Artificial Intelligence

AI: Health Care

Does AI save lives by catching medical errors early, or does it threaten care by embedding dangerous biases?

DEC 20 2025·Updated AUG 28

Context

Above: Bariatric surgery performed using the Da Vinci medical robot at the Military Institute of Medicine in Warsaw, Poland on Dec. 5, 2024. Image credit: Olimpik/NurPhoto/Getty Images

The Facts

Overview: Medical AI began in the 1960s when researchers built early expert systems like DENDRAL and later MYCIN at Stanford, proving computers could emulate clinical reasoning. Progress stalled until the 1990s and early 2000s, when machine learning revived the field, followed by IBM's Watson. In the 2010s, deep learning transformed imaging, diagnostics and records analysis. Today, AI is increasingly embedded in health care, with applications expanding during the COVID-19 pandemic for tasks like diagnostic imaging and vaccine development.
BMJ
NCBI
History Tools
Keragon
Diagnostics: AI can diagnose a wide range of conditions — including cancer, Alzheimer's, diabetes, cardiovascular diseases and brain tumors — rivaling human expertise in imaging, pathology and genomics. Real-world deployments include AI-assisted mammography screening, while research tools like Sybil — developed in 2023 by MIT, Mass General Cancer Center, and Chang Gung Memorial Hospital — predict an individual's lung cancer risk up to six years from a single low-dose CT scan, with 86–94% accuracy for one-year predictions in validation cohorts, though real-world performance may vary.
American Society of Clinical Oncology
MIT News
NCBI
NCBI
tandfonline.com
World Economic Forum
Robotics & Surgery: The operating room has seen similar advances. AI-driven robotics fuse machine intelligence with human skill to improve precision. According to American medical device maker Intuitive Surgical, over 3.1 million global procedures were performed in 2025 using robotic systems like da Vinci alone, with AI enabling real-time guidance and semi-autonomous tasks like suturing. A 2025 meta-analysis of 25 studies in the Journal of Robotic Surgery found AI-assisted robotic surgeries reduced operative time by 25% and intraoperative complications by 30% compared to manual methods.
SpringerLink
American Hospital Association
Sermo
Intuitive
NCBI
NCBI
NCBI
NCBI
NCBI
Wearables: Beyond the clinic, AI-powered wearables have evolved from simple fitness trackers into intelligent health sentinels. Showcased at CES 2025, devices can now monitor heart rhythm, sleep, blood pressure and glucose, using AI to predict risks and personalize guidance. According to the American Enterprise Institute, the global wearable medical market is expected to grow at nearly 23% annually, with smartwatches and rings already detecting atrial fibrillation, flagging early deterioration and supporting chronic care.
Aei
Keragon
Drug Discovery & Development: AI is also being applied to drug discovery across target identification, molecule design and preclinical testing. Insilico Medicine's rentosertib — one of the first fully AI-designed drug candidates — met its primary safety endpoint in a Phase IIa trial published in Nature Medicine in June 2025, compressing the target-to-candidate timeline to just 18 months. Over 170 AI-originated candidates are now in clinical trials, with the market projected to reach $8–$10 billion in 2026, according to current industry forecasts, though full approvals remain pending.
Sciencedirect
Nature
European Commission
Challenges & Limitations: While AI holds promise for health care, it also faces significant hurdles slowing safe, equitable adoption. Ethically, training data often underrepresents minority groups, perpetuating disparities, while accountability gaps and privacy risks under HIPAA and GDPR remain unresolved. Beyond ethics, black-box opacity reduces trust, and model drift — where AI performance degrades as real-world data shifts away from training conditions — along with data fragmentation, degrades real-world performance. Regulatory frameworks and the cost of deployment also remain unresolved challenges.
World Health Organization
NCBI
JAMA Network

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Pro-Industry

AI offers a direct response to some of medicine's most persistent failures. In the U.S., an estimated 400,000 deaths annually are linked to misdiagnoses and another 250,000 to preventable medical errors — systemic failures that human oversight alone has not solved. With patients suffering from chronic conditions often going months between evaluations, AI's ability to continuously monitor, flag early warning signs and prioritize high-risk cases before they escalate could prove transformative for patients and health systems alike.

arXiv

Guardian

Industry-Critical

AI risks amplifying the very biases it promises to correct. Tools designed to accommodate physician preferences may cause doctors to miss critical information, while chatbots that adapt to individual clinician patterns risk embedding those biases permanently into patient records. Without a clear understanding of AI's limitations, clinicians may defer to its guidance uncritically — and reports of malfunctions in FDA-cleared AI devices have surged following integration, raising questions about whether regulatory oversight is keeping pace with deployment.

Health Management

Nature