> Sounds like the bigger issue is that you're able to get "spatiotemporal" data in the first place?
Almost all data is spatiotemporal data, people just aren't used to thinking about it like that. Everything that "happens" is an event with associated times and places.
Tagging of events with spatiotemporal attributes, or with metadata that can be used to infer spatiotemporal attributes, is pervasive. Every system data passes through, even if not the creator of it, observes the event of the data passing through it. Event observation is not trying to track things but it implicitly and necessarily creates the data that makes tracking and spatiotemporal inference possible.
These kinds of analyses rely almost entirely on knowing the events occurred; you could encrypt the contents of the data and it wouldn't matter. Software leaks spatiotemporal event context everywhere across myriad systems, internal and external, that incidentally collect it. There isn't anything nefarious about most of it and much of it is required for reasons of criminal and civil liability.
What people underestimate is that you can analytically stitch together many unrelated sparse data sources with spatiotemporal attributes, many of which are quite crap or seemingly unfit for purpose, to reconstruct a dense high-quality graph. Counter-intuitively, diverse and seemingly irrelevant data sources often produce better data models. It surfaces bias, errors, manipulation, and processing artifacts in individual sources you might otherwise miss.
It is much more difficult to access the obvious first-party data sources than it used to be, mostly because people with that data are far more selective about who they give access. It doesn't really matter, that is a speed bump for the unsophisticated. The exponential growth in the scale and diversity of network-connected telemetry of all types pretty much guarantees these data models will always be constructible.
The historical limiter has always been the absence of data infrastructure platforms that can handle these kinds of analytics at scale.
>Tagging of events with spatiotemporal attributes, or with metadata that can be used to infer spatiotemporal attributes, is pervasive. Every system data passes through, even if not the creator of it, observes the event of the data passing through it. Event observation is not trying to track things but it implicitly and necessarily creates the data that makes tracking and spatiotemporal inference possible.
>These kinds of analyses rely almost entirely on knowing the events occurred; you could encrypt the contents of the data and it wouldn't matter. Software leaks spatiotemporal event context everywhere across myriad systems, internal and external, that incidentally collect it. There isn't anything nefarious about most of it and much of it is required for reasons of criminal and civil liability.
>What people underestimate is that you can analytically stitch together many unrelated sparse data sources with spatiotemporal attributes, many of which are quite crap or seemingly unfit for purpose, to reconstruct a dense high-quality graph. Counter-intuitively, diverse and seemingly irrelevant data sources often produce better data models. It surfaces bias, errors, manipulation, and processing artifacts in individual sources you might otherwise miss.
That's a lot of technobabble for what essentially sounds like "there's some ad SDK that's phoning home with your gps/ip geolocation every few minutes, if you cross reference that with when flights are, you can guess what flight someone took". How far off am I? Or is there some galaxy brained AI that can infer that from disparate facts like that you stopped posting on twitter for 12 hours, your car's license plate was caught by an ALPR to be heading towards the airport, and 3 weeks ago you visited some portuguese tourism site that had an ad beacon installed?
Thank you for cleaning that up.
Education starts in the home: if it's not locally runnable and useable offline, it does not exist. We need to teach people how to be sneaky kids trying to sneak all things past authoritarian parents. That mindset is what will drive otherwise lemmings to doing things like making Qubes OS primary, getting a Google tablet and installing GrapheneOS on it, building a 48 hour battery life "comms bag" which is an LTE modem (or 5G) + a good OpenWRT capable router + battery packs and charging equipment.
Idea is: baseband is divorced from application processor and packed away into a separate radio station which can be brought online completely under the owners' control. That will be my next "cell phone."