2+ Million Views Viral Tweet On IPVM Dahua Protester Report, Boosted By Tech Investor

Published Jun 01, 2023 12:10 PM
PUBLIC - This article does not require an IPVM subscription. Feel free to share.

As fears about AI and the PRC rise, an IPVM reporter's tweet went viral, amassing over 2 million views, boosted by a prominent tech investor.

IPVM Image

The May 30th tweet shared our report - Dahua Selling Protestor / Banner Alarms, Deletes Evidence:

IPVM Image

Balaji Background

The virality was boosted by a prominent tech executive. Balaji Srinivasan is a tech investor who has nearly a million followers regularly making strong claims, e.g. Srinivasan recently bet - and lost - $1 million on Bitcoin reaching $1 million, boasting "I just burned a million to tell you they're printing trillions":

Srinivasan was formerly a General Partner at Silicon Valley VC Andreessen Horowitz and the CTO of Coinbase; he currently describes himself as an angel investor.

Goes Viral Claiming "True Dystopia"

On May 31, Srinivasan quote-tweeted IPVM's Dahua article, claiming the "true dystopia" would be PRC police identifying every protestor in a crowd and immediately freezing their WeChat accounts, a "human EMP" since WeChat is essential to PRC citizens' daily lives.

The tweet went viral garnering over 2 million views and almost 2,000 re-tweets:

IPVM Image

In a lengthy follow-up tweet Srinivasan claimed this "face freeze" could be done by PRC police officers wearing facial recognition glasses, sharing a 2018 WSJ article on the topic. For protestors wearing masks, Srinivasan said gait analytics could be deployed although he admitted this "may be hard in a crowd":

IPVM Image

Srinivasan claimed it is now possible for the PRC to build a "true dystopia" where police equipped with facial recognition glasses and gait analytics instantly match protestors and freeze their social and financial accounts.

Another prominent tech investor and former A16Z partner also shared and commented on the IPVM's reporter's tweet:

IPVM Image

As well as a BBC China Correspondent:

IPVM Image

Face Recognition Glasses Feasibility Examined

Our facial recognition testing and ranking indicate facial recognition glasses would be extraordinarily error-prone unless people were forced to stand still and look at the camera. Indeed, in the 5 years since that WSJ report, we have seen few even make marketing claims about doing this, which signals that the technology is fairly impractical.

In protests, it is much harder to get usable footage for a crowd of protestors since they would not fit well into the FOV. Also, a police officer using these glasses at a protest may get unstable/shaky footage given how hectic protests can be.

More broadly, using face recognition to instantly recognize faces in a protest is very challenging since the technology suffers significant accuracy issues when presented with harsh angles, low light, face masks, hats, glasses, etc, all of which are more likely during a protest.

Providers often tout 90%+ accuracy but this is only possible under good conditions. IPVM tested Hikvision's facial recognition NVR in 2023 and found accuracy plummeted to 0-10% when subjects wore hats, sunglasses, and/or had moderate-to-harsh down-tilt angles.

The above accuracy issues mean false positives are possible, which, under Srinivasan's scenario, would result in random people getting their WeChat/financial accounts frozen, causing unnecessary trouble for PRC authorities.

Gait Recognition Claims

Gait recognition is also unlikely to be of much use during a protest since protestors move or stand around in atypical ways. Further, gait recognition is far less accurate and less field-tested than face recognition, with one PRC company (see IPVM's Watrix Gait Recognition Profile) focusing on it which has raised ~$61m, unlike face rec which has numerous large competing PRC firms like SenseTime, Megvii, Yitu, CloudWalk, and DeepGlint who have collectively raised billions.

For more, see Gait Recognition Examined and Single Frame Gait Recognition From Michigan State and Osaka University Examined.

PRC Surveillance During Protests

The PRC already has extensive live facial recognition camera (Sharp Eyes) systems that can help recognize suspected protestors and pick out key leaders, which is more practical than instantly finding all protestors in a screenshot and freezing their WeChat. PRC police also have other methods to identify protestors, like using cell signal towers to pull all phone numbers from a protest location, The Washington Post reported.

The PRC does use WeChat to hamper protests, albeit in a different fashion than Srinivasan's scenario. In 2019, a BBC reporter's WeChat account was temporarily blocked as soon as he posted photos of people holding candles during a Tiananmen Massacre memorial in Hong Kong. The reporter had to submit a selfie ("faceprint") to get his account unlocked.

Sci-Fi Solutions Unnecessary

Overall, PRC police are already well-equipped to track down protestors without needing sci-fi methods like face rec sunglasses, gait recognition, and instant WeChat freezes.

Shows Increasing PRC Tech Scrutiny

On the other hand, the virality of IPVM's and Srinivasan's tweets shows that PRC surveillance technology is more unpopular than ever in the Western world, attracting negative attention for its deployment against protestors, Uyghurs, and others.

This is a contrast to the mid-2010s era when PRC surveillance tech boomed thanks to its low cost, with few commentators caring about how it was being used in the PRC.

Comments are shown for subscribers only. Login or Join