When I finished my graduate training, "artificial intelligence" wasn't a curriculum topic. It wasn't even a sidebar. We were focused on neuropsychological assessments, ethical codes, and the meticulous, painstaking work of understanding human behavior, concussion, and trauma. I thought I knew what tools I would need to practice rehabilitation psychology effectively. I was wrong.
Today, the clinical frontier is shifting under our feet. AI is being woven into the fabric of our work, from administrative note-taking to the adaptive platforms we use for cognitive rehabilitation. This transition brings a lot of excitement—and an equal amount of apprehension. It’s a strange feeling, being a professional who suddenly feels like an apprentice, trying to figure out how these tools fit into a practice where the core remains human-to-human, person-centered care. If you're feeling that same uncertainty about how to balance the technological hype with the ethical responsibilities we hold, believe me, you aren't alone. It’s not just a technological question; it’s an ethical one. We need to navigate this, carefully, while keeping our foundational discipline front and center.
Not Just Hype: Where AI Can Serve Us
To dismiss AI as a passing fad is to ignore its immediate utility. In the realm of rehabilitation, AI offers genuine advantages, provided we use the tools with a critical eye.
Think about cognitive rehabilitation. My patients, often dealing with the aftermath of traumatic brain injury (TBI) or long COVID, require consistent, repetitive practice to consolidate gains. Traditionally, this meant intensive in-person sessions followed by homework instructions that were often forgotten or abandoned. Adaptive algorithms now allow us, as clinicians, to assign computer-based training programs that track performance in real-time. These tools adjust the difficulty based on where the patient is at, giving us data points we could never collect manually.
Beyond therapy, there’s the sheer efficiency aspect—the "drudge work" that eats into our clinical time. Natural language processing models are already assisting professionals in streamlining documentation, cleaning data for research, and even helping to summarize vast amounts of literature (Alnattah et al., 2025). When it’s used to handle administrative tasks, it can free us up to spend more time with the patient, where the actual healing happens.
Additionally, AI has matured into a vital advocate for accessibility. It is frequently employed to build disability-friendly technology, such as advanced voice-recognition software and predictive text tools that make the digital world more open to people with various impairments. These are tangible, positive outcomes, and they are why we cannot simply opt out of this technological shift. The question is not if we use it, but how we incorporate it into our existing practices without losing our ethical footing.
The Hidden Pitfalls: Bias and Cognitive Offloading
However, I find the uncritical embrace of AI deeply concerning, especially when I see it being used as a substitute for thought.
The most acute risk is the reinforcement of baked-in socio-cultural biases. These algorithms are not neutral; they are reflections of the datasets they were trained on, which are often riddled with historical, medical, and socio-economic biases. When we use tools that haven't been scrutinized, we risk exacerbating disparities for underrepresented groups (Hansen & Kerkhoff, 2025). The NYC Bar Association (2025) has noted, for example, that image-generating models often portray disability in narrow, dated, and stereotypical ways—depicting it only as a wheelchair user, or framing disabled individuals as inherently lonely or isolated. If your diagnostic tool or treatment optimizer is leaning into that kind of bias, you are already hurting your patient before you’ve even started the session.
Then there is the issue of privacy. If you’re utilizing a readily available, public-facing AI tool to transcribe clinical notes or generate treatment plans, you are fundamentally violating the principle of confidentiality. Once you input that data, it’s not yours anymore. It’s being added to the model’s training pool. That’s a massive breach of trust, and arguably, a massive ethical failure.
Finally, we need to address the "cognitive offloading" problem—what some are calling the "critical thinking epidemic." I’ve seen this personally. My patients dealing with TBI, who I want to see engaging in the hard, independent work of problem-solving, are sometimes using chatbots to do it for them. When we offload the cognitive work of writing notes, creating test batteries, or planning treatments to an LLM, we are, by definition, reducing our own independent analytical development (Jose et al., 2025; Tian & Zhang, 2025). If we stop doing the hard work of synthesizing information ourselves, our ability to do so will inevitably erode over time. Dependence isn’t just a risk for our patients; it’s a risk for us, too.
For deeper context on this pattern of automation-induced skill atrophy, see The AI Dependency Paradox: How Chatbot Reliance Weakens Independent News Verification, which explores similar cognitive offloading effects in media consumption.
Maintaining a Human-Centered Clinical Stance
So, how do we proceed? I don’t believe in turning back the clock—that wouldn't be useful. Instead, I believe in practicing with intentional, persistent skepticism.
"Knowing better" must become our operational default. This goes far beyond just understanding how your specific software works. As rehabilitation psychologists, we need to be literate across fields. We need to be checking in with the medical, legal, and educational communities to understand the broader implications of these algorithms (Hansen & Kerkhoff, 2025).
We also need to implement, firmly, the "human-in-the-loop" rule. The clinician must always be the final judge. You use the AI to help you build a framework? Fine. But you don't use it to make the final clinical decision. You must be the one to proofread, verify, challenge, and ultimately endorse the accuracy of the output. If you can't justify the conclusion because you don't understand how the AI arrived there, the conclusion isn't yours to use.
When I’m working with my patients on their cognitive rehabilitation goals, we talk openly about the tools they want to use. We don't just ban them. That's a losing battle. Instead, we examine them. We ask: "What did this tool do well?" and "Where did it lead you astray?" We build our treatment plans based on direct, active collaboration. The strategy here is not to replace the clinician but to leverage our unique human judgment to turn these black-box tools into something more transparent and deliberately focused on the patient's actual, real-world needs. We have to be the ones, ultimately, who decide whether a tool serves the patient or just serves the convenience of the provider. And that, in my opinion, is a choice we have to make in every single clinical interaction.
The Road Ahead
AI is part of the clinical landscape, and it doesn't look like it’s checking out anytime soon. But it is not, and it cannot be, a replacement for the essential clinical work that defines our profession. The heart of therapy, of diagnostic assessment, and of empathetic rehabilitation will always reside in the human interaction—in the moments where we, as professionals, can account for the ambiguity, the complex history, and the unique humanity of the person sitting across from us.
If we approach this new frontier with humility, with a commitment to continuous learning, and, most importantly, with an unwavering commitment to our existing ethical frameworks, we can harness the power of these tools without surrendering our integrity. Tradition, as Mitchell (1993) wisely noted, is best served by a framework that balances continuity with change. We are in the midst of change, but it is up to us to ensure that, as we move forward, we don’t lose the essential elements that make our work therapeutic in the first place. Stay curious, stay skeptical, and keep the patient at the center. Always.
For related perspectives on AI-driven emotional support tools, see The Rise of Para-Therapy: Redefining Emotional Well-Being in the Age of AI.