The introduction of intelligent technology is a subject that is simultaneously exciting and terrifying. The smart trinity of artificial intelligence, machine learning and deep learning has the potential to analyze data at a speed the human brain could never reach.

That the resulting insights could have profound beneficial effects for society isn’t debated. But there is a paradox around how to make data simultaneously open and shareable and retaining privacy and security.

“This intersection of privacy and AI is at the core of Intel’s data-centric mission,” said Casimir Wierzynski (pictured), senior director of AI products at Intel Corp. “How can you also respect the privacy and the security of the underlying data while still being able to train and use AI systems?”

Wierzynski spoke with John Furrier, host of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the RSA Conference in San Francisco this week. They discussed security and privacy issues with sharing data for machine-learning models and how Intel’s research in homomorphic encryption could solve the problem. (* Disclosure below.)

To share or not to share

One issue with machine learning is that it is almost always a multiparty interaction, according to Wierzynski. One party owns data, while another owns the model, which in turn is running on yet another party’s hardware. This makes sharing data an exercise in trust — not a good plan when the data may contain sensitive personal information that is a goldmine to criminals. Yet the promise of machine learning cannot come to fruition without sharing data.

One example is in the health care sector. “You have multiple hospitals that have patient data,” Wierzynski said. “If somehow, they could pool all their data together, you would get much more effective models, much better patient outcomes. But for very good privacy reasons, they’re not allowed to do that.”

The answer lies in a technology known as homomorphic encryption, he said. This leading-edge technique allows encrypted data to be shared and accessed for analysis. Any results are also encrypted, and only the actual owner of the data has the encryption key.

The inventor of homomorphic encryption, Dr. Craig Gentry, explains it as being like a locked glovebox. Anyone can put her hands in the gloves to manipulate what’s inside the box, but only the owner of the box can unlock it and retrieve the results.

“It seems like magic,” Wierzynski said. “With this capability, you enable all kinds of new use cases that wouldn’t be possible before where third parties can act on your sensitive data without ever being exposed to it in any way.”

One-click encryption for data scientists

Unfortunately, most data scientists aren’t also cryptology experts. So, Intel created a product that democratizes homomorphic encryption. The open-source package is called HE Transformer — where the HE stands for homomorphic encryption.

“It allows the data scientists to do their normal data science in Python or whatever they’re used to,” Wierzynski said. “But then they flick a switch and suddenly their model is able to run on encrypted data. If you can know that as long as [personal] data is moving around and people are operating on it but it’s staying encrypted the whole time, not just in transit, that gives a much higher level of comfort.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the RSA Conference:

Photo: SiliconANGLE