Staff Reports
A new study in a major medical journal says an artificial intelligence can look at a person’s face and estimate their “biological age” and even predict short‑term survival for cancer patients. The model is called FAHR‑Face and its builders claim it learned from tens of millions of photos. That sounds clever — until you think about who might use those guesses and how they could be turned against ordinary people.
What the study actually did
Foundation model, two medical tools
Researchers at a big hospital system trained a foundation face model on more than 40 million facial photos, then made two tools from it: FAHR‑FaceAge (it estimates how “old” your face looks) and FAHR‑FaceSurvival (it predicts cancer survival risk from one photo). The survival tool was fine‑tuned on about 34,363 photos of cancer patients and tested on many more. The authors report the top risk group had roughly a three‑times higher mortality risk than the bottom group (adjusted hazard ratio ≈ 3.23), and short‑term discrimination numbers in the mid‑0.60s (AUC ≈ 0.66 at three months). Raymond H. Mak, MD, and Hugo J.W.L. Aerts, PhD, who led the work, say the model could help when other biomarkers aren’t available.
Why the results are not the miracle the press makes them
Promising, but not ready to run your care
Yes, the paper shows patterns. No, it does not mean a selfie should replace a doctor. The authors and outside experts agree this is a proof‑of‑concept, not a validated clinical decision tool. Photo lighting, pose, makeup or facial expression can change the reading. The model appears to pick up signals tied to appearance — things like weight loss, pallor, or soft‑tissue change — not the underlying biology. In plain language: correlation, not causation. Before anyone treats a score as gospel, we need prospective trials that show using the score improves real medical decisions and helps patients.
Privacy, fairness and the patent problem
Handing health data to whoever is watching
This is where it gets worrying. Face photos are everywhere — social media, driver’s license databases, security footage. A health score built from faces is cheap to compute and could be used by insurers, employers or advertisers to deny coverage or raise prices. The research group released some code and model weights for non‑commercial use, but withheld the survival weights and filed a provisional patent. So they publish enough to show the idea, but control the commercial path forward. That raises questions about consent, who profits, and whether the technology will be audited for bias. Early analyses showed uneven performance in subgroups. For example, in some tests the facial‑age link to survival was not significant for Black patients after adjustment — not acceptable if these tools touch real people’s futures.
Bottom line: hands off until rules catch up
We need guardrails, not hype
FAHR‑Face is an impressive technical exercise. But impressive does not equal harmless. Conservatives who believe in privacy, fair markets and personal liberty should be skeptical of handing a biometric health score to faceless systems or profit‑seeking firms. Regulators and lawmakers should demand transparency, strict limits on insurer and employer use, independent audits for bias, full disclosure of patents and commercial plans, and real clinical trials before any deployment. Otherwise, the next time someone asks you for a selfie, you might want to ask who will profit from your face.
