Not a computation data type as such, but ability to use semantically stored information that can be queried in a decentralised fashion and queried using map-reduce type algorithms across the network. This will be added to the ability to handle small computations per node using zk-snark type tech to ensure ‘halting (or more accurately lack of)’ is not an issue but using group consensus as opposed to snark validation of computation. Reason we will use snark is only non halting issue, the group consensus then takes care of the environment issues that zk-snarks are not good at (OK you run the code but can I be sure of your environment, was that really a real blockchain or similar you wrote to etc.).
This mix is potentially very powerful and allows the post Google type intelligence, so not searching but asking questions and getting answers via all of humanities information. a real goal of mine.
Sorry for my very fast messy answers