ideasIdea
Publicly funded cancer AI must release open weights and model cards
If public or charity money paid to build a cancer AI model, the model itself (not just a paper about it) must be released so others can test, improve and use it.
Most publicly funded cancer AI is described in papers but the trained model is never released, so it cannot be independently validated or built on. The proposal makes release of weights, code, a model card and evaluation data (or an evaluation API where data cannot be shared) a condition of grant funding, mirroring open-access and data-sharing policies, with a governed-access route for models with genuine dual-use or privacy concerns.
Hypothesis
Open weights will lead to independent external validations for a majority of funded models within two years of release (versus almost none now) and to reuse in downstream tools, increasing the return on public AI funding.
Rationale
Open-source releases in general machine learning are reproduced, audited and extended within weeks; closed medical models are neither validated nor used beyond the originating lab.
What would test it
One funder adopts the policy for a funding cycle; count external validations and downstream uses of funded models at 24 months versus a prior cycle.
Maturity
speculative
Who has to act
philanthropy
Cost to try
Small (under $1M)
Years to first evidence
2
Bottlenecks it attacks
- AI that is built but not validated or deployed · Thousands of cancer AI models are published; a handful are in clinical use, and fewer have shown they help patients.
- Preclinical results do not reproduce · Fewer than half of landmark cancer biology findings reproduce when someone else tries.