A type of encryption called homomorphic encryption is gaining traction among major tech companies as well as agencies like the NSA (via Cyberscoop).
Right now it’s impossible to run computations on encrypted data. In order for algorithms like machine learning to work with data, it has to be unencrypted. With homomorphic encryption, data could be encrypted and still worked with, greatly increasing security.
We think that whoever cracks homomorphic encryption is going to have a lock on the next generation of cryptography. There’s a lot of people who don’t quite understand what it means, or assume it’s just marketing and it can’t possibly work. But it’s all really cutting edge stuff, and there is a lot of money going against it because we all know we’re getting close.Mark Horvath, Gartner analyst
Current methods of this encryption are too slow for practical use. In the ’80s and ’90s researchers tried to come up with homomorphic schemes. Some worked with OR gates (addition) and others worked with AND gates (multiplication). But the schemes couldn’t work with both (NAND gate).
Achieving NAND computation is a big deal because you can derive all of the other logical gates from it, and work on NAND schemes is currently being researched. Just don’t tell Australia.