The end of anonymity in the crowd is near

Not an endorsement of this technology or the use of it, just stating that it will happen.

Given sufficient measurement precision, all humans have unique genomes and fingerprints, but also faces and voices. The first two are well known and somewhat difficult to measure. However, the last two are very easy to measure, even at a distance. In the next few years, massive datasets will be built of public, semi-public and leaked private data linking people between all services with available data, for all available time periods. This first and foremost includes social media like Facebook, Instagram, Linkedin, but also Youtube, and every dating and porno sites. There are a host of voice-only services like SoundCloud that currently handle anonymous users, which is also true for Youtube and porno sites. All these people will be automatically identified and linked in the near future. It will not be possible to take part in a public demonstration without a mask (illegal in many places) which cannot later to matched to you. It will not be possible to take part in amateur or paid-porn without a mask and maybe without being silent (even moans can be matched in all likelihood).

Because of the above, services will open whose business is based on this. These can be completely legal, or maybe illegal depending on the local jurisdiction. However, they will inevitably be created. There will be sites like and which have obvious and enormous markets. The less moral, depending on your view, services will send automatic blackmail to you, saying that you don’t pay a monthly fee, your family, spouse and friends will be informed of your misdeeds, whatever they are. Don’t think they can’t find your relations — you put them on Facebook, Linkedin, and in any case, family can be inferred from name and facial similarity anyway.

What shall we do about this technology? One can distort the voice to make it (probably) unmatchable, but one probably cannot do anything about the face except for masks, which are inconvenient.

Update: 2020 March

This paper from 2014 predicted the same stuff as me, and even carried out a proof of concept experiment effectiveness:

In 1997, the best computer face recognizer in the US Department of Defense’s Face Recognition Technology program scored an error rate of 0.54 (the false reject rate at a false accept rate of 1 in 1,000). By 2006, the best recognizer scored 0.026. By 2010, the best recognizer scored 0.003 an improvement of more than two orders of magnitude in just over 10 years.

In 2000, of the approximately 100 billion photographs shot worldwide, only a negligible portion found their way online. By 2010, 2.5 billion digital photos a month were uploaded by members of Facebook alone. Often, those photos showed people’s faces, were tagged with their names, and were shared with friends and strangers alike.

This manuscript investigates the implications of the convergence of those two trends: the increasing public availability of facial, digital images; and the ever-improving ability of computer programs to recognize individuals in them.


Leave a Reply