The Profiling Machine (PM) is a distributed system emerging from1 the interaction between a database DB and a discrete set of processing elements A, whose elements are denoted as a2.
DB is a database formed of two sets: a (public) set O of objects o and a (private) set P of profiles p. An o may denote any object whatsoever -- details are beyond the current scope, although it is usually expected that os are similar in type/kind. For any p in P, p is uniquely mapped to an a, the information signifying so-called "knowledge" in DB about A.
as may interact with DB in the following ways: they may add objects o to O; they may remove any previously-added objects o from O; they may query O for an object, which is selected by PM according to some internal logic; and upon receiving an o, an a may "approve" it or otherwise "reject" it3.
The so-called object selection algorithm is in general a function of DB in its entirety. More specifically, when a new a is added to A (it "joins" the system), PM may sample os uniformly from O; when an o is approved, PM will update pa (in P) with a relation of weight w between a and o. As new os are approved by a, P will in fact converge to a graph containing relations (of various weights) between as and os, and correlations between (on one hand) as and (on the other) os. While convergence may be avoided by introducing some fixed or variable level of noise, it is expected that as the cardinality of O goes asymptotically towards infinity, correlations will lead to clusters (grouped subsets) of os on one hand and as on the other.
More precisely: if a1 and a2 are in (strong) relation to objects in Ra1 and Ra2 respectively, then the intersection between Ra1 and Ra2 forms a correlation between a1 and a2. Conversely, if o1 and o2 are in strong relation to agents in Ro1 and Ro2 respectively, then the intersection between Ro1 and Ro2 forms a correlation between o1 and o24. Both these correlations are fed back into P and thus into the selection algorithm of PM.
The primary goal of this feedback loop, and of PM itself is for DB to maintain a P that is as accurate a reflection as possible of A, under the assumption that as in A are equipped with quasi-stable5 criteria for (dis)approval of objects in O. The so-called correlation clusters are particularly interesting, since they may converge over large populations, i.e. large particular subsets of A may tend to approve large particular subsets of O. This approach may even prove to be useful for marketing, and in particular for those with such inclinations, may end up an outright implementation of a so-called tick-tock.
Anyways, the particulars of such a system and (especially) its failure modes might be worth studying some other time.
-
Distributed systems "comprise" distinct, perhaps disparate objects that interact, and thus they are fundamentally emergent objects. For example, a computer, in the von Neumann sense, is a distributed system emerging from processing units, memory, peripherals and the interaction between all these elements. Similarly, a computer network is not simply composed of computers and the wires between them, but it also specifically requires the underlying phenomena of interaction, i.e. "communication protocols" in order to be considered an object we can talk about, that is, a so-called "system".
Believe it or not, this is not mere pedantism, but rather an attempt at rigorousness for the sake of avoiding many of the traps befalling this field of computing. ↩
-
afor "agent", that is, an autonomous computational unit. ↩ -
This may be implemented in several ways. For example,
PMmay explicitly expect a reply to the message containing theo, or otherwise if theadoesn't reply within some timeframe, it may consideroimplicitly rejected. Moreover, the same mechanism may be complicated indefinitely, such by introducing arbitrary "levels" or "scales" of approval. All these details are beyond the scope of this specification. ↩ -
Note however that, while
as are completely opaque toPM,os are actual objects inDB, so they may also be correlated according to their particular properties.In case this wasn't clear by now,
PMaims to obtain a profile of eachait interacts with. It may of course be the case thatas also aim to obtain information aboutPM, but this particular bit is out of the scope of this specification. ↩ -
Stable at some particular point in time, but not necessarily over a sufficiently long period of time, however this "sufficiently long" may be defined, or otherwise determined. ↩
[...] and the what is as important as the analogy. Anyway, the system under scrutiny is an extension of The Profiling Machine. The input of this new system consists of particular states of DB, in particular of P, as per the [...]
[...] the interaction between humans and machines would not only make machines "more human", but also the other way around! I'll admit, I didn't realize this out for a long while either, until one fateful day when I had to [...]
[...] for the whys behind this act of dehumanization: why does the state collect taxes from you? why does TickTock show you some particular piece and not the other one? why does Google's machine, or for that matter [...]
[...] fashion, except unfortunately the whole thing works both ways: if you consistently spend your days netflixing, then at the end of it all there won't be much "you" [...]
[...] to the sacred principles of their founding fathers? What do you think? Were the Netflixes and TikToks of the time satisfying enough for the good folks down there? and for how [...]
[...] machines, all the way from before ye olde Tubes to today's platforms, are merely cheap surrogates for systematized education or portions thereof. That they breed [...]
[...] the second guy pops his head out of the crowd to talk... what do you think could happen? Well, the machine pops in for a course correction, what else the fuck could it do? If you're demoralized by the [...]
[...] recall, I expect Iu Da, or The Profiling Machine: both pieces deal with such platforms, from their respective concrete and abstract perspectives. [...]