Whether you like it or not, artificial intelligence (AI) is a fixture in health care. AI programs are now reviewing imaging studies, generating documentation, summarizing medical literature, assisting with diagnosis, generating risk assessments, and planning treatments. There is no doubt that AI can handle complex data processing better than us.

As health care grows increasingly complex, these capabilities will further improve efficiency and may reduce certain types of human error. Yet despite the rapid expansion of AI in medicine, important questions remain: what should human oversight of AI in health care look like and which physician skills must be preserved to ensure that AI strengthens rather than diminishes patient care?

In other words, if AI is here to stay and ever improving, how can we ensure that we also maintain a voice when it comes to diagnosis and management decisions?

Evidence suggests that humans and AI tend to have different and almost complementary strengths and weaknesses.1 AI excels at managing complex data and doing it fast. It performs better than humans on tasks of convergent thinking, where a single-best answer can be found.2 On diagnostic tasks, AI systems can match or exceed the performance of trainees.3

In contrast, AI often struggles with ambiguity, rare presentations, and situations that fall outside the patterns represented in its training data.1 While expert clinicians still make errors of imperception or misperception, experts tend to be better than AI in divergent thinking tasks, including creative idea generation and adaptability.2,3

This distinction is highly relevant in spine medicine, where diagnosis and treatment occur within a complex biopsychosocial context. Patients bring unique functional limitations, expectations, risk tolerance, resilience, and goals of care that are often as important as imaging findings, diagnostic labels, or risk calculators in treatment decisions. Contextual understanding and motivational interviewing remain some of the expert clinicians’ greatest strengths.

History and more recent studies provide cautionary lessons about overreliance on AI in medicine.1 Early computer-aided mammography systems were intended to improve cancer detection but produced the opposite effect.4 Radiologists became overly focused on computer-generated findings, a phenomenon driven by attention capture and confirmation bias, and performance worsened. Modern AI systems carry similar risks.1

If clinicians consult AI too early in the diagnostic process, they may anchor to its conclusions and prematurely narrow their thinking. Human oversight, therefore, should mean more than simply reviewing an AI-generated answer. It should involve independent clinical reasoning first, followed by using AI as a cognitive cross-check.1,5 In this role, AI can potentially augment expertise rather than diminish it.

Perhaps the greater long-term concern is not diagnostic error but the gradual erosion of expertise itself. Emerging evidence suggests that excessive reliance on AI may create a form of cognitive debt, in which individuals increasingly offload thinking tasks and gradually lose the skills those tasks once developed.6

A recent randomized study at MIT showed reduced cognitive engagement and poorer recall among individuals who initiated an essay-writing task with the use of AI (“AI-first”) compared to those using a “brain-first” approach.7 Frighteningly, when the groups crossed-over, the original “AI-first” group still showed a cognitive debt when later asked to take on a second essay-writing task without the use of AI. In other words, using AI too early in the creative process may actually reduce our ability to think creatively on a future task.6

My primary point is that the physician skills most likely to remain valuable in the AI era are the same skills that have always distinguished master clinicians from average ones: adaptive reasoning, contextual understanding, ethical judgment, empathy, and creative problem-solving. To remain relevant with AI integration in health care, we need to place new emphasis on these low-tech human skills that allow us to apply knowledge wisely, creatively, and compassionately in circumstances where there is no single correct answer. I hope trainees can still learn these skills despite AI.

On that note, I want to encourage everyone to attend this year’s NASS Annual Meeting in San Antonio. I've written about the scientific and community building benefits of in-person attendance in the past.8 The benefits of building and maintaining a spine community are made even clearer in this issue’s President’s Message from NASS Vice President Christopher Kauffman, “NASS: The Home of Spine Professionals” and the article on “Burnout, Purpose, and Longevity in Spine Care” from the NASS Governance Committee.

The July/August issue of SpineLine is also packed with high-value information about the conference, along with its usual technology updates, coding tips, and 2 high-yield journal article reviews, one from NASSJ and one from TSJ.

I’d like to also highlight the excellent Invited Review article on gabapentin and dementia by John Hesling and Austin Boos, as this kind of writing exemplifies the broad, multidisciplinary nature of NASS and the value of spine specialists digesting data coming from nonspine journals and generating reasonable and practical recommendations for the spine community.