Decentralizing public safety
Using millions of cameras without sacrificing privacy
Millions of cameras on the streets feeding videos with our faces to anonymous government monitors is as scary as dystopian visions come. The good news is that it doesn’t have to be so, because we already have the tech that helps balance the growing capabilities of machine learning with the need to protect individual rights.
In 1949 when George Orwell wrote the famous “1984” novel, public key cryptography had not been invented. It took several decades before the idea surfaced first under classified military research and years later in mass use with internet protocols. As the internet gave birth to radically improved communications, it also made surveillance easier. Cryptography shifted that power from the government back to the individuals. We got the best of both worlds — improved communications, but without giving up our privacy.
Just like public key cryptography saved our communications from the overreach of governments, it can do the same for our faces on cameras in public spaces.
Cryptography for public safety
At Supervaisor we’re focused on traffic and what crowdsourcing of videos can do to increase public safety. We’ve collected over a quarter million anonymized videos already in Tallinn, Estonia. With all cars turning into sensors, we can all guess where this constantly updated “Google StreetView” is headed. It’s a matter of years, not decades until our surroundings are full of smart cameras.
While EU was busy proposing legal frameworks that would throw the baby out with the bathwater, we were busy building tech and intellectual property that helps strike the best balance between growing capabilities of machine learning and the resulting growing need to protect individual rights.
Here’s how our tech works. Consider this Google StreetView image from my hometown for example:
Let’s pretend for a second that the cat is dead and the kids are the main suspects in this horrific premeditated murder. Since this is EU, where personally identifiable information can not be stored without consent or other valid legal bases, we can assume that Google does not hold the original data and despite discoverying what could be important evidence in the context of a crime, Google would be unable to help the government in the identification and prosecution of these two hardcore criminals.
With our solution Google (or a third party) can use public key cryptography and scramble the faces using a government provided public key and do all that on an edge device. This way no faces would ever land on Google servers and they would not be able to reverse the scrambling, yet they could still help the government solve the crime.
AI can continue processing this media with the goal of identifying criminal behavior and when identified, Google can still decide if there is enough context to justify government involvement. In case the evidence is convincing they can provide the media in question to the government who, having received the newly discovered suspicions, has both the legal grounds and the only key to reverse the scrambling.
Here a private company does not effectively have personal data, yet in justified circumstances can provide data that can be transformed into personal data by a party with legal grounds and the only key to do so.
Our cars can help others
With the growing amount of vehicles with cameras around us constantly growing, the question for automotive companies effectively becomes —
Would you like your vehicle to be able to help the victims of crimes?
We can take this a step further and actually get private companies or citizens involved not only in capturing public safety related data, but also in helping identify crimes (mainly by competing on writing AI software that does so). With the privacy protections in place the process of identifying crimes or even voting for whether the events should be looked at by the government can be outsourced to the public. We tested interest for this with Traffic Tinder and are positive that with the right incentives and privacy protections in place, the public can be actively involved in improving our safety.
This isn’t in fact a new idea, but an idea proposed in 2019 by Nick Bostrom from the University of Oxford in his work — The Vulnerable World Hypothesis:
“In theory, some of the monitoring could be crowd-sourced: when suspicious activity is detected by the AI, the video feed is anonymized and sent to a random 100 citizens, whose duty is to watch the feed and vote on whether it warrants further investigation; if at least 10% of them think it does, the (non-anonymized) feed gets forwarded to the authorities”
Getting crowds more involved to develop laws that are more informed
Juries are a form of crowdsourced judgement and taking jury sizes to statistically safer (more significant) grounds in limited contexts is likely how it starts. Discovery of laws that the population does not consider worth enforcing could be a valuable signal as well.
With the pandemic our trust in governments ability to handle our safety has been shaken. Arguably it is the private sector along with the market forces that has both the competence and agility to solve problems most efficiently. Perhaps now more so than ever, we should all be putting our heads together to work on tech that can better bridge the two. We’ve cracked an important piece of the puzzle for the visual space and can’t wait to see where it all takes us.
If you’re interested in using our privacy preserving solutions for video, reach out on email or DM me on twitter ( @keskkyla ).
If you found the article interesting, don’t forget to “clap” below to show your appreciation.
