Even though I came across homomorphic function while studying discrete mathemtical structures in undergrad, I only had a vague idea what application it had. An IBM fellow has found an encryption mechanism that is homomorphic. That is, if the encryption function is f(), then it is homomorphic on operation + then
f(a+b) = f(a) + f(b)
This means that if a is encrypted and b is also encrypted (with function f), then to add them, there is no need to decrypt a and b, to add them. You can operate on the encrypted versions themselves. Why? Because f is homomorphic.
This has profound implications on its applications. If everything is seen as operation on data, then one need not really know what the data is, but can still operate on the data by working on their encrypted versions.
When more of the computing is moving to cloud, data privacy needs to be assured for (see, for example, how Facebook tried to take ownership, but failed). Homomorphic encryption is a technical solution to the problem, as opposed to policy based solution (licenses, MoUs etc.)
This is just theory so far. Not all operations can be made homomorphic for example. Research has to be done along this dimension to find ways to make all operations homomorphic. This seems to be going to take long time as the article mentions that it will take at least a decade to get it done. 10 years is too much a time in this modern world to bet on a technology for long time. Who knows what ‘s in store for future?
Moreover, the functions may not be purely homomorphic. After some operations, the original data may be corrupted (who knows whether is function is purely homomorphic until it is proven?). The solution proposed by the inventor is to double encrypt the data and periodically re-encrypt the inner encryption layer by decrypting it.
What about the key management problems?