes in cognitive function, emotional regulation, and risk for psychiatric disorders
• Cannabis-induced psychosis is increasingly common in emergency departments, particularly among young men using high-potency products daily

This is the clinical reality I confront in my cannabis research at Columbia. The data are clear: today's cannabis products carry genuine psychiatric risk, particularly for youth, and we are not doing nearly enough to address it.

The Treatment Gap

Despite the growing burden of cannabis use disorder among young people, the treatment system is failing to meet demand:

• Less than 10% of adolescents and young adults with cannabis use disorder receive any form of specialty treatment
• Stigma remains a major barrier -- many young people do not perceive cannabis use as a problem, and the cultural normalization of cannabis makes it harder for those who are struggling to seek help
• The mental health workforce shortage means that even when young people want treatment, they often cannot find a provider with availability, particularly one trained in adolescent substance use
• Traditional treatment models -- weekly in-person therapy with a trained clinician -- are effective but not scalable to the magnitude of the problem
• Geographic barriers exclude rural and underserved communities from accessing specialty substance use treatment

This gap between need and capacity is what motivated the development of PAWS. We need solutions that can reach young people where they are -- on their phones, in real time, between clinical encounters -- without replacing the human clinician at the center of care.

What PAWS Is

PAWS -- Preventing Adverse outcomes With Screening -- is a digital therapeutic built around a large language model-powered conversational agent designed specifically for youth experiencing cannabis use disorder.

I am developing PAWS in collaboration with Xuhai "Orson" Xu, PhD, at Columbia's Department of Biomedical Informatics (DBMI). Orson brings deep expertise in human-AI interaction, conversational AI systems, and health behavior change. He is also visiting faculty at Google Research, which means PAWS benefits from both academic rigor and state-of-the-art AI engineering.

How the LLM-Powered Agent Works

The core innovation of PAWS is its conversational agent -- not a rule-based chatbot that follows scripted decision trees, but a large language model fine-tuned for clinical interaction with young people experiencing substance use problems. The agent is designed to:

Why a Digital Therapeutic

The rationale for a digital approach to cannabis use disorder treatment is grounded in several clinical realities:

• Scalability: A digital therapeutic can serve thousands of users simultaneously, addressing the workforce shortage that limits access to traditional treatment
• Youth engagement: Young people spend an average of 7+ hours per day on screens. Meeting them in a digital environment they already inhabit reduces barriers to engagement
• Real-time monitoring: Unlike weekly therapy sessions that rely on patient recall, PAWS can capture data continuously, identifying patterns and risk factors as they emerge
• Between-visit support: The highest-risk moments for substance use often occur between clinical encounters. PAWS provides support when patients need it most -- late at night, on weekends, during social situations where cannabis is available
• Reduced stigma: Many young people who would not walk into a substance use treatment program will engage with an app on their phone

The NIH Funding

PAWS is funded by the National Institutes of Health through a UG3/UH3 mechanism totaling $335,500. This is a significant endorsement from NIH for several reasons:

• The UG3/UH3 mechanism is specifically designed for projects that require an initial development and feasibility phase (UG3) before proceeding to a larger clinical trial (UH3). This milestone-driven structure ensures that only projects demonstrating feasibility progress to the expensive trial phase.
• The funding comes through NIDA (National Institute on Drug Abuse), reflecting the agency's recognition that digital therapeutics for substance use disorders are a high-priority area of innovation.
• NIH peer review of the PAWS application evaluated the scientific rigor, clinical significance, team qualifications, and innovation of the project -- a validation that this approach meets the highest standards of biomedical research.

The Advisory Board

One measure of a research project's credibility is who agreed to advise it. The PAWS advisory board includes some of the most distinguished researchers in addiction neuroscience, psychiatry, and clinical trials:

Scientific Advisors

Clinical Collaborators

This advisory structure is intentional. Digital therapeutics for substance use disorders must be grounded in the neuroscience of addiction (Nestler, Krystal, Hurd), the clinical realities of youth treatment (Gray, Levy, Walkup), the methodological rigor of clinical trials (Wall, Kalin), and the developmental psychiatry perspective (Kalin, Walkup). PAWS has all of these.

Why This Approach Is Different

The mental health app space is crowded with products that make large claims and provide little evidence. PAWS is different in several fundamental ways:

The Broader Context: Digital Therapeutics in Substance Use

PAWS builds on a growing evidence base for digital therapeutics in behavioral health. The FDA has already cleared digital therapeutics for substance use disorders -- reSET and reSET-O by Pear Therapeutics (now acquired) demonstrated clinical efficacy in randomized controlled trials. These products established that software can function as a regulated medical intervention.

What PAWS adds to this landscape is the power of large language models. Earlier digital therapeutics relied on rule-based systems and fixed content libraries. PAWS can generate personalized responses in real time, adapt to conversational context, and engage users in a way that feels more like talking to a knowledgeable person than navigating a menu. This represents a genuine step change in the clinical utility of digital interventions.

The future of substance use treatment will involve a combination of in-person clinical care, medication when appropriate, and digital tools that extend the reach of the treatment system. PAWS is designed to be a key part of that future for cannabis use disorder -- the most common illicit substance use disorder among young people and one of the least well-served by existing treatment options.

Connection to My Cannabis Research Program

PAWS is one component of my broader cannabis research program at Columbia. My other cannabis research includes:

• CASNY (Cannabis Screening for New York Youth) -- a validated screening instrument for adolescent cannabis use
• Cannabis psychosis research -- epidemiological studies on the relationship between cannabis potency and psychosis risk
• JAMA publications on cannabis policy and clinical implications
• NYP Advances presentations on cannabis and adolescent mental health

PAWS represents the translational arm of this program -- taking what we know from epidemiological and clinical research and applying it directly to patient care through technology.

Frequently Asked Questions

What is PAWS?

PAWS (Preventing Adverse outcomes With Screening) is an NIH-funded AI-powered digital therapeutic for youth with cannabis use disorder. Developed at Columbia University's Department of Biomedical Informatics, it uses a large language model-powered conversational agent to provide personalized, evidence-based support between clinical encounters. It is not a chatbot dispensing generic advice -- it is a clinically informed AI companion designed to work alongside human treatment providers.

How does the PAWS AI conversational agent work?

The PAWS agent uses a large language model fine-tuned for clinical interaction with young people experiencing cannabis use problems. It delivers motivational interviewing techniques, monitors use patterns and triggers in real time, provides personalized coping strategies, tracks mood and functioning through brief daily check-ins, and escalates to human clinicians when clinical thresholds are crossed. The agent adapts its approach based on each user's engagement patterns and treatment stage.

Who funded the PAWS project?

PAWS is funded by the National Institutes of Health through a UG3/UH3 mechanism totaling $335,500. This milestone-driven funding structure requires demonstration of feasibility before proceeding to a larger clinical trial. The funding comes through NIDA (National Institute on Drug Abuse) and was awarded following rigorous NIH peer review.

Why is cannabis use disorder in youth a growing problem?

THC content has risen from approximately 4% in 1995 to 15-25% today, with concentrates reaching 80-90% THC. Daily high-potency use increases psychosis risk approximately 4-fold. Cannabis use disorder rates are rising among young people, yet less than 10% access specialty treatment due to stigma, workforce shortages, and access barriers. The gap between need and available treatment capacity is growing.

Who are the advisors on the PAWS project?

The advisory board includes Eric Nestler (Mount Sinai, NAS + NAM, h-index 216), John Krystal (Yale, NAM, EIC Biological Psychiatry), Ned Kalin (Wisconsin, NAM, EIC American Journal of Psychiatry), and John Walkup (Northwestern). Clinical collaborators include Yasmin Hurd (Mount Sinai, NAS + NAM), Kevin Gray (MUSC), Sharon Levy (Harvard/BCH), and Melanie Wall (Columbia biostatistics).

Further Reading

• AI in Psychiatry: Research Overview
• Cannabis Research Hub
• Research Collaborators
• The Sultan Lab at Columbia
• Publications