Understanding what AI is—and what it isn’t—helps us better grasp the complex relationship between technology and therapy. Can AI support mental health? Yes. Can it replace the human connection found in therapy? That’s where things get more nuanced.
Artificial Intelligence (AI) refers to computer systems designed to perform tasks that typically require human intelligence—like learning, problem-solving, and decision-making. While AI has been around in theory since the 1950s, its real-world applications have exploded in recent years, especially in healthcare and mental health.
AI in mental health is not just about chatbots or symptom checkers. It includes:
- Predictive analytics to identify at-risk individuals
- Natural language processing to analyze therapy transcripts
- Personalized interventions based on behavioral data
- Virtual companions for support between sessions
These tools can augment care, increase accessibility, and reduce barriers—especially for those in underserved areas or on long waitlists.
But we need to note – augmentation is not replacement.
What Makes Human Therapy Unique?
When you meet with a therapist—whether in person or via telehealth—there are subtle, deeply human elements that shape the experience:
Body Language
Nonverbal Cues
Facial Expressions
Tone and Voice Changes
These cues help therapists understand what a client is feeling, even when words fall short. They allow for real-time adjustments, deeper empathy, and a sense of being truly seen and heard.
Human vs. Machine: The Therapeutic Relationship
When considering the difference in human therapy vs. AI therapy, it’s important to look at two specific terms:
Empathy & Intuition
Empathy is the ability to understand and share the feelings of another person. This allows a therapist to consider:
- Emotional Resonance: Feeling what someone else is feeling.
- Cognitive Understanding: Recognizing and making sense of another person’s emotional state.
- Compassionate response: Responding with care and concern.
In short terms: Empathy allows a clinician to deeply connect with a client’s experience which helps the client feel heard, seen, and validated.
Intuition is the ability to understand or know something immediately, without the need for conscious reasoning.
- Picking up subtle cues [body language, tone, energy changes]
- Tips the therapist when to challenge or support a client
In short terms: That “gut feeling” or instinctive insight.
Considering that both empathy and intuition take into account that clients current presentation which cannot be fully assessed through a chat, the question poses then on how can AI properly recognize and provide appropriate care for the client.
The question would be: How can AI replicate how to make someone feel heard or seen or would it be able to truly understand someone’s suffering and provide an individualized and empathic response. The answer is likely no; a therapists ability to demonstrate empathy and intuition comes from years of training, lived experience and moment-to-moment presence.
Food for thought: “If therapy is about being deeply seen and understood, what role should AI play in that process—and where should we draw the line?”
Ethical Considerations
Let’s chat about how the use of AI relates to ethical thought processes.
Our core question is: Is AI held to the same standards as human clinicians?
Ethics [in the mental health world] refers to the principles and guidelines that guide mental health professionals in making responsible, respectful, and just decisions in their work with clients. Ultimately following ethical guidelines and standards ensures that a client’s care is delivered in a way that protects the clients rights, promotes well-being and upholds professional integrity.
Data Privacy and Confidentiality: How is your information being stored, what is truly confidential and is AI being held to the same HIPAA/privacy standards and laws that a therapist would be?
Many of us are from the era of snapchat and one thing snapchat has taught us – nothing on the internet ever truly goes away.
The AI Algorithm: If you are familiar with AI and technology, you are aware that it works off of an algorithm like script. A significant concern with this is that it would essentially think like the model of “one size fits all”. For those of us in the mental health field, we are very aware that therapy is not a “one size fits all” and truly requires the individualized aspect. Lacking the individualization increases chances of bias which leads to:
- Misdiagnosis and/or inappropriate treatment recommendations
- Inappropriate or unfair disparities between marginalized or underrepresented groups
Let’s think about: Can AI understand the full context of a person’s experience?
A good example of this is looking at someone who might be presenting with hallucinations [seeing, hearing, feeling, smelling and/or tasting something that isn’t actually there] or delusions [firmly held false belief that is not based in reality and is resistant to logic or contrary evidence]. Without understanding any individual aspects of this person, we might begin to go down the psychosis, schizophrenia or schizoaffective route. What we could be missing is understanding:
- Spirituality: Some cultures, religions, etc. have a belief in connecting with spirits, ancestors, shadows, etc. Someone who comes in discussing this may appear to describe a hallucination or delusion, but in reality it is how they feel connected to their culture or spirituality.
- Trauma: Trauma can cause symptoms that may present like hallucinations or delusions, but we try to analyze if they are rooted in truth.
- Someone who has experienced violence against them may struggle with “feeling like they hear things that no one else does”. Could this be a hallucination? Yes. It could also be hypervigilance as a result of what they have experienced.
- Substance Use: Substances can cause these symptoms both during active use and during withdrawal. It can also be important to note that some substances can have what we call “Post Acute Withdrawal Symptoms” or PAWs that can occur up to a year after the individual stopping use. PAWs can often include symptoms or experiences that could be mistaken for hallucinations or delusions.
Taking those 3 examples, and understanding that they are more, emphasizes the piece about AI not understanding the individualized aspect.
Decision Making:
How does AI arrive at its recommendations?
Much of mental health decision making is rooted in clinical understanding through assessments, risk ratings, etc. Again, these cannot always be black and white as there is the individualized piece and ensuring that recommendations are tailored to the actual client. So how do we truly know that AI is providing an individualized recommendation?
Accountability: In all ethical systems and frameworks we have to identify the responsible party. Essentially, who is responsible should a mistake be made. When we look at a therapist who look at if they have a licensed supervisor [if the therapist is pre-licensure], the agency or clinic they are at and potentially their malpractice insurance, if applicable. We also know that therapists are subject to oversight by different regulatory bodies and in clinical settings. If you are licensed you are responsible for maintaining your continuing education credits, ensuring proper up to date training, documentation expectations and ethical standards of care per your licensure board. Additionally, you are being held to standards indicated by your place of work and any licensing or oversight [245G, CARF, JCO, DHS, Federal Statutes, HIPAA, etc.] that regulate pieces of your practice.
Who’s watching the Machines?
Who or what regulates AI? Who provides oversight, ensures it’s up to date when diagnoses change [think DSM-IV to DSM-V changes], changes in best practice recommendations or even statute changes. Lacking proper oversight and accountability can have some of the most significant impact. It creates the question of, Who is responsible for when AI gets it wrong?
Think of this: In the change from DSM-IV to DSM-V one of the notable changes was the removal of Asperger’s Disorder which had been under the umbrella of Pervasive Developmental Disorders. In DSM-V, it was eliminated as a separate diagnosis and the broader category of Autism Spectrum Disorder (ASD) was created. Another example is Hypochondriasis [define] was removed with DSM-V and replaced by Illness Anxiety Disorder and Somatic Symptom Disorder. Now maybe this seems like a lot of big words and large definitions, but for those of us within the field these are significant changes. Diagnosing is how you obtain insurance approval for services AND it helps us understand what drives/impacts/defines our mental health. Being given an outdate diagnosis or something that isn’t considered best practice/recommended can not only provide inaccurate information but it can have detrimental effects to the client.
Connecting Ethical Considerations to Regulatory and Professional Guidelines
Let’s consider 2 main points, FDA regulation of mental health apps as well as licensure implications.
Not all mental health AI resources are regulated by the FDA [Food and Drug Administration]. The FDA requires that FDA oversight is required for:
- Apps that deliver therapeutic interventions
- Apps that claim to diagnose mental health conditions
- Apps that function as clinical decision support tools for providers
It does not require FDA oversight for:
- Apps that provide general wellness tips
- Mood trackers or journaling tools
- Educational content about mental health
So considering the above, tools that are providing actual clinical or therapeutic services do require FDA oversight.
Some examples of tools that are regulated by the FDA include:
- Rejoyn
- Focuses on treating major depressive disorder [MDD] in adults.
- Prescribed by a clinician and is designed for individuals who don’t fully respond to antidepressants.
- EndeavorRx
- Treats ADHD in children 8-12.
- Available by prescription to be used alongside other treatments.
- Somryst
- Treats chronic insomnia in adults
- Prescription-only
- Freespira
- Treats panic disorder and PTSD
- Clinician-supervised; typically over a 4 week period.
These are an example of prescription digital therapeutics (PDTS) – software based treatments that require clinical oversight and are regulated as medical devices.
When looking at licensure implications, the major theme is ensuring that AI does not replace the therapeutic relationship but enhance it and ensuring that it is documented and disclosed to clients. Ultimately it is not recommended by licensure boards or regulatory guidelines to rely solely on AI for client care. The responsibility still falls on the clinician to both ensure it is accurate [if helping write notes] and that the client is aware [inclusion in informed consent].
Current Applications: How is AI being used currently for Mental Health?
When we think of AI and mental health, the more common known ones will be the use of chatbots, services like ChatGPT or Copilot, Mood tracking, AI-assisted diagnostics and personalized treatment recommendations.
So how can AI help?
Rather than replacing therapists or relying solely on AI, it can:
- Support clinicians with insights and documentation
- Offer psychoeducation and coping tools
- Bridge gaps between sessions
- Scale access to underserved populations
Final Thoughts: Walking the Line Between Innovation and Integrity
As artificial intelligence continues to evolve, its role in mental health care will undoubtedly expand. From increasing access to care to supporting clinicians with data-driven insights, AI holds real promise. But with that promise comes responsibility.
Mental health is not a one-size-fits-all field. It’s deeply personal, shaped by culture, trauma, identity, and lived experience. While AI can simulate support, it cannot replicate the human connection, empathy, and intuition that define therapeutic relationships.
Ethical considerations—like data privacy, algorithmic bias, decision-making transparency, and accountability—aren’t just technical concerns. They’re human concerns. They remind us that behind every dataset is a person, and behind every recommendation is a life that could be impacted.
As clinicians, clients, and creators, we must ask:
Are we using AI to enhance care—or to replace what makes care truly healing?
The future of mental health may include machines, but the heart of healing will always be human.
Disclaimer: For the purpose of this article, I was not familiar or aware of all the applications, in fact, I used AI [copilot] to assist in researching different ways to incorporate AI with mental health therapy.
Resources:
- https://www.mdpi.com/2076-0760/13/7/381
- https://mental.jmir.org/2024/1/e58493
- https://www.unitedwecare.com/wp-content/uploads/2024/05/The-Regulatory-Landscape.pdf
- https://ejnpn.springeropen.com/articles/10.1186/s41983-023-00735-2
- https://www.apa.org/practice/artificial-intelligence-mental-health-care
- https://www.scu.edu/ethics-spotlight/generative-ai-ethics/the-ethics-of-ai-applications-for-mental-health-care/
- https://www.psychiatry.org/File%20Library/Psychiatrists/Practice/DSM/APA_DSM_Changes_from_DSM-IV-TR_-to_DSM-5.pdf
- https://www.yourceus.com/pages/dsm5577-section-vii-dagnostic-categories-removed-replaced-in-the-dsm-5
- https://www.fda.gov/medical-devices/digital-health-center-excellence/device-software-functions-including-mobile-medical-applications
- https://www.verywellmind.com/fda-approval-and-mental-health-apps-5193123
- https://www.psychiatry.org/News-room/APA-Blogs/Exploring-Digital-Therapeutics
- https://www.counseling.org/resources/research-reports/artificial-intelligence-counseling/recommendations-for-practicing-counselors
- https://www.psychologytoday.com/us/blog/navigating-the-serpentine-path/202309/ethical-principles-in-mental-health-care
