Systems thinking, feedback, and cybernetics.
- Fritjof Capra, 1995. The Web of Life: A New Scientific Understanding of Living Systems. chapters 1-4.
- Arturo Rosenblueth, Norbert Wiener, and Julian Bigelow. 1943. “Behavior, Purpose, and Teleology.” Philosophy of Science, 10(January): 18-24. [link]
Please post comments below.
Entries (RSS)
April 11th, 2007 at 3:13 pm
So I guess I’ll end the game of chicken and start my reply.
The second reading I found funny mostly because of what it was trying to do. By “trying to do” we mean it was implementing a “behavioristic system” of science, guided towards some attempt at a definition of its purpose rather than its components. To accomplish that goal, the authors looked at the various components of “purpose”, and came up with a branching tree diagram including “first-” “second-” and “third- order” perdiction rather than trying to get at a behavioristic generalization of the kinds of purpose. That is, I would have expected the authors to come up with a holistic account of holistic science, rather than a reductionist account.
The question is, of course, what would that look like? For one, we would need to consider prediction wrought large. What characterizes it? Simply that one can know the purpose of the various behaviors in the world, and then a recursive understanding that prediction is also a kind of behavior. I would contend that beyond that there is very little difference in kind between the various orders of prediction. Any differences between me and a cat would then be a difference of performance, and not one of competence, to take the linguistic terms out of context.
Unfortunately I have to run off to class, I will edit finish this comment when I get back, but hopefully I broke the dam?
April 11th, 2007 at 4:27 pm
Back… no editing system… so here goes:
After an hour of tennis in which to collect my thoughts, I’m actually interested in looking at at least this paper (and probably the first, and probably a lot of subsequent in terms of the competence/performance distinction from linguistics.*
That is, a behaviorist program largely rejects a notion of “competence”: that there is something in us that knows how to do a thing (like say, producing language) that is distinct from our “performance”, that is what we can actually see. Or at least, the behaviorist program rejects that competence is meaningful: if it is overwhelmed by other factors, then you might as well not consider it at all.
* For the less linguistic-y inclined: competence is taken to be your knowledge of language, and performance is what you can actually do with that knowledge. For instance, the difference between
1) ???The doctor sent for the patient arrived. (Bad!)
2) The flowers sent for the patient arrived. (Good!)
is a performance problem, while things like:
1) Who did John say that I killed?
2) *Who did John wonder whether I killed?
is taken to be a competence problem. In the former case, you just screwed up. In the latter (goes the theory) there’s something that tells you that “wonder whether” won’t allowed a question word into its complement.
April 11th, 2007 at 5:34 pm
The beginning of the first reading (The Web of Life) struck me as full of sweeping unsupported claims and unconvincing rhetorical tricks.
As examples of unsupported claims, consider the following paragraph: “Stabilizing world population will be possible only when poverty is reduced worldwide. The extinction of animal and plant species on a massive scale will continue as long as the Southern Hemisphere is burdened by massive debts. Scarcity of resources and environmental degradation combine with rapidly expanding populations to lead to the breakdown of local communities and to the ethnic and tribal violence that has become the main characteristic of the post-cold war era.” Granted, the purpose of the text is not to convince us of these claims but rather to use them as examples of interconnected problems where systems thinking can help us come to a deeper appreciation of the issues involved. In that sense these claims are presented to be self-evident, perhaps preaching to a choir that I do not belong to. This kind of sloppy argumentation so early on made me skeptical of the rest of the author’s points.
As examples of unconvincing rhetorical tricks, consider the choice of terms for “deep” ecology and “shallow” ecology. Even knowing nothing about the reading, by the way the debate has been framed one can quickly surmise on which side of the deep/shallow divide the author stands. (After all, who wants to be shallow?) By labeling the opposing viewpoint as “shallow” the author seems to believe he can ignore the task of actually logically arguing AGAINST the opposing viewpoint and IN FAVOR of his own. In any case, “deep” ecology’s privileging of “life” above all other things is just as unjustified as “shallow” ecology’s privileging of “human life” above all other things. Using the author’s own tricks, one could easily speak of a “transcendental” ecology that does not distinguish between living and non-living things, and views living things as just one particular strand in the web of existence. (Note that we have not actually made a coherent argument against “deep” ecology. But doesn’t “transcendental” so so much, well, deeper, than “deep”?)
Regardless, the first chapter of the reading has very little to do with the guts of the rest of the reading. The author fancies himself immersed in a Manichean philosophical struggle of epic porportions waged between “mechanism” and “holism”, “Cartesianism” and “Romanticism”, “Physics” and “Biology”, “domination” and “influence”, “balance” and “imbalance”, “parts” and “wholes”, “hierarchy” and “system”, “analysis” and “synthesis”. In summary, nothing short of the ultimate battle between good and evil. The religious undertones are so strong that the author even introduces the prophet (Wiener) and the Judas (von Neumann) of systems thinking.
Perhaps systems thinking was so succesful in its origins that it was commonplace in my early education, but I really don’t find anything groundbreaking about the form of thinking the author advocates. It seems natural to me to think in terms of relationships between elements, between parts and wholes, to shift between levels of interpretation in order to achieve more general or more specific understandings. It seems, to me, as natural as “Cartesian mechanistic” thinking, and not in any way conflicting with it. The history and the applications of systems thinking were interesting to read, but I am far more interested in exploring the explanatory power of systems thinking. To my taste, the reading spent too much time trying to inflate its own importance and too little time actually achieving importance.
The second reading (Behavior, Purpose, and Teleology) is philosophically interesting in the attempt it makes to sidestep the problem of determinism by distinguishing between “purpose” and “effect”. However, this is merely terminological sleight of hand. “Purpose” is defined as “behavior that may be interpreted as directed to the attainment of a goal [effect]“. The trick here lies in the word “interpreted”. Interpretation is not well understood philosophically, and its use is problematic in this case. Consider, for example, a stream of “encrypted” data, for example, the text of JRR Tolkien’s The Lord of the Rings. (For those with some knowledge of cryptographic protocols, assume the text is encrypted with a one-time pad so that the message and the key are both the same length, and the encrypted text is just the XORing of the message and the key). To the clueless (keyless) observer, this stream of data will be indistinguishable from random data. The omnipresent Eve of cryptogrpahic lore will be oblivious to the journey the Fellowship takes right before her eyes. To clued-in (keyed-in) observer, however, it is clear that it is not random data, but in fact data that can be “interpreted” (decrypted) to be the text of Tolkien’s most famous work, even though it would be unrecognizable as such to the naked eye. The catch is, to another observer in possession of the WRONG key (well, more precisely, the correct wrong key), the stream of “random” data might be interpreted as CS Lewis’s The Chronicles of Narnia. To yet another observer (with, coincidentally, just the right incorrect key), it might be some subset of the Goosebumps series. Each of these observers is “interpreting” this “random” data differently, and each coming to a different conclusion as to what the “purpose” of the data is. Which interpretation, then, is the correct one? No two people who look into Galadriel’s Mirror see the same thing.
April 11th, 2007 at 5:40 pm
So, my thoughts were geared less toward the second reading; and since David hit that one up pretty strongly, I guess I’ll talk a little bit about the first instead.
I suppose I was introduced to aspects of “systems thinking” pretty early on — ideas of emergence, feedback loops, etc. — which makes it difficult for me to internalize the first article with anything but agreeable moderation. I think the article does a good job, if indirectly, of supporting this moderate approach.
First it showed that systems thinking should play an important role in the world, at least for some healthy balance. I think the fundamental problem that these thinkers reacted to was a close-mindedness in defining our (individual or society’s) goals and purposes, and ignoring other consequences that our actions (could) have on the grounds of ignorance or apathy. In this vein, the systems ideology had moments of taking on a self-righteous tone. But it brought up relevant examples: serious problems like “world hunger” which are both caused by, and unsolvable because of, afflicted tunnel-vision.
On the other hand, it showed that the reactionary approach of “then generalize! zoom-out!” is self-defeating. As we all probably suspected as the reading progressed, too much consideration makes for overly-complex, unsolvable problems. The article asserted, under it’s Ethics section, something along the lines of systems-like thinkers (or deep ecology thinkers, or related group) believe that all life is equally valuable. I disagree. But in any case, this is a bit of a misnomer if taken as a metaphor. Because not all variables in all situations are equally valuable to us depending on our goals, and this is where the pendulum starts swinging back in the other direction… to simplifications and approximations.
I am left claiming, disappointingly, that the balance must be decided on a case-by-case basis. Because so many variables hang in the balance of every question, and because each problem is so very different, I don’t think we could ever develop a rule-of-thumb for what types of approximations are okay, and where we must be more holistic in our considerations. How we personally determine this is probably reflective of what relevant variables we do and do not value.
I would imagine, and got the impression, that many of the ideas discussed in the first reading went through intense buzzword phases: “systems”, “deep ecology”, “ecofeminism”, “feedback”, … and though we are obviously focusing on systems ideas and feedback loops in this class, I also suppose that most of us will resist subscribing to any of these ideas as the end-all, be-all of worldviews. I’m very curious and excited to hear where other’s feel the line should be drawn. To what degree a holistic perspective will be appropriate and productive in the cases that we study, and how applicable “systems” ideas are to our models.
And I can’t help but wonder what feedback loops are influencing the prevalence of the performance/competence theory.
April 11th, 2007 at 5:47 pm
I will comment on the second paper for now, since I am more interested in the point the paper is trying to make. Specifically I’m intrigued by the connection between humans and machines in light of the authors’ ideas about purpose. Although the authors of the paper “stress the importance of purpose,” in my opinion, they fail to make a meaningful distinction between having a purpose and behaving purposefully, at least in the realm of machines. The reason “the view has often been expressed that all machines are purposeful” (19) is because all machines were created in order to fulfill a specific human purpose. Machines can be made for aesthetic purposes, convenience, entertainment, etc. In fact, if a machine had no purpose, it would be a useless assemblage of parts and there would be no reason to keep it around. However, the underlying function of a machine does not ever dictate its behavior. Think of the following analogy: suppose, for example, that a mother wants her ex-husband to take care of her small child for the weekend. Because of their strained relationship, though, she gives the child a set of detailed instructions as to how to effect this behavior in the child’s father. The child is to gaze adoringly into his father’s eyes, say “Daddy, can I spend time with you? I really miss you when you’re gone,” and then bite his lower lip in a way the mother has calculated will warm the father’s heart. In this case, the child’s behavior is certainly planned with a purpose in mind. Yet while the child is performing the set of actions laid out by his manipulative mother, he is not acting purposefully, or at least not with the same purpose that underlies his actions. The child probably bites his lip out of respect for his mother and a desire to obey his wishes, whereas the real purpose of such an action is to sway the father to take a specific action.
Similarly, we have not yet created machines with desire or express purpose. Machines, then (like the child sent to do his mother’s bidding) are guided by a purpose that is not their own. Consequently, their actions cannot be intentional; machine actions are merely by-products of external inputs and a machine’s intrinsic set of instructions. Therefore, we should think of a gun’s purpose not as “to fire a bullet so that it hits a certain target,” but “to fire a bullet so that it goes where the firer intended.” By this definition, every machine acts with purpose and no machine acts purposefully.
Keeping this in mind, it’s interesting to consider how this sense of purpose would change if humans could create a machine to act in a human-like way. To quote the article, “The ultimate model of a cat is, of course, another cat.” So how does this sense of purpose factor into an autonomous computer that mimics human behavior? Endowing a computer with the capacity to determine its own purpose still means that the creator of the computer has a specific end in mind. But if the best model of a human is another human, then the computer’s actions should be guided exclusively by its own desires…which is problematic if a human’s desires dictate its structure. Perhaps introducing randomness into the system solves this dilemma? In any case, I think it’s something interesting to think about.
And my personal favorite part of the article: the idea that human-like robots should be made out of organic material in order to be most realistic. J
April 11th, 2007 at 7:13 pm
The Capra reading, which gave a historical account of the shifting worldview, from a more materialistic/Cartesian to ecological/holistic perspective, helps to frame the paradigm in which the study of cybernetics emerged. This work also stresses the importance of systems thinking, not only to understand human cognition but to solve global problems as well. While I felt some of Capra’s introductory comments were unsupported, this did not distract from the purpose of his paper.
While I agree that the general lens through which we view science has changed, I think it might be too soon to say that the “building blocks” mentality of science has been abandoned. For example, in psychology, we often belief that we can glean a better understanding of brain functioning by understanding the purpose of particular parts of the brain—and this is an area of science in which systems thinking plays a large part, more generally. Nevertheless, I think that the notion that the whole is more than the sum of its parts has merit. Linguistics, for example, continues to fail to produce a unified theory of grammar, perhaps because the level of specificity they have obtained has caused them to lose sight of some overarching mental process which cannot be described by deeper detail.
(As an aside) I found this reading quite refreshing, as some SymSys majors look at me funny when I say that I am focusing on the same subject area (identify formation as related to our experiences, and how our interactions with race, ethnicity and class help to foster these experiences) through my concentration in Decision Making and Rationality in Symbolic Systems and in Comparative Studies in Race and Ethnicity. As a result, I find Capra’s discussion of the emergence of systems thinking involving not only mathematicians and engineers but also social scientists particularly striking. The relationship between computers and human cognition emerged in some sense as a result of more social, or more apparently social, studies. And Wiener recognizes this, too, as he states on page 62 of the Capra reading: “It is certainly true that the social system is an organization like the individual, that is bound together by a system of communication, and that it has a dynamics in which circular processes of a feedback nature play an important role.” Indeed, as we consider how our Symbolic Systems courses may be applied to human interaction, one can understand how people (or even computers) process things on an individual level are often modeled by group interaction. Learning processes, for example, which we may model using neural networks have feedback loops embedded in them—networks that receive positive feedback tend to repeat those same actions in a continuous cycle; indeed, humans behave in a similar fashion from the perspective of more social studies–the notion of “positive reinforcement” in teaching children how to potty train or even the processes we use to develop stereotypes have similar feedback loops attached to them.
Meanwhile, the Rosenblueth, Wiener, and Bigelow paper seeks to establish a link between feedback loops in machines and in humans, arguing that they are similar in regard to negative-feedback goal-oriented behavior, or that machines are capable of performing goal-seeking and purposeful activities. Interestingly, they appear to argue that in the case of machines goal-oriented behavior is independent of future causation. I’m not sure if I see the link here to human action; hopefully we can discuss this topic in class. Nevertheless, I admit after reading the Capra paper, I was surprised, like David states in his comment, about the format of this paper, as I expected a more holistic discussion of this topic than one so abstract as the one presented in the paper.
April 11th, 2007 at 7:50 pm
Sorry this is a few minutes late. I planned to post this afternoon but I have started to get pretty sick and so I took a nap, and ended up sleeping much longer than I expected. I’m still feeling sick so I will miss out on class today, unfortunately.
I’ll be focusing on the first paper here. In particular, I’m interested in how some of the ideas relate to Kant, and while Kant is briefly mentioned in the reading I’d like to say add a few more ways in which he is relevant to the reading.
An important distinction that Kant draws is the difference between things-in-themselves (“noumena”) and things-as-appearance (“phenomena”). Kant points out that we can never know things as they are in themselves but only as they appear to us. This is echoed in the reading in several places, but most strongly when the author writes about the connection between epistemology and science. The author quotes Heisenberg saying that we never know nature itself but only nature “exposed to our method of questioning.”
This seems like a pretty obvious point, at least a couple of centuries after Kant, but it leads to a difficult question with a not as obvious answer. The author writes that, while we may never know reality itself, our science and knowledge approximates reality. The problem is, how do we know that our science approximates reality if we can’t actually know the reality to compare it with? Another way to put this problem is, how are phenomena related to noumena, and how can we know they are related in this way?
One possible way in which phenomena are related to noumena is that noumena cause phenomena. I see my computer screen because there is some object in the real world (that I can’t know) that is causing me to perceive a computer screen. However, a problem with this kind of answer is that the object in the “real world” causing me to see a computer screen could be some sort of object like a computer screen, but the cause of the perception could be something totally unlike the perception, especially since Kant points out that we can’t really say anything at all about noumena – we don’t know if any of the qualities we use to describe our phenomena apply to phenomena. If this is the case, then it’s not even clear why we need to believe in (or have grounds to believe in) the existence of an objective world of noumena, and we end up with something like simple idealism or solipsism.
Another way to answer the question of how phenomena relate to noumena, however, is not to say that a thing-in-itself causes some separate thing-as-appearance, but rather, that the thing-as-appearance is our relationship with a thing-in-itself. Not only does this remove the problematic gap between phenomena and noumena, but it also changes what Kant (and the author of the reading) are saying from something that poses a huge problem for our ability to know reality into something that is obviously true – of course we can’t know anything apart from our relationship with it, because how else can we know something if we don’t relate to it?
This connects to another part of Kant that I think is relevant to the reading. We never know the world in itself – instead, we only know the world as a relationship to us. Kant therefore seeks to discover facts about the world as we know it by investigating our own faculties of cognition. In his investigation he discovers certain “synthetic a priori” cognitions – space, time, and categories of logic – that we perceive the world through. These cognitions are within us, but through analyzing them Kant derives laws about the world, such as the law of cause and effect. I think this brings up an interesting point that is relevant to holism but wasn’t really touched on in the reading. I think that Kant is engaging in a type of “holistic” understanding here – by only understanding a single part (the self), Kant is able to reach knowledge of the whole (the world). This may seem reductionist in that the whole is understood in terms of the part, but reductionism breaks down the whole into all of its parts, while here Kant is only understanding a single part, but by understanding its _relationships_ he is able to reach knowledge of the whole. I think this also relates to the concept of recursion in computer science; we only need to understand a part of a problem, but if the part somehow resembles the rest of the problem, we can solve the problem recursively while understanding only a part of it. This may be a point where reductionism and holism meet – we understand a part to understand the whole (reductionism), but we must also understand the part’s relationships (holism) in order to reach this knowledge.
April 18th, 2007 at 4:48 pm
Fritjof Capra’s “The web of life” is a very entertaining but flawed introduction to Systems theory. The writing style of the assigned passage is conversational; suggesting that Capra geared the book towards the uninitiated. However, targeting a book towards the general public does not excuse the lack of rigor and the wholesale rape of common sense that Capra engages in throughout the assigned passage. Norbert Wiener, James Greer, Ross Ashby, and Von Neumann, all System Theorist and Cyberneticians, went to great lengths to ensure that their work was rigorous. Capra, however, chooses the spectacular; making declarations like “Physics has now lost its role as the science providing the most fundamental description of reality” to the life sciences. In making this statement, Capra is making the claim that reality or “all that is the case” is either a living system or an intricate network geared towards the emergence and sustenance of life without describing the rationale behind such a sweeping statement.
Capra’s new way of thinking –“Systems thinking” – is all about connectedness, relationships, and context. I agree with Capra about the interconnectedness of “all that is the case”. I also agree that one cannot talk about phenomena without regards to the context or environment in which it is expressed. But if “all that is case” is “all that is the case” then it is not expressed within any context, thus cannot be examined or understood using Capra’s version of system thinking which is very context sensitive. Also one cannot talk about phenomena with reference to a given context without differentiating it from the context. Although this differentiation is arbitrary, one cannot describe phenomena even in Capra’s worldview without making it, thus giving rise to either a contradiction with Capra’s notion of wholeness or an infinite regression context in the Meta level.
The second theory is very insightful and more or less shows the evolution of the concept of negative feedback and its importance in the regulation of organisms and the nature of purposeful behavior. It is a classical cybernetics paper, abstract and very general but unlike Capra, Arturo Rosenblueth, Norbert Wiener, and Julian Bigelow were specific, to the point, and defined most of the domain specific terms they used.