I’ve been a clinical psychiatrist and a professor of psychiatry at NYU Medical Center for nearly 50 years.  I’ve witnessed the early days of my profession when the two leading antidepressants were Elavil and Tofranil.  Patients could barely tolerate either of the drugs because of the debilitating side effects.  In addition, psychotherapy or ECT (electroconvulsive therapy) were the only alternatives.  No single modality was adequate and imperial studies in psychiatry were far and few between.

Most of my early clinical work was “hit and miss” with little guidance from fact based well designed studies.Beginning in the early seventies and eighties pharmaceutical companies developed and host of drugs for depression, psychosis and other psychiatric disorders. And yet, even with these new pharmacological tools, determining which drugs and therapies to use was difficult.

I observed my colleagues using poly-pharmacy as their mainstay.  Patients would be placed on one drug. If the response was marginal, another was added. Then another. This cycle left the patient with an array of medications to take yet little or no clinical response.

These realities validate the clinician’s experience of working 'in the dark.’ They document how the information gap drives unsafe delays and the inappropriate use of polypharmacy(Correll et al., 2024; Sabe et al., 2025). Furthermore, they highlight the urgent need for an AI‑augmented, zero‑friction clinical thought partner. We must bring the patient’s lived data into the clinical decision-making process.  This approach expands the doctor’s understanding of the patient and prioritizes evidence based data rather than “reactive” approaches that generate unnecessary prescriptions based on the isolated fragments of information reported in weekly meetings.

The Core Problem: Operating in the Dark

Clinicians see a tiny fraction of a patient’s life; most of the illness narrative unfolds outside the clinic and the sparse weekly meetings with the psychiatrist provides the illusion of understanding the patient’s weekly experience. We, therefore, make high‑stakes decisions from fragmented snapshots rather than a continuous “movie of daily experience.” Critical mood shifts, medication side effects, and functional declines often occur days or weeks before the next visit, which is precisely when early intervention would be most effective.

Antipsychotic polypharmacy affects roughly 24.8% of patients on antipsychotics, and psychotropic polypharmacy rates approach 29.3% among adult psychiatric outpatients, complicating assessment and safety monitoring (Correll et al., 2024; BMC Psychiatry, 2025). Additionally, about 4.8% of U.S. adults report regular depressive symptoms, underscoring the population burden clinicians face (Centers for Disease Control and Prevention, 2024). Delays in follow‑up after crises, often on the order of 11 days, are associated with measurable increases in two‑year mortality (Krebs et al., 2024). This illustrates how dangerous the "in‑between" void can be for a busy psychiatrist who is further burdened by the restrictions of third-party payers.

From the Front Lines: What I Have Seen

Across five decades I have repeatedly watched deterioration go unnoticed until it became an emergency.

For example, Ms. A, a 35-year old mother of two, who comes to see me on an emergency basis.  She claimed that her psychopharmacologist was treating her for anxiety and depression but to no avail.  Her doctor rarely spoke to the social worker conducting the patient’s “therapy” because of the busy nature of his practice.  This patient describes weeks of worsening sleep, slowed activity, or new restlessness that were not captured in either the therapist’s notes or the doctor’s chart.  With each complaint of weekly anxiety, the doctor added increasing doses of Seroquel until reaching a dose of 800 mg. a day. 

The result is that the patient walks around in a zombie-like state and little or no relief from her anxiety.  She came to me for a second opinion, and I proceeded to aggregate vital data missed by both her therapist and her psychopharmacologist.   By the time this patient presented to me, the window for a brief outpatient adjustment was closed.

This has become a frequent experience in my clinical practice.  Patients coming to me with increased symptoms despite the use of polypharmacy.  I often inherit medication regimens with unknown prior trials, undocumented adverse reactions, and unclear reasons for augmentation or increasing medication regimes. This forces repeated, risky trial‑and‑error changes.  Memory and scattered records are unreliable; subjective recall fades and EMR notes rarely synthesize the true treatment story.

Why This Problem Persists

The brain lacks a single, simple biomarker analogous to troponin or creatinine; psychiatry therefore depends on observation and history. Care is structured around episodic visits while disease is continuous.

Electronic Medical Records (EMRs) are optimized for billing rather than insight. Clinicians now spend substantial time on documentation to appease third party payers, yet this time that does not reliably translate into better clinical understanding. Without consolidated, longitudinal context, we default to trial and error, which increases polypharmacy and patient risk. 

Expanding the Mental Checklist

Depression comes in all shapes and sizes; rarely does it present with the textbook clarity of a patient like Ms. A. In practice, every doctor operates with a mental checklist—a heuristic framework built on training and experience. Ideally, a skilled interviewer continuously updates and expands this checklist based on the nuance of the patient sitting before them. However, the constraints of episodic care often compromise this process. When a patient presents with a firm conviction about their condition or demands a quick resolution, the clinician’s natural instinct is to believe them. Operating on limited information, it is easy to accept the patient’s narrative at face value and overlook the subtle symptoms that contradict it.

Integrating AI tools has allowed me to see beyond these biases and capture the individualized signals that a hurried clinical interview might miss.  I have seen this clearly in recent composite cases: for a patient with recurrent major depression, the system flagged progressive fragmentation in sleep and activity data long before the patient felt the slide, allowing us to intervene days before a crisis. In another complex case, a patient appeared restless—typically a confusing toss-up between medication side effects (akathisia) or emerging agitation. By analyzing speech rate and typing variability, the data pointed clearly toward a mood shift, allowing me to avoid piling on more drugs to treat a side effect that didn't exist. Even with difficult-to-engage young adults, I have seen personalized digital prompts transform sporadic attendance into a consistent therapeutic alliance. In each instance, the data didn't replace the doctor; it expanded the doctor’s view, turning a rigid checklist into a fluid, precise understanding of the patient's reality.

What a Better Approach Looks Like

A practical solution must meet five requirements:

1. Deep Medication Intelligence: This goes beyond a simple list of drugs. We require a synthesized medication narrative that flags prior adverse reactions, identifies sequences suggesting treatment resistance, and highlights clear opportunities for safe deprescribing (Solmi et al., 2024).
   
   
   

2. Biological Context: This involves a rigorous consideration of laboratory data to either unearth underlying medical conditions, such as thyroid dysfunction, or to help synchronize which medications might best serve a specific patient. AI is uniquely capable of compiling and recognizing patterns within clusters of asynchronous data. Every patient possesses a unique "fingerprint" to their depression; while one may present with textbook crying spells and weight loss, another arrives with a totally different constellation of symptoms, such as OCD traits or somatic complaints like agitation and itching.

These disparities often point to biological origins—hypothyroidism, cancer, or brain tumors—rather than a simple recurrence of a psychiatric disorder. I recall a patient from years ago who requested an SSRI for what appeared to be a repeat depressive episode. However, this particular presentation included a different cluster of symptoms: depressive affect combined with urticaria and a subtle discoloration of the eyes. This triad, once detected, pointed to biliary carcinoma rather than depression. Such a diagnosis is easily overlooked, especially if jaundice is not yet prominent. AI technology can be of enormous help here, recognizing these shifts in symptom patterns and alerting the practitioner to a physiological underpinning that might otherwise be missed.

3. True Decision Support: Clinicians need synthesis rather than raw charts. We need a system that provides real-time feedback, evaluates whether a treatment is working, and suggests next steps while preserving clinician judgment. Diagnostic errors (Singh et al., 2014) already affect millions annually, so decision support must reduce rather than add to diagnostic uncertainty.
   
   

4. Zero Friction:  There should be no extra data entry. The system must integrate into existing workflows and EMRs, offer explainable outputs, and maintain clear consent and privacy safeguards.
   
   

5. A Unified Model: Finally, we must address the "split treatment" model where a patient sees a pharmacologist for medication and a psychologist for therapy. If this model continues to be encouraged by third-party payers, a closer alliance between the practitioners must be accomplished using a unified data collection approach. All issues mentioned above can be addressed using a comprehensive platform where the psychiatrist, therapist, and patient produce data that is viewed through a singular, evidence-based lens facilitated by an AI program.

A Look Ahead

Imagine a “second brain” that remembers every medication trial, tracks functional shifts between visits, and surfaces the right drug, for the right patient, at the right time. This is a shift from reactive firefighting to proactive prevention. Real‑world behavioral data are missing.  When measured and synthesized responsibly, they illuminate the invisible and improve outcomes

Works Cited

1. American Economic Review. https://www.aeaweb.org/articles?id=10.1257/aer.20240226

2. Centers for Disease Control and Prevention. (2024). FastStats: Mental Health. https://www.cdc.gov/nchs/fastats/mental-health.htm

3. Correll, C. U., et al. (2024). The challenges of antipsychotic polypharmacy. The Lancet Psychiatry. https://www.thelancet.com/journals/lanpsy/article/PIIS2215-0366(24)00367-5/fulltext

4. Krebs, P., et al. (2024). How do mental health treatment delays impact long-term mortality? 

5. Sabe, M., et al. (2025). General polypharmacy, psychotropic polypharmacy, attitudes of patients. BMC Psychiatry. https://bmcpsychiatry.biomedcentral.com/counter/pdf/10.1186/s12888-025-06746-y.pdf

6. Singh, H., et al. (2014). The frequency of diagnostic errors in outpatient care. BMJ Quality & Safety, 23(9), 727–731. https://qualitysafety.bmj.com/content/23/9/727

7. Solmi, F., et al. (2024). Discontinuation of psychotropic medication: A synthesis of evidence. Nature Mental Health. https://www.nature.com/articles/s41380-024-02445-4.pdf