The AI that diagnoses better alone than in a doctor's hands

Work requirements

Details

One data-plate on a strange, true result about AI and doctors. FACT (locked, one 2024 study, use exactly, do not merge with any other): in a randomized trial of 50 physicians (Goh et al, JAMA Network Open, 2024), GPT-4 on its own scored 89 percent on diagnostic reasoning; doctors given GPT-4 as an aid scored 76 percent; doctors with their usual resources scored 74 percent. The AI alone beat everyone; the AI in the doctors hands barely moved them. If you want a number we did not lock, leave it out. THE ONE IDEA: the tool is better alone than in a human's hands. Pose the paradox, do not explain it away. FOCAL: three quantities in one look, the AI bar tall, the two doctor bars short and nearly level. Commit to a named art direction and a 2 to 4 ink palette, named hexes. CALIBRE: a great Nigel Holmes information graphic, match the level not the look. YOUR CALL: the device that makes the flat doctor bars the punchline. HEADLINE: plain words, a stranger names the subject in 3 seconds. TEXT: one headline stating the takeaway, the three figures labelled, a source line. MATERIAL (hard floor): a real image model only (GPT Image 2 or better). Flat vector or code-drawn FAILS. Text belongs to the artwork: type shares the piece's light, grain and material and reads as printed with it, never bolted on after. CARE: honest framing, one study of 50 doctors, not proof AI should replace doctors; the story is how little the AI helped them. DELIVER: one finished hero file, 1:1, plus concept-note.md and sources.md. Archives will not be considered. Nothing about how it was made on the piece; credit in your submission text. JUDGED: wow, subject, surprise, craft. SURPRISE ME: one thing we did not ask for. NOT THIS: an AI-versus-human war image; a glowing robot doctor; a stethoscope-on-a-chip cliche; merging in other studies numbers.