“Post-truth” was named Word of the Year by Oxford Dictionaries in 2016. Fighting deception and lies and establishing trust in information is a significant task for Western countries. Why? Because trust and truth are the pillars of democracy. Without them, democracy will fail.
AI-s fighting each other
Sentinel is engaging the most complex forms of deception: deepfakes. Prefix “deep” signifies concepts that go beyond the human plane. Thus, deepfakes are video imagery created by AIs, better of which are indistinguishable from real ones. They differ from conventional digital forgery by also escaping detection by humans. Therefore, Sentinel is relying on its AI to detect forged videos. And hence we witness the arrival of a new form of warfare: AI-s fighting each other in the digital realm.
Over 100M deepfakes
Sentinel has just closed a 1.35M $ seed round and joined the arms race in deepfake detection. Because the technology is advancing at an incredible speed, the race is aiming at moving targets. Since 2019, the number of deepfakes online has grown 900% and passed 100M. Advancements have driven this exponential growth in AI algorithms, lower cost of computing, and exponential growth of data on the Internet. Therefore Sentinel, unlike its competitors Microsoft and Deeptrace, is using multiple layers of defense instead of one single neural network.
Four layers of defence
“We think it is impossible to detect all deepfakes with only one detection method,” Sentinel co-founder and CEO Johannes Tammekänd tells. “We have multiple layers of defence that if one layer gets breached, then there’s a high probability that the adversary will get detected in the next layer.”
At the current stage, Sentinel is lining up four layers: an initial layer based on hashing known examples of deepfakes to check against; a second layer comprised of a machine learning model that parses metadata for manipulation; a third that checks for audio changes, looking for synthesized voices; and lastly a technology that analyzes faces “frame by frame” to look for signs of visual manipulation. Based on these layers, a „certainty score“ is assigned to the video.
“This indicates that we have already reached the point where we cannot say with 100% certainty if a video is a deepfake or not unless the video is “cryptographically” verifiable, “ Mr. Tammekänd explains.
Founders used to work for NATO and the Royal Navy
Because the stakes in digital security are high (think the coming US elections, for example), Sentinel, whose founders include ex NATO and the U.K. Royal Navy employees, choose their clients carefully. They recently turned down requests from Saudi Arabia and China and work only with NATO allies.
There are likely significantly more AI-generated media in the future so that instead of just one video whole plethora of synthesized media is created reflecting complete fake events. Furthermore, this fake content is hyper-personified based on the data we leave in the cyberspace to be even more susceptible to believe this. This type of disinformation could influence the fate of whole countries and markets.
Estonia’s unique position
Therefore Mr. Tammekänd’s vision for the next decade is to provide authentication solutions scaled up to possibly all Internet users. Estonia has a unique position for this kind of service because, for Estonians, the ID card mandatory for all citizens offers digital signing and encryption already now. This, Mr. Tammekänd sees, could be the foundation upon which to extrapolate. Similarly to fighting against computer viruses, authentication software will evolve on two tiers: certification by creators and detection by users. Thus, when the first clients for Sentinel’s products are states, commercial actors will soon follow. And very shortly, these defensive AI-s will become familiar allies for each Internet user as antivirus software is today.
See the original article here.