The cryptography behind bitcoin solved a paradoxical problem: a currency with no regulator, that nonetheless can't be counterfeited. Now a similar mix of math and code promises to pull off another seemingly magical feat by allowing anyone to share their data with the cloud and nonetheless keep it entirely private.
On Tuesday, a pair of bitcoin entrepreneurs and the MIT Media Lab revealed a prototype for a system called Enigma, designed to achieve a decades-old goal in data security known as "homomorphic" encryption: A way to encrypt data such that it can be shared with a third party and used in computations without it ever being decrypted. That mathematical trick—which would allow untrusted computers to accurately run computations on sensitive data without putting the data at risk of hacker breaches or surveillance—has only become more urgent in an age when millions of users constantly share their secrets with cloud services ranging from Amazon and Dropbox to Google and Facebook. Now, with bitcoin's tricks in their arsenal, Enigma's creators say they can now pull off homomorphically encrypted computations more efficiently than ever.
http://www.wired.com/2015/06/mits-bitcoin-inspired-enigma-lets-computers-mine-encrypted-data/
[Paper]: http://enigma.media.mit.edu/enigma_full.pdf
(Score: 2) by Justin Case on Wednesday July 01 2015, @07:43PM
OK, thank you, that at least I can follow.
Is it any good for anything beyond simple arithmetic... which you wouldn't "outsource" when you can do it yourself for way less than the cost of the crypto?
This seems like it might be headed toward massively parallel protein folding or some such thing, where I don't want "the cloud" to patent my new drug before I can, but I want "the cloud" to do the work for me.
(Score: 2) by No Respect on Wednesday July 01 2015, @08:55PM
I have seen an application of this as a component of an online voting system. People submit encrypted ballots and the server is able to tally the ballots without being able to see the details of each ballot individually. The computational requirements are relatively heavy. There are also probablistic tests that can be run to verify the authenticity of the submitted ballots. I probably have some of the terminology wrong here, but that's the general idea. Each run of the tests provides an indication that the results are correct with 51% probability. After a significant number of runs one can say with high probability that the results are mathematically correct.
(Score: 0) by Anonymous Coward on Thursday July 02 2015, @01:39PM
But couldn't you reveal the content of a ballot by simply running the whole algorithm on that single ballot, and then looking who won that one-voter "election"?