Physical therapists are discovering that AI works best not as an autonomous decision-maker, but as a thinking partner that amplifies human expertise and fills specific gaps in patient care. Rather than debating whether artificial intelligence will enter clinical practice, therapists are now focused on how to implement it responsibly, ensuring technology supports rather than undermines the relational and interpretive nature of treatment.

Why Patient Adherence Is the Real Problem AI Can Actually Solve?

One of the most persistent challenges in physical therapy is patient adherence to home exercise programs. Patients struggle for many reasons: uncertainty about whether they're performing movements correctly, difficulty fitting routines into daily schedules, or simply not understanding how the program addresses their condition. This is where AI-powered computer vision technology is making a tangible difference.

Rather than simply correcting form, computer vision systems can teach patients new exercises, provide real-time feedback during movement, and gather valuable data about how patients are progressing. When a patient struggles with an exercise or finds it too easy, the system can recommend modifications or progressions for the physical therapist to review, eliminating the need to wait until the next appointment. The therapist remains the primary decision-maker, but now has richer information to guide adjustments to the treatment plan.

AI chatbots and conversational agents are extending physical therapy care beyond clinic walls in similar ways. These tools can help patients build habits around exercise routines, reiterate educational components like pain neuroscience education, and gather information that therapists can quickly review to determine whether treatment modifications are necessary. The goal is engagement and support, not replacement of the provider.

How Can AI Reduce Bias in Clinical Assessment?

Clinical decision-making in physical therapy is not purely mechanical or logistical; it's interpretive, contextual, and relational. This is where AI's strength lies: reducing human variability and bias. Research has long documented that physical therapists can vary widely in their clinical judgments, and studies show measurable inconsistency in assessments like joint palpation, movement analysis, and range of motion measurement.

A 2016 study published in the Journal of Orthopaedic and Sports Physical Therapy demonstrated anchoring bias, where therapists measuring wrist passive range of motion produced different results depending on what historical information they received about the patient beforehand. AI algorithms can help reduce this variability by providing stable baselines and highlighting discrepancies that might otherwise be missed.

Steps to Implementing AI Responsibly in Clinical Settings

• Involve Clinicians in Development: Any AI tool created to support clinical work must include expert clinical input from brainstorming through implementation, daily use, and regular iteration. If practicing clinicians perceive AI as opaque, intrusive, or misaligned with their goals, they will disengage or override it reflexively.
• Maintain Clinician Authority: Achieve synergy between AI and human clinicians so that the best capabilities of each complement one another. Clinicians should neither dismiss nor defer to AI, but engage with it as a thinking partner that strengthens their intuition with reliable, reproducible input.
• Verify Source Quality: While AI can summarize bodies of evidence and surface important research quickly, many tools cannot yet consistently assess the quality of sources or study design. The responsibility for determining reliability still rests with clinicians, who must critically evaluate what the AI presents.

What Should Therapists Do When Patients Bring AI-Generated Health Information?

About one in six adults, and roughly a quarter of adults under 30, used chatbots to find health information at least once a month in the past year, according to a survey from the health policy research group KFF cited by The New York Times. The problem is that the online advice landscape is noisy, with countless voices offering recommendations that can worsen symptoms. A 2025 study examining widely used large language models (LLMs), which are AI systems trained on vast amounts of text data, found high variability and inconsistent accuracy compared to published clinical practice guidelines for lumbosacral radicular pain.

"Because it's no secret that patients are already using LLMs like ChatGPT to learn about their conditions, I invite them to bring what they find to our sessions. This presents a valuable opportunity to explain how I combined their story, exam findings, imaging when relevant, and their goals with the best available evidence to build the plan," explained a physical therapy clinician with 13 years of experience.

Physical Therapy Clinician, 13 years clinical experience

This approach allows therapists to coach patients toward safer self-education and correct misinformation in real time. These teachable moments build health literacy and keep care grounded in clinical guidelines and context. AI should be seen as a bridge, not a barrier.

Where Is International Research on AI-Driven Health Technologies Happening?

Beyond clinical practice, academic institutions are becoming hubs for developing the next generation of AI health tools. Khalifa University's Digital Medicine Lab in the United Arab Emirates is attracting international researchers focused on wearable sensing, biomedical signal processing, and AI-driven health technologies.

Two self-funded PhD students from Spain are currently conducting research at the lab, including work on wearable photoplethysmography (PPG) sensing, which measures blood flow through the skin using light. One student is developing AI-powered wearable tools for sleep apnea screening using PPG signals, aiming to provide a low-cost and scalable alternative to laboratory polysomnography, the gold-standard sleep study test. The work integrates signal quality assessment, pulse rate variability analysis, and deep learning approaches to discover cardiovascular biomarkers related to apnea.

The lab's director, Dr. Mohamed Elgendi, has been recognized among the world's top 2% most-cited scientists from 2021 to 2025 in the annual Stanford University-Elsevier ranking of global scientific impact. Recent outputs from the lab have appeared in leading Nature Portfolio journals, including npj Digital Medicine, Communications Medicine, and Communications Engineering.

"The presence of talented PhD researchers from leading European institutions at our Digital Medicine Lab reflects the strong international appeal of Khalifa University as a hub for cutting-edge biomedical innovation. Our laboratories provide an interdisciplinary environment where expertise in wearable sensing, artificial intelligence, and biomedical engineering converges to address real-world healthcare challenges," stated Dr. Mohamed Elgendi.

Dr. Mohamed Elgendi, Assistant Professor, Biomedical Engineering and Biotechnology, Khalifa University

How Are Healthcare Systems Building Practical AI Solutions?

At Stony Brook Medicine, a clinical data hackathon brought together 21 clinicians, data scientists, and researchers who worked with Google engineers to tackle real-world healthcare challenges using cloud and AI tools. The full-day event, held in April 2026, demonstrated how interdisciplinary collaboration can rapidly prototype solutions with clinical impact.

Teams worked on projects ranging from ICU hypotension forecasting and arterial waveform analysis to operational systems like "Rescue Radar" designed to improve hospital response times. Participants handled the entire end-to-end lifecycle of AI development, from initial literature and data discovery through data engineering, model training, app development, and final presentations.

The hackathon provided practical experience working with gold-standard clinical datasets, including MIMIC-IV, TCIA, and the Imaging Data Commons. Teams built AI-powered clinical note summarizers and multimodal radiology analysis pipelines, collaborating alongside Google engineers to prototype solutions. The event exceeded expectations and highlighted the potential of combining clinical insight with advanced technology to drive meaningful advances in healthcare and research.

The bottom line is clear: AI's promise in healthcare lies not in autonomy of either the clinician or the machine, but in balanced collaboration that bolsters the strengths of each. When implemented thoughtfully, with clinician input and oversight, AI can reallocate clinicians' cognitive energy toward the type of thinking that is uniquely human: interpreting data with a lens that includes clinical history, patient beliefs, and experience gained through years of treating patients.