Patients beware: the AI system reading your dental scans may be optimized to find problems that aren’t really there. That’s the warning from former Wall Street Journal columnist Joanna Stern, whose new book I Am Not a Robot: My Year Using AI to Do (Almost) Everything exposes how dentists across the country are weaponizing AI diagnostic tools to push unnecessary procedures, according to Futurism AI.
Stern shared her findings on The New York Times’ “Hard Fork” podcast, and her personal story is the smoking gun.
The Setup
During a routine dental visit, Stern’s office ran her scans through Pearl AI, a system that markets itself as catching “37 percent more disease and deliver 24 percent more care to patients in need.” The verdict: she needed periodontal treatment across four sessions, costing thousands, with no guarantee of insurance coverage.
Her reaction was skepticism. Her teeth felt fine. She’d never needed this work before. So she got second, third, and fourth opinions. Every other dentist disagreed with the AI’s assessment. Better home care, they said, would fix it. She skipped the treatment and was right to do so.
The Pressure Behind the Screen
What Stern uncovered next is the part that should worry every patient. Dental office employees told her their bosses were aggressively pushing AI adoption because the readings gave them a sales lever. Staff who didn’t act on the AI’s findings got grilled: “Why didn’t you drill it? Why didn’t you sell the periodontal treatment?”
That’s not diagnostic assistance. That’s a sales quota dressed up as clinical software.
Why This Moment Matters
Medical AI is having a real moment. Radiology is the poster child, where AI can flag early signs of breast cancer or other serious conditions that justify the cost and the follow-up. The stakes match the intervention.
Dentistry breaks that logic. A cavity flagged by AI isn’t life-threatening. The downside of acting on a false positive isn’t a missed diagnosis. It’s a drill, a bill, and an enamel cut you can’t undo. Tools like Pearl AI and Overjet are now fixtures in dentist offices, but the incentive structure rewards over-treatment, not patient outcomes.
This is what AI deployment looks like when the buyer (the practice) and the user (the patient) have opposite interests.
What’s Actually Changing
Three dynamics worth watching:
- AI as a billing engine. Vendors selling “more disease detected” as a marketing pitch are quietly selling “more revenue per chair.” The metrics are aligned with the practice, not the patient.
- Authority laundering. A screen showing a heat map of “problems” carries more persuasive weight than a dentist’s eyeball assessment, even when the dentist’s eyes are more accurate.
- Regulatory blind spot. Dental AI tools aren’t getting the scrutiny that radiology AI gets, despite being deployed at massive scale in independent practices with zero oversight.
Practical Takeaways
For patients:
- Always get a second opinion on AI-flagged work that costs four figures or requires multiple sessions.
- Ask whether the diagnosis came from the dentist’s exam or a software report. Then ask to see both.
- Treat “the AI says” the same way you’d treat a used-car salesman’s pitch.
For AI practitioners and healthtech builders:
- Detection sensitivity is a feature, but tuning it without accounting for downstream financial incentives turns your product into a predatory tool.
- The customer paying for the software and the human absorbing the consequences are rarely the same person. Design for that gap or own the fallout.
For businesses deploying medical AI:
- Audit how staff are being measured against AI outputs. If “acted on AI finding” is a KPI, you’ve built a system that will eventually attract a class action.
Stern’s reporting is one of the clearest case studies yet of AI’s dark pattern problem in healthcare. The tech isn’t broken. The incentives are. Full details are available at the original Futurism AI report.