Creating a new kind of blockchain-based data structure that is essentially combination of a neural network and a blockchain. You could call it a "blocknet" or "blockmesh" or some other mash up.
Where blockchains excel in the management of value and decentralised sharing of data, blocknets would be able to be used to efficiently delineate knowledge from raw data at scale, all the while incentivising users to provide more data in order to be rewarded with tokens for doing so.
Users supply data and information which is checked by other users that receive rewards for maintaining the integrity of the network, and once the data is confirmed as being accurate and relevant to the goals of the network then it would be added as part of a new block.
The neural network would then use the data from all previous blocks, along with the data from the newest block in order to try to do whatever the tasks is that it is designed to do.
ie. mining companies could conduct the process of exploring for new resource-rich areas (undiscovered oil fields etc) by feeding the data from surveying into a blocknet one block at a time, and when each new data package is added then the network can make one cycle in order to try to identify new resources.
Over time the network would continually become better at the task that it is designed to complete, and this growth in performance would be achieved in a completely decentralised fashion - without anything occurring other than users being incentivised to make the network smarter and increase the knowledge stored within it, and then as a result, a tool that will progressively perform better over time.
It might take a while, but if you can find a good core team - particularly some postgrads studying DLT and blockchains, you should have a crack at being able to develop a basic model to get started.
There are (at least) 3 core types of this model by the way - a chain that takes data from the end to determine all previous block values, a chain that takes all earlier data and uses that to determine the most recent block values, and bi-directional chains that determine both earlier and later block data with all other related data.
