Kurt Rohloff: As far as the work that I’m doing on fully homomorphic encryption, is in some sense quite groundbreaking, and sometimes I think it’s also the primary breakthrough in theoretical computer science of the 21st century. For a long time people would use encryption technologies to protect their data. The data come once the data is actually encrypted, you would be able to give it to someone – say as a cloud host or someone else – where that cloud hose can pull the data for you. But if the data was encrypted, the host would not really be able to do much with the data, other than give it to someone else. With homomorphic encryption, it’s a way of encrypting a set of data so no one can touch it, no one gets access to it. But then you can give the encrypted data to someone you don’t necessarily trust – like the Amazon cloud – and then run computations on the encrypted data without sharing keys, without giving any access to the encryption to the cloud host. That cloud host would run a computation on the data, get an encrypted result, give the result back to you. You could then decrypt that result, and that decrypted result is the same as if you had run the original source computation on the original data without encryption. So in some sense is a fundamental black magic, in that you can actually enable computing on encrypted data. This has tremendous implications for multiple industries, for government, and for the society at-large. Particularly, let’s look at the healthcare industry. In healthcare there are large issues of privacy. All this data is being generated by all these patients at all these hospitals and all these various insurance agencies across the world. There is very strong both privacy concerns associated with patients not wanting to share that data, and also very strong commercial competitive concerns, where insurance companies don’t want to share information about their subscribers and give their competitors any kind of information about who they are receiving their money from. At the same time, there would be a large benefit if researchers, particularly with funding with the FDA or other agencies, could get access to broader sets of patient data to run computations on those data to compute, for example – what is the most beneficial treatment under certain observed symptoms? Or what is the most effective drug to prescribe to a patient under certain symptoms to maximize patient outcomes, minimize patient death, and for overall betterment of patient health?
The trouble with this though is because the patient, because the insurance companies, because the hospitals have these very strong privacy concerns – and very valid privacy concerns – they’re less willing to share the data. With homomorphic encryption, we provide a way – if enabled – were patients, doctors, insurance agents could encrypt data, share the data, and enable competitions on this data to better provide better healthcare for their patients. The trouble with homomorphic encryption nowadays though is although it’s been demonstrated to be theoretically feasible, it is still not completely practical. So my work with DARPA and my work here at NJIT and my work with my collaborators – both in industry and academia and the government – has been focused on making homomorphic encryption practical. By practical, I mean that making it run fast enough. There are very, very strong runtime concerns associated with homomorphic encryption. In particular, original implementations of homomorphic and corruption have showed that just encrypt into bed and running a [bit wise and] operation, probably the simplest operation you can run on logical data, took a half-hour. With advances I’ve been making with DARPA, we basically reduce that by five to six orders of magnitude in the past three years to the point where we’re streaming data from iPhones to Amazon cloud and running computation on encrypted data in Amazon cloud with practical applications. Other implications of this research include it could help, for example, the financial industry – for example, the Department of Treasury to be on the better scan to find folks who are not fully honest on their taxes, while still maintaining privacy off our tax base. It could also help in general society, general commercial industry, where we’re constantly at threat through cyberattack – both through malicious insiders and by encrypting data that we compute on using homomorphic encryption. We can better protect ourselves from these malicious insiders. Another concern is that industry often are shy about sharing information about cyberattacks they’ve come under, primarily because it would cause them embarrassment. It might cause shareholder issues, where it would decrease stock prices. However if industry was better able to share information about cyberattacks, to share information to identify what the common attack factors were. Who might be propagating these attacks? By sharing this information, we could all better protect ourselves, protect our banks, protect our power companies, protect our healthcare providers. Homomorphic encryption provides a path that would enable victims of cyberattacks to privately share information about the existence of cyberattacks so that we as a society can better defend ourselves from the cyberattacks and better share information without fear of the victims of cyberattacks becoming unnecessarily embarrassed by the existence of these attacks while still getting benefit from the knowledge of their attacks.
In terms of where the research is going, so up to now we’ve had I think some really nice successes about making homomorphic encryption real. One of the challenges with it is that it’s quite hard to use in that it has runtime issues. It’s a very difficult compute model. It’s difficult to use these technologies. Even more pressing, the software that we’ve been writing for these technologies is in some sense also very difficult to use. It’s in some sense, a lot of the technology that’s out there is called proprietary. It’s not freely available for use. One of the things that I’ve been focusing on with my lab and my researchers and my collaborators is to develop tools, particularly open-source tools, that will make this technology much more broadly available and much more usable; and developing implementations that will make this technology more broadly applicable – so particularly writing very fast implementations for standard enterprise computer architectures, even for high-performance commodity computer devices like GPUs, and developing tools such as a programming language and compilers that general developers can use and not specific experts in homomorphic encryption. Where I see this work going in the next several years is that we’ve been having some very good successes, general improvements to this technology. But my belief is that this technology is going to primarily be driven by application use going forward. So I’m looking and working with industry and government to identify where the pressing needs are and identify how we can apply this technology and how we can adapt our tooling that we’re developing to meet the specific needs of both industry and government. So I’m doing this with both my students and my collaborators, and I’m looking forward to doing this over the next several years.