-
Notifications
You must be signed in to change notification settings - Fork 25
JAR_ICE_notes
Because of exposure to corrupting influences (e.g. Wittgenstein's Philosophical Investigations) at a tender age, I squirm when I hear uses of relations such as "denotes", "is about", "mentions", "means", "prescribes", "specifies", "intends", and so on that putatively hold between an "information artifact" and something else (a specific thing, a class, a property, a verb, a state of affairs, an inclination, etc.). All these words, unless accompanied by an explanation, are mysterious and they hide complex processes that we don't usual have the energy to analyze. When these things can be tied to some particular theory or mechanism, I'm not too worried, as in the mathematical framework of denotational (compositional) "semantics" where "meaning" is defined as a point in a mathematical space, without any reference to the world. In some cases I'd even be willing to accept maybe one or two unexplained relationships (like one of the terms I listed) hoping that through further use and testing their nature will become clear. But in a context where you would strive for scientific ideals such as objectivity, reproducibility, and falsifiability, nailing down any of these terms in a satisfying way just seems excruciatingly difficult.
Although it is subject to all sorts of criticisms, the BARG (basic action-response game) model of ethology is appealing to me because it seems to be on the road to being falsifiable. And now that the connection of IAO to OBI, together with the idea of attaching the sender's is becoming clearer to me I am seeing some connections between BARG and OBI, and this may point the way to better explanations. So I will try to expand on this below.
I'm not suggesting that ethology necessarily has any relevance to IAO other than as a level of analytical quality that we should try to match.
BARG works (or is described) like this: There is an agent A, and it has some deficit or objective O, for example to eat or to mate. It creates and sends a message M through some medium. M may received by an agent B (perhaps not, or perhaps by more than one). On reception B performs some action X. The outcome of X is some payoff (perhaps negative) for A, and perhaps some payoff for B. Based on the two rewards, and the idea that A and B are drawn from populations that behave more or less similarly, we can perform the usual sort of behaviorist and/or evolutionary analyses of why A and B did as they did regarding M, and what they might do in the future.
The first analysis done by ethologists is to decide under what circumstances A will "tell the truth" or, equivalently, in which B will correctly interpret the "meaning" of M. The patterns one sees in nature will have truth-telling at some adequate frequency since lie-telling strategies will be extinguished. (The problem of truth has been placed outside the scope of IAO, if I understand correctly.)
For example, a male widowbird wants grandchildren, so it grows a long tail (M), and displays it (visual medium). A female B reads M, measures its length, and on this basis judges A's assets and earning potential (its suitability as a mate). If the outcome X is for B to mate with A, and A really is as fit as he seems (likely given that he had discretionary income that let him buy the long tail), the payoff to both will be grandchildren.
The connection with OBI is that in OBI A can compose a message M in order to satisfy some objective; M might be, for example, a plan (specification) that someone (e.g. B) can carry out in order to achieve A's objective (conduct an experiment). This seems nicely operational to me. Not only is it an instance of BARG, but it's close to the phenomenon of imperative programming languages, a domain in which "semantics" is unproblematic, if you disregard the computer's relations to other entities and pay attention only to what happens on its boundary (I/O).
So we need an accounting of the "generically dependent continuant" angle. In the above M is a quality (member of Quality class) or some other "dependent continuant", I think. The distinguishing feature of things like M is their ability to be copied (a mysterious term - let me leave its nature hanging for a while). That is, there a "copying" processes that create new qualities M', generally inherent in some new bearer (not M's bearer that is), where M' is related to M in that M' will have the same effect on B that M would have - namely it will help to achieve O.
B is drawn from some population of B's that are able to help A achieve O. O may not (does not?) care which B "reads"/"executes"/"interprets" the message, as long as O is achieved.
We might say that M is original (it did not originate in a copying process) and M' is a direct copy of M. Now define (as a term of art; fix the label later) is a copy of as the reflexive and transitive closure of is a direct copy of. So starting with an original M, there is a population of qualities {M, M', M'', ...} of copies of M. One flavor of the information content entity story is that the ICE induced by an original M is generically dependent on each member of that population - and on no other such qualities. That is, this ICE could never inhere in a quality M2 that is not a copy of M.
I think this is a stronger definition of ICE than what IAO has now. Of course it may be wrong! It rules out the possibility of multiple origins of the "same thing", and that's OK for unlikely M's. But independent origin might happen through lawful processes of composition, such as concetenation. (This is the EQ / EQUAL problem in LISP.) I haven't figured this out yet.
We also have translation and format conversion to account for, but this fits. The message M is composed with a particular population of enabling agents B, B', ... in mind. But there may be other agents C, C', ... that, given a different quality M+, could achieve the objective O. So an information conversion process is one with quality M as input and M' as output, where M' enables an agent drawn from a different population to induce the achievement of O.
This is not the end of the story; to turn this into a complete "realist" or BFO-like story we have to say what these qualities M really are, and what objectives are, and check that the resulting system isn't circular. For example, are these qualities in 1-1 correspondence with objectives (are they "meaning"), or are they simply syntactic qualities that are lucky enough to have a pedigree? How do we circumscribe the classes from which M-like things are drawn in such a way that we curation consistency and testability (or at least some semblance of objectivity)?
One thing about the qualities M is that they are intentional in that not only did A compose M in order to achieve O, but they did so with some reasonable expectation that it might do so, by virtue of being interpreted by some B. That is, they are not random, frivolous, or coincidental.
How do we account for parts of ICEs, atomic ICEs such as words, and ICEs that are "informative" or "declarative" rather than objective-enabling? Words and noun phrases do not seem to have "objectives"; and parts of documents may not have adequate "meaning" when removed from "context". -- I'm not sure, but I would be inclined to try to figure out, or invent, "objectives" for these things, or to figure out how to broaden "objective" so that it can apply to these things. Here we might start wandering into theories like Montague semantics, where parts of utterances are interpreted as functions of (dependent on) the things that are near them. If there is a remix, and part P that had been with Q1 is now with Q2, then the function remains the same, but the objective (P Q2) will differ from the objective (P Q1).
In any case "originality" seems an interesting property. You can apply it to words and identifiers, where it means that the assignment of meaning (objective) is fresh and not derived from in a lawful way from the meanings of its constituent parts (letters, digits). But it also applied to creative cultural works, those subject to copyright. These are "original" in the sense of "really unlikely to have been produced by chance", and as a result copying can be detected, argued in a courtroom, and protected.
It should come as no surprise that I am applying here some of the apparatus I'm familiar with from computer science, evolutionary biology, and copyright law. Which of these ideas is appropriate or useful for applications of IAO is hard for me to judge.