Blogs

A tighter hold on personal data ensures more secure AI

As edge AI and federated learning become practical technologies, our ability to ensure the privacy of individuals is respected in the processing of personal data will develop. MyData thinking will also likely become more common. Professor of Computer Science at the University of Helsinki and artificial intelligence expert Teemu Roos believes that the disparity between users and service providers will be reduced by regulation and new technologies.

Using AI image recognition, service providers like Facebook can identify various social networks with a high degree of accuracy.

“At the moment, we aren’t necessarily able to determine exactly where personal data ends and what other data services combine it with,” says Teemu Roos, Professor of Computer Science at the University of Helsinki, who specialises in AI research.

He is optimistic that regulation and the technologies that make it possible will be improving our privacy protections in this regard in the near future.

“The EU’s proposal for an AI regulation published earlier this year especially has been a wake-up call for operators that utilise AI,” Roos explains. The proposal promotes secure, reliable and ethical AI development and, consequently, supports the EU’s protection of fundamental rights.

“The EU doesn’t want to compromise when it comes to the rights of EU citizens. The realisation of individual rights takes priority over the competitiveness of service providers,” Roos says.

Not everything belongs in the cloud

Methods that ensure that the privacy of individuals is respected in data processing have started moving off researchers’ desks and into real-world implementation. According to Roos, the key terms in this development are edge AI and federated learning.

“When data is increasingly being processed on users’ devices, on the edge of the web, it no longer finds its way en masse into large data reserves in the cloud, where it is processed and combined with the data of other users,” Roos explains.

Federated learning has the same goal: to not combine user data. Using federated learning, data that is scattered across devices can be used to construct predictive machine-learning models for providing services in such a way that individual users are not identifiable.

In short, Roos expects that the current disparity between users and service providers will be reduced by regulation and new technologies.

“The opposition will hopefully no longer be between individual users who have nothing but their own data and giant service providers like Facebook that can utilise the data of hundreds of millions of users,” Roos says. This would allow users to manage their data in such a way that combining it with the data of other users is only allowed in the contexts of their own choosing.

You own your own data

In the next few years, Roos would also like to see users’ control over their own data increase with the popularisation of what is called ‘MyData’ thinking.

“MyData completely topples the prevailing balance of power. Currently, services like Netflix effectively own the data related to my browsing history. But MyData thinking would mean that I control my own data. If I decide to switch to a competing service, I can take my data with me,” Roos explains.

This way, new services would not need to build up the browsing history they use to recommend content to users from scratch. Instead, it will be up to date from the moment users join the service, because they bring their own data with them.

Achieving gradual results

The content users watch on streaming services may soon be getting entirely new features, thanks to the development of AI technologies.

“It’s really fascinating to see what creative industries can come up with when creators get their hands on AI tools and start exploring their possibilities in innovative ways,” Roos says.

New, complicated technologies do not generally become fully functional pillars of everyday life in one fell swoop. Teemu Roos hopes that voice user interfaces and chatbots, which have been gradually coming into use for some time now, will soon be reaching a level of maturity where users can be very comfortable using them. Roos remarks that, for at least the past five years, he has been saying that these technologies will be revolutionising our everyday lives within ten years.

“Even though I’m well practiced with touchscreens, it’s still easy to accidentally hit the wrong thing, especially if you’re moving while using them. With any luck, we’ll be taking a big step forwards in 2022 that will let us interact with our devices using speech in a way that’s flexible and smooth enough that we’re not on the edge of a mental breakdown every time we try it,” Roos chuckles.