Prizes are a really good way of funding science but are not used enough. All sorts of prizes from longitude to private space have speed up technological development in their area.
Kaggle is an interesting example of current innovation incentivised using prizes. Kaggle defines a metric over a dataset and people try an build models that predict that well. This is then tested over another dataset to prove your predictions are the best. For not very much money a large number of very skilled people work on a defined problem for fun, kudos and the possibility of profit.
Typical Kaggle competition lasts 3 months, offers $25,000-100,000 in prize fund and attracts around 1000 specialists
At least top 10% of those specialists, ~100 persons are of prime quality, many others 'just' good.
Could a similar thing be done for discovering materials with properties we want? There are all sorts of problems that could be solved with materials with new properties. There are commercial incentives to develop many useful materials already. A more efficient solar panel or energy dense battery could have such commercial value lots of people are working on making them already. But some properties of materials are known to be useful but not commercialised yet to have huge competition or budgets trying to create them.
If we agree prizes are a useful incentive. And that we need new materials with useful properties how might a kaggle for such prizes work?
Kaggle has test datasets unknown to competitors held out to prove later which prediction model is best. For a materials version proving specific qualities in the lab would, initially, be too expensive. Relying on the scientific peer review process, of quality journals, would probably be enough initially. Occasionally issues in published materials research comes to light. But that happens in kaggle competitions too. And with prizes fairly low the incentives for shenanigans are not huge. Use a paper being published by a high quality journal as proof that a material has the defined property.
How long would the competitions be open for? The millenium prize for solving known big hard maths problems are open ended. Kaggle competitions are shorter a few month time periods. The time needed for journal papers to be approved and the difficulty of making materials means a year is probably more practical. Having a year competition timeline makes it more open to tinkerers than big grand ambitious challenges. I think 'A prize to someone who makes a wire room temperature superconductor at atmospheric pressure' is too ambitious. 'A prize to someone who publishes the warmest superconductor in a substance that can be turned into a wire by December 2022' is more practical. Competitions for improvements of a defined characteristic over a period of a year or two seem to be the best prize incentives to me.
How much would the prizes be for? Using journals keeps the cost of running the competitions low. But also means huge amounts of money cannot be involved. Also unlike kaggle a company putting up a dataset is not paying the prizes. The prizes would just be donated by people who can see that creating a product with these properties would be useful for improving our lives. As such a Patreon like model where donations are collected would probably be best. Stripe lets foundations setup as non profits set up these sorts of donations. Prizes would probably be in the 100K to -> 1 million range but initially to prove the concept would be much lower.
What would be a good initial material quality to test this Patreoned Kaggle for materials idea on?
No comments:
Post a Comment