Serendip is an independent site partnering with faculty at multiple colleges and universities around the world. Happy exploring!
The Classification Problem: Implications for Intentionality
![Paul Grobstein's picture Paul Grobstein's picture](https://serendipstudio.org/exchange/files/pictures/picture-2.jpg)
The Emergence of Form, Meaning, and Aesthetics
Evolving Systems:
Open Conversations
The Classification Problem: Implications for Intentionality
Wil Franklin
5 November 2009
(to be continued 12 November)
The issue (excerpted from Wil's discussion notes):
There are three interesting intersection between the Classification Problem (CP) and evolving systems - particularly with the storytelling, intentional agents:
- First, do categories exist independent of the classifier? What is the relationship between Nature vs Classifier; Code vs Decoder; Reality vs Perceiver and perhaps Maker vs Audience.
- Second, how does one classify or even identify intentionality?
- Finally, if intentionality is reduced to the assumption of a “maker”, then what happens if we let go of the assumption of a “maker”? What becomes of human intentionality if the universe and all that is in it has no beginning, no end, no maker, only the interaction between code and decoder? Does it even matter whether we assume there is an intentional maker or not, if there is no why to distinguish between the two alternatives?
5 November meeting summary (Paul)
Wil introduced the session be picking up on prior discussions of intentionality/story telling and the problems of classifying images as reflecting or not reflecting intentionality/story telling, and related that to his own interests in biological systematics and evolution. How could one infer an underlying process (evolution) from a set of classifications of organisms (eg fungi) if the classifications themselves might be subject to question?
A number of interesting issues quickly arose
- To what degree have contemporary biologists actually "removed subjectivity" from the classication process (as, for example with cladistics or the use of DNA sequences as relevant characters)?
- Is there a "natural" classification for biological entities, ie one that is observer independent?
- Given the emphasis in evolutionary thinking on slow continuous change over time, are there actually any discrete biological categories?
- Is there actually "clustering" of biological organisms independent of observers? Over time? At any given time?
- Can one "acknowledge subjectivities and still not throw out the baby with the bathwater"?
- What is the point of teaching/talking if everything is dependent on a variable observer context?
The active conversation around these issues precluded Wil expanding on the relation of these issues to that of intentionality/story telling, and it was agreed to continue this discussion at a special meeting next Monday.
12 November meeting summary (Paul)
Conversation about the issue of whether classifications are actually "out there" as opposed to constructions from "in here" was advanced by Wil bringing to the table the case of color: in rainbows (or color wheels) there is continuous change from one hue to another rather than discrete reds, greens, blues etc. "Categorical perception" in this case is clearly a matter of brains (which may do it somewhat differently) rather than of discrete categories out there that exist independent of an observer.
That there are commonalities among humans in categorical perception, and indeed commonalities in some characteristics across most organisms (lateral inhibition networks) raises the possibility that the evolutionary process has found/converged on ways of revealing aspects of an external "reality." An alternate perspective, however, is that evolutionary processes settle on mechanisms that "work," and that there may well be other mechanisms, unexplored by evoution, that might work equally well. From this perspective, evolution is not evidence for a unique "reality" with particular categories; it only discriminates between categories/descriptions that have worked and others that either didn't or haven't been tried.
Against this background, conversation went on to the issue of 'intentionality/story telling," with the potential to generate counter-factuals as a key element. Trees, it was argued, contain and update "models" of their surroundings, as evidenced by spontaneous movements that prepare them for changing environmental conditions. The question that arose was whether everything we do we do the same way that trees do things. One perspective was yes, another yes with a small difference due to memory. An intriguing idea was that trees explore counterfactual possibilities on a population basis, through the process of evolutionary change, whereas humans have the potential to explore counterfactuals within individuals over individual lifetimes.
From all of this came some intriguing new questions
- Can one tell whether an organism is using "counter-factualization"? If not, is it worth thinking about?
- Could counter-factualization be modelled? Is it worth trying?
Continuing conversation, in on-line forum below
Comments
using categories to escape the limits of particular categories
Thanks all for a conversation that notably pushed my own thinking along in several directions. Some notes for myself, and anyone else interested ...
The lack of categories "out there" in the case of color I'll happily add to my own list of relevant observations/arguments for giving up the notion of a fixed external "reality." Yes, the hues of rainbows/color wheels are continuous rather than "categorical." Another relevant example, brought to mind by some recent reading in the history of structuralism and its relation to linguistics, is the speech stream. Many of the demarcations we use to decipher a spoken sound stream don't exist in the sound stream itself. And speakers of different languages will parse a given sound stream differently. Clearly "classification" can involve a fully observer dependent creation of discontinuities in a fully continuous input. The distinction commonly made between females and males is worth thinking more about in this regard. Also worth thinking more about is the notion that we needed technological/scientific advances to become aware that categories like color are observer dependent. A similar story holds for the observer dependence of space and time. Does this mean we are coming closer to "reality" as we reveal observer dependence, or that we can't escape observer dependence? I bet on the latter, for the reasons given in the 12 November meeting summary above vis a vis evolution. What we have at any given time is "less wrong" than what it replaces, but there is no reason to believe there aren't entirely different ways of making sense of things that would equally well work.
I'm very much intrigued by the increasing focus in the story telling story on counterfactualization. My intuition is that there is a lot of room for productive inquiry along these lines, and perhaps a way around some of the barriers that have impeded thinking about "consciousness." Yes, I think they are closely related ideas, and yes, I think that counter-factualization, like consciousness, is probably not something that one can establish the existence of in another person (or another entity) "from the outside" using current techniques of observations. But no, I don't think that means that counterfactualization isn't worth exploring. And no, I don't think that means that one can't model it, or look for its emergence in an emergence model. Yes, I can't do either yet, despite several years of playing with it, but I'm more than encouraged by our conversation to increase my own efforts along these lines.
Equally importantly, our conversation strengthened my intuition that there is indeed an answer to the "so what?" question, even without new observational techniques or models. Does counterfactualization make a difference? Yes, if it gets us beyond the Godel limitation of formal systems, if it allows us to get to places that we couldn't get to without it, ie places one can't reach either inductively or deductively from a fixed set of starting positions using a fixed set of rules. I think we're getting closer to a position in which one can both learn from and go beyond William James: paraphrased as "My first act of counter-factualization shall be to believe in counter-factualization." We can't do it without believing first that we can do it, but if we can do it we can perhaps show both how it is done and in what ways it is "less wrong."
Categorization == Perception
Not sure where to have the conversation here... too many places to "reply" to! In any event, thanks, Wil, for bringing up some demonstrations I hadn't seen before, especially the shaded checker board. That really drives home the point as to what the brain does so quickly and effortlessly to construct what I would call an "isomorphic" view of what it thinks is really "out there." This is so different from what a robotic vision system does that it makes my head spin. It also reminds me of the point: perception is categorization, and categorization is perception. These two processes can't be separated. That make it harder to model and implement, but together they are a tight dance resulting in the representation in my head. I'm still wrestling with understanding how this could be connected to "intentionality" and "counter-factuals"...
Pattern as a false positive
For me, classification is not the core of the issue, but is instead symptomatic of a larger issue. Regardless of the accumulation of information (I’ll keep a cautious distance from any notion of facts), story tellers attempt to reconcile their perception of reality both with and using that information (whatever its source). Thus, story tellers are classifiers/decoders (in some sense these are synonymous). If we cannot distinguish bits of information from one another, then there is no hope for understanding the whole. Some variation in information, we ignore because the decoding has no perceived effect on our working knowledge of a system (i.e., the decoding doesn’t influence our fitness). We focus on other particular variation because our decoding in those cases does affect our interaction with ‘reality‘ (i.e., our fitness is enhanced or degraded). So, given how selection (as one source of modifying stored information [genes]) provides differential consequences for our decoding, I am not surprised that we perceive what might in fact be continuous variation as quite clumpy. Across the entire range of possible continuous variation (in whatever), why should we assume that decoding the whole range should have equal consequence to our understanding (or fitness)?
Complementing this idea is another point, which I think is an interesting consequence of our decoding. Paul expressed some discomfort with the relaxation of the assumption of a ‘maker’, the notion of ‘no beginning, no end’, and the interaction between code and decoders. Oddly, this relaxation is an articulation of a long held notion of mine. The making of stories is a perception from which we’ve reconciled inconsistencies in what we perceive as real. One story line for which Westerners have particularly fixated is that time is finite, or at least it must have ‘started’. Interestingly, from an Eastern perspective, ‘no beginning, no end’ would seem very logical. Thus, different contexts (environments) have provided quite different decoding of similarly available information. So I am not surprised or distressed ‘That we are always working in "subjective" worlds that may be different in different people’. That said, statistics might be a decent metaphor for making sense of the noise problem, i.e. the goal of statistics is to discern signal from noise. This idea is similar to Wil’s ‘Only with many viewpoints do we dilute individual bias.’ Of course, given our common descent, do we really have all that many viewpoints? And thus, a sobering lesson from statistics is that any analysis with a small, autocorrelated sample size is quite susceptible to the acceptance of a hypothesis as supported when it is indeed false.
classification, decoding, and story telling
I'm glad we agree there is no reason to be either surprised or distressed "that we are always working in 'subjective worlds' that may be different in different people." And that "Only with many viewpoints do we dilute individual bias" (see The Objectivity/Subjectivity Spectrum). And that statistics (and humans, see The Odds of That) "is quite susceptible to the acceptance of a hypothesis as supported when it is indeed false." Against that background, let me push the argument a little further?
Yes, story tellers are inevitably "classifiers" but I don't think that is "synonymous" with "decoders," for reasons that relate to the issue Anne raised in an earlier session about sending an understandable message into the future. The notion of "decoding" contains an implicit assumption that there is something fixed and unique (if not "intended") that is represented in the coded form and that needs to be recovered, that there is "meaning" or "information" independent of the coder and decoder. That seems to me tantamount to slipping an only slightly disguised "reality" back into the discussion. My own guess is that neither information nor meaning (to say nothing of reality) exists independently of a decoder/encoder (see Information?: An Inquiry). Statistics can tell us whether there is a deviation from randomness but nothing about the "meaning" of that deviation. For that we need not a decoder but a story teller who classifies for particular reasons. And have to face the consequence that different story tellers with different reasons will classify differently.
"So what?", as Peter says, "does it matter?" I think it does, since it suggests that our task as inquirers is actually to create information/meaning, rather than to uncover/discover it (cf Empirical Inquiry: Limitations and Possibilities). And that's a quite different sort of task than the one we're used to/were trained to do. And it means that different stories may be different without any of them being dismissible/wrong. Along those lines, the distinction between "time is finite" and time with "no beginning, no end" is directly relevant. One can make a serious argument that "time" is a construction, a story element, rather than a fixed something out there (cf Timing the conscious and the unconscious). Hence it makes perfectly good sense that there exist multiple stories about it.
"what might be" as a replacement for "what's out there"
Having perhaps had more experience with reactions to the idea that "patterns ... [are] an artifact of our filter," I'm less surprised than Wil is by encountering resistance to it, and more interested in the reasons for that resistance - and in trying to figure out what story reflecting the relevant observations would achieve more interpersonal agreement.
For me, the relevant observations are both historical and neurobiopsychological. The history of science is one of repeated efforts to achieve a definitive description of what is actually "out there," independent of an observer, each followed by recognition that such descriptions are not actually definitive, that there are previously uconceived ways of making sense of existing observations (cf The Nature of Science: The Problem of "Unconceived Alternatives and its Significance and Science as Story Telling and Story Revising). The historical pattern suggests not asymptotic movement toward a final description but rather that each successive description is itself the progenitor of new descriptive possibilities.
The relevant neuropsychobiological observations have to do with ambiguous figures and what they reveal about human perception (cf Getting It Less Wrong, the Brain's Way and Reality: Construction, Deconstruction, and Reconstruction and Reality Reconsidered, From a Brain Perspective and From the Subjective/Personal to the Objective/Interpersonal and Back Again). What we see (hear, feel, etc) is, if we presume no special access to the mystical/eternal, reflects a combination of "infomed guessing" and "conceiving alternatives," and hence is always subject to revision based on new observations and/or the conception of new alternatives. And may always be different from what other people see.
One might argue that neither the historical record nor the specifics of how the brain works eliminate the possibility that there is actually a definitive something "out there" but simply account for our existing inability to discern it. We have a "blind spot." And that could of course be. But if so, its a limitation not of science to date nor of the brain but rather of any inductive/empirical process of making sense of things. No matter how many observations one has available, there will always be an infinite number of different ways to make sense of it. In short, any description we provide will always be a context-dependent one.
What's "really" out there? If we eliminate all the filtering, as well as all the guessing and conceiving of alternatives? My guess is noise, nothing interesting or meaningful at all (cf Chance in life and the world), that we create meaning by the act of perceiving. Whether that's so or not, the more general point is that we can't either see or talk about what is "actually" out there. All we can see or talk about is the meaning we make of it, individually and collectively. As Wittgenstein said, "what we cannot speak about we must pass over in silence."
That we are always working in "subjective" worlds that may be different in different people is a disturbing idea not only in the context of science but also in the context of human interpersonal interactions and social organization (see From ambiguity to skepticism to social organization). It seems to require a change in not only how we think about the objective of science but about how we think about the objective of interpersonal and social exchange as well. The challenge was well put in our conversation
"What is the point of teaching/talking if everything is dependent on a variable observer context?"
We all tend to presume that "teaching/talking" has as its objective reaching some shared agreement, and so is threatened if we each see things from different perspectives. An alternate perspective, consistent with the evolving systems framework (and a revised conception of science) is that "teaching/talking" is not actually a way to become more similar but rather a way to make clearer our differences so that they in turn can generate "previously unconceived alternatives."
"Objectivity" is perhaps rooted in "subjectivity" and in turn leads to enhanced "subjectivity" that creates new "objectivity" ad infinitum? And the point of "teaching/talking" and science and evolving systems generally is not to get to a particular place but rather to conceive new places? Would "patterns ... [are] an artifact" be less disturbing in the context of that story?
How to proceed?
Thank you all for the feedback so far. I was(am) surprised at the resistance to the the suggestion that all variation might be continuous and that any patterns we recognize from that variation is only an artifact of our "filter" - our innate ability to decode. Perhaps, it could be a bit of both. Perhaps much of what we discern as discrete, non-overlapping patterns in nature is due to our filter, but in addition, nature may have some characteristics of "clumpiness."
How would we know difference? Here, the best solution I know of to date is not trying to discover the truth from "nowhere" - completely unbiased, but instead attempt to suss out the truth from the "point of view from everywhere". Only with many view points do we dilute individual bias.
It is more than likely we will not be able to agree on this supposition, but for the time being we could agree to entertain the idea that at least most of reality is only an amalgam of biased/filtered constructs. Individuals come together with their own unique(weird) constructs and the consensus that emerges after debate and collaboration is our best model of "reality". If we can agree to entertain this for the moment, we can proceed to investigate the ramification of such a view of nature on intention, meaning, and aesthetics.
On the other hand, no reason to proceed if the starting assumptions are still highly suspect. All I really wanted to suggest in the first half of my presentation was that patterns (hence constructs of reality) are observer dependent. That may seem obvious, put it does have interesting ramifications - I think?
Post new comment