Using existing data differently
In a recent case AI didn’t find new data, it used old data differently.
Set the task of re-analysing data from 376 rare disease cases it surfaced 18 new diagnoses long after experts had given up.
Researchers used an AI reasoning model to connect clinical features, genetic variants, and evolving scientific literature to produce evidence-linked hypotheses for clinicians to interrogate.
After expert review, testing, and confirmation, 18 diagnoses emerged from cases that had already passed through multiple specialists
Roughly half of rare disease patients remain undiagnosed even after genomic sequencing, not because the signal isn’t there, but because knowledge is fragmented, records don’t connect, and science moves faster than human reanalysis cycles.
The constraint isn’t data. It’s interpretive bandwidth.
WHY IT MATTERS
Organisations suffer from the similar pathology:
→ Decisions “close” too early
→ Backlogs become graveyards
→ People anchor to prior conclusions
→ Revisiting feels like failure, not intelligence
AI changes the behavioural economics of attention and lowers the cost of re-opening decisions.
It introduces:
→ Systematic second looks (norming curiosity, not closure)
→ Evidence-linked reasoning (nudging better judgement, not just faster answers)
→ Attention triage (guiding experts to where reinterpretation is most valuable)
It is an opportunity to design systems where decisions are not endpoints, but living hypotheses. Where AI nudges re-evaluation behaviours, normalising them, rewiring organisational learning loops.
WHAT TO WATCH FOR
→ Re-analysis loops embedded into workflows organically (not just post-mortems)
→ AI outputs that explain rather than rank (shifting trust dynamics)
→ Backlogs reframed as option value rather than operational debt
→ New norms removing blame: “When did we last revisit this?” replacing “Who signed this off?”
→ Organisations moving from decision finality to decision fluidity.
LIMITATIONS.
The model didn’t diagnose, experts did. A notable share of findings were rediscoveries, pointing to system integration issues as much as scientific ones.
The real lesson may be less “AI is brilliant” and more we’ve been structurally bad at storing and revisiting our own knowledge.
SOURCE
https://openai.com/index/diagnose-rare-childhood-diseases/
BESCI AI OPINION
How many organisations have the data, but don't have the systems, process, knowledge, people to do something with it. This is where AI, and its ability to quickly make sense of a lot of text based data comes in.
What could we learn if we looked differently at the past?
