,
20 tweets,
4 min read
Read on Twitter
1. Here is a pattern I've been noodling for years, and I think it is salient to the #notStupidEnterpriseBlockchain challenge (day 13). Let me see if I can lay this out. Tugging on the edge of something here. Patient #Crypto friends, read on:
2. Party 1 wants to share data from its System A with remote Party 2 and a System B, which it controls locally. Recent example: Facebook sharing data with Cambridge Analytica. (Simplified...could be n systems.)
3. System B needs to perform a series of operations on a corpus U that joins data from System A and B in a way that requires all data to reside locally in System B. Parties A and B agree that only these operations are allowed. Party A permits no other use of System A data.
4. Party A can't be sure that Party B isn't a bad actor willing to surreptitiously copy System A data while performing the agreed-upon operations. Normally the risk of data appropriation would be too high, but...
5. Both Systems A and B implement a common protocol and agree to share a trie of hashes that represent the world state of their respective systems, so that protocol compliance can be confirmed by each System observing specific hash markers in the remote System's trie.
6. This should be done in such a way that compliance is the only thing a remote system can learn about a local System, so that the internal functions of each Party can remain private.
7. System A and B send messages to each other at appropriate events to ensure that neither has changed its state and tampered with the implementation of the protocol. Let's call this "Entanglement" just for fun.
8. In this embodiment, the protocol will perform several operations: a) check that System B's state is unchanged from previous verification, and if fail, send error and delete all data provided by System A.
9. b) check that System B is running on a High-Security-Module (HSM) and is compliant with other security measures, per original agreement with Party 1. If fail, delete System A from System B and send error.
10. c) If a, b pass, receive System A's Part of a shared encryption scheme required to operate on System A's data locally in System B. The scheme is implemented in such a way that the received Part can't be used in subsequent operations without obtaining a new part from System A.
11. d) Protocol 'watchdogs' System B during the execution of the permitted operations (say for example, machine learning model generation on U) and ensures that System A data is not copied during process or used by non-sanctioned local operations.
12. e) System A and B re-encrypt or verifiably delete System A data at the conclusion of the permitted operations.
13. The notion here is a regime that says Entangled Systems trust a shared protocol to provide a kind of "restraining bolt" (nod to Star Wars) whose presence is verified before joint events can occur (initiation, copy, decryption, message passing).
14. More extensibly, this kind of verified Entanglement might be used perhaps for things like system orchestration where different parties control different parts of a system. Future electrical grid SCADA systems come to mind.
15. final. I'm pretty sure I'm missing lots of devils in the details (so please, would be grateful for thoughtful replies along this thread to help me find the gaps).
Perhaps it is impossible to implement a subset of shared software objects such that, without knowing all elements of the total remote system, can verifiably prevent the unobserved objects from performing unsanctioned operations.
For avoidance of doubt, Party 2 controls System B, not Party 1.
One other thing: it's essential that the notion of "excommunication" be emphasized. Meaning, that System A will excommunicate System B if the branches of the trie that all entangled Systems expect to correspond to the restraining protocol don't pass hash function.
And as my old friend Gari Singh (one of the world's great distributed system architects) says, the real trick would be not so much the higher level software layer state attestations but down into machine level and system level.
Maybe in @MIT's #neilgershenfeld's world where computing and physics are the same (qc?), and there is no abstraction between the atoms, the qbits, the info, the logic, then entangled systems?
