The world is complex, but acknowledging its complexity requires an appreciation for the many roles context plays in shaping natural phenomena. In Unsimple Truths, Sandra Mitchell argues that the long-standing scientific and philosophical deference to reductive explanations founded on simple universal laws, linear causal models, and predict-and-act strategies fails to accommodate the kinds of knowledge that many contemporary sciences are providing about the world. She advocates, instead, for a new understanding that represents the rich, variegated, interdependent fabric of many levels and kinds of explanation that are integrated with one another to ground effective prediction and action.
Mitchell draws from diverse fields including psychiatry, social insect biology, and studies of climate change to defend “integrative pluralism”—a theory of scientific practices that makes sense of how many natural and social sciences represent the multi-level, multi-component, dynamic structures they study. She explains how we must, in light of the now-acknowledged complexity and contingency of biological and social systems, revise how we conceptualize the world, how we investigate the world, and how we act in the world. Ultimately Unsimple Truths argues that the very idea of what should count as legitimate science itself should change.
{"itemData":[{"priceBreaksMAP":null,"buyingPrice":21.36,"ASIN":"0226532623","isPreorder":0},{"priceBreaksMAP":null,"buyingPrice":15.03,"ASIN":"0199798109","isPreorder":0}],"shippingId":"0226532623::oupqq4KBGVruqFTMaMh2%2FOfNaRL%2BFQX4MFLanLjIPGTjvBT6FgTgQWzM37Qwoe7%2BFRlB2tBlLQG1%2BHOn3j9E0ah%2Ft1vY3gMtV5Ugyqu1Ahz6j1B5gw8RJw%3D%3D,0199798109::4mDOBt16yLY2AfdsKB%2B%2Bzv9X%2FbRSP%2ByAF09E%2BKTTkgQNbqSjOYX68%2Bv6WIlbXG5oCglYAsCQYwdxs9U8ZQxxram5hSiqS0kw9LHTehoenaGgPcBEhs9mjQ%3D%3D","sprites":{"addToWishlist":["wl_one","wl_two","wl_three"],"addToCart":["s_addToCart","s_addBothToCart","s_add3ToCart"],"preorder":["s_preorderThis","s_preorderBoth","s_preorderAll3"]},"currenyCode":"USD","shippingDetails":{"xy":"same"},"tags":["x","y","z"],"strings":{"addToWishlist":["add to wishlist","Add both to Wish List","Add all three to Wish List"],"addToCart":["Add to Cart","Add both to Cart","Add all three to Cart"],"showDetailsDefault":"Show availability and shipping details","shippingError":"An error occurred, please try again","hideDetailsDefault":"Hide availability and shipping details","priceLabel":["Price:","Price for both:","Price for all three:"],"preorder":["Pre-order this item","Pre-order both items","Pre-order all three items"]}}
“Very stimulating. . . . [Unsimple Truths] is clean and spare and fun to read. And to argue with. What more could one ask of a philosophical treatise?”
(Michael Ruse Quarterly Review of Biology )
“Drawing on nicely handled examples from psychiatry (e.g., major depressive disorder) biology (e.g., recent genetics and genomics, drug discovery, the study of insect societies), and the policy world (e.g., climate change and economic problems), Mitchell develops and illustrates a philosophy of science suited to the complexities scientists face. The result is a compact and elegant presentation of a philosophy of science she calls “integrative pluralism,” challenging many orthodox positions in the philosophy of science.”
(Richard M. Burian, Virginia Polytechnic Institute and State University BioScience )
About the Author
Sandra Mitchell is professor in the Department of History and Philosophy of Science at the University of Pittsburgh and is the author of Biological Complexity and Integrative Pluralism.
Product Details
Hardcover: 160 pages
Publisher: University Of Chicago Press; 1 edition (November 15, 2009)
Sandra D. Mitchell is Professor in the Department of History and Philosophy of Science at the University of Pittsbrugh. She has degrees from Pitzer College, Claremont, California; The London School of Economics; and the University of Pittsburgh. More information can be found at her web page: http://www.pitt.edu/~smitchel/
I've been thinking about complexity for years, so it was a joy to discover this outstanding book. I agree with everything Sandra Mitchell says, and I felt like she was reading my mind. As a consequence, I may not have gained any major new ideas from the book, but it certainly helped in crystallizing my understanding in many areas.
Below, I've attempted a summary of the key points from the book, along with providing my own fairly detailed thoughts (not in the book) on how these ideas can be applied to the problem of cancer. My hope is that these thoughts will be both illustrative and useful in themselves. ____________________
Prevalence and Behavior of Complex Systems:
(1) Complex systems are ubiquitous, especially in biological and social domains. As described below, cancer exemplifies complexity.
(2) The behavior of complex systems tends to be multilevel and full of "messy" causal interactions. The biological complexity of cancer involves molecular, organelle, cellular, tissue, organ, and organismic levels, with lateral, upward, and downward causation within and across all of these levels. In addition, beyond biological complexity, the cancer problem also involves psychological complexity related to individual knowledge and decision making, as well as social complexity related to paradigms in cancer research and clinical oncology, institutional structures and practices, funding mechanisms, peer review processes, methods for disseminating information, drug approval processes, legal considerations, etc. Further, there are abundant upward and downward causal interactions across the biological, psychological, and social levels, thus making the complexity of the cancer problem truly intertangled, multifaceted, and encompassing....
(3) The behavior of complex systems is context dependent, in terms of the boundary/environmental conditions which can affect the system. With cancer, context dependence is especially reflected in the extensive heterogeneity of cancer outcomes (which is observed both with and without treatment), thus reinforcing the importance of accounting for context by individualizing treatment.
(4) The roles played by individual elements in complex systems can't be determined by studying those elements in isolation, or even by altering those elements individually while in the system (eg, gene knockout experiments), since the role of each element depends on the dynamic configuration and state of the rest of the system. Moreover, each element can contribute to many effects (eg, pleiotropy), and each effect can be influenced by many elements (eg, epistasis), possibly resulting in homeostasis, redundancy, and robustness. Systems can even reorganize their network topology in order to preserve functionality. The extent to which these characteristics are present limits our ability to develop models based on a modular approach. All of this applies directly to cancer:
* Tumor robustness is demonstrated by the basic fact that most cancers are resistant to most treatments, regardless of the intended targets, specificity of targets, and mechanisms of action of the treatments.
* Identification of signaling pathways has been a popular modular approach in cancer biology and treatment during the past decade, but signaling pathways are increasingly being found to be connected by cross-talk and may even sometimes topologically reorganize themselves naturally or in response to treatments. In addition, (a) hitting an intended molecular target within a signaling pathway can result in unexpected side effects because the target also contributes to normal functions (eg, the experience with antiangiogenic agents such as Avastin), and (b) hitting an intended molecular target may not be effective because pathologic function can still be maintained by a variety of other mechanisms (eg, upregulation of other pathways). These factors suggest that treatments based on targeting molecules within signaling pathways may not be effective for most cancers, especially as single agents, and clinical experience during the past decade has indeed revealed this to be the case. Using multiple targeted agents in combination could be more effective, but this approach also involves greater uncertainties because the number of potential interactions (both beneficial and adverse) increases rapidly as more agents are added. A possible strategy here is to use large mixtures of natural compounds, with use of natural compounds hopefully reducing toxicity, and with use of many such compounds hopefully overcoming robustness and thereby increasing efficacy (see the New Earth Biomed website for details).
* Moving up in physical scale, treatment approaches in which the targeted modules are cellular organelles may be more promising (eg, targeting mitochondria using DCA).
* Moving up further in physical scale, robustness could potentially be overcome by targeting entire tumor cells rather than specific molecular targets on or within cells. For example, effective immunotherapies may enable phagocytosis of entire tumor cells, metabolic treatments may cripple energy supply to glycolytic tumor cells, anti-invasive treatments may physically "trap" tumor cells and thus limit tumor malignancy, and treatments which substantially alter the tumor microenvironment may shift tumor cell phenotype towards normalcy (thus linking developmental biology with cancer treatment).
* Targeting entire tumor cells is essentially a modular approach in which the modules are cells, but there is also evidence that looking at individual tumor cells in isolation may not be adequate in many (or most) cases, since it's well established that tumor cells extensively interact with each other and with the complex tumor microenvironment. Ecological models are a promising way to model this complexity, and some research along these lines has begun within the past few years.
* Finally, looking at the cancer problem overall, the disciplinary hyperspecialization in the war on cancer so far is a modular approach which has resulted in knowledge fragmentation and thus inefficiency (including duplication of effort) and ineffectiveness. If we continue to let the cancer problem divide us, it will continue to conquer us. Instead, we need an integrated cross-disciplinary approach which collaboratively draws together knowledge and skills from many different areas.
(5) Complex systems involve deep and perhaps unavoidable uncertainties due to lack of predictability. This lack of predictability is because (a) models can provide only partial representations of the many elements and causal interactions in complex systems, (b) nonlinear dynamics can result in chaotic sensitivities to inputs and contextual boundary conditions, and (c) randomness. In the case of cancer, as noted above, the responses of patients to particular treatments and their overall outcomes vary widely, regardless of how we try to diagnostically and prognostically categorize patients and tumors, and we're therefore unable to reliably predict how individual patients will fare, which is why we resort to clinical trials and statistical approaches. Techniques such as genetic/molecular characterization of tumors, chemosensitivity and chemoresistivity testing, and development of mathematical/computational models may reduce uncertainty to some extent and thus provide meaningful benefit, but it's likely that a large degree of uncertainty and lack of predictability will remain no matter what we do.
(6) Multicausality and nonlinear sensitivity conspire with randomness to produce system behavior which displays historical contingency and path dependency. For cancer, this means that treatment decisions should be based on a patient's full history, including history prior to the diagnosis, prior treatments used, the response to those treatments, and the particular timing of these events. Since the state of a tumor and its potential to respond to various treatments depends on this full history, ignoring the patient's history amounts to discarding valuable information which could pivotally influence outcomes.
(7) As complex systems dynamically change, they can produce altogether new behaviors (crossing thresholds to produce phase changes/emergence) and can change their internal structures. This is clearly the case with cancer where, according to the somatic mutation theory of carcinogenesis, the development of malignant tumors is a multi-step process, with invasion typically being the threshold beyond which precancer crosses into becoming malignant cancer. Once malignant tumors form, another threshold is the tumor size (about 1 to 2 mm) at which neoangiogenesis is triggered to produce vascularized tumors. As tumors continue growing further, additional transitions include development of qualitatively different tumor zones (eg, necrotic cores and proliferating boundaries) and the different steps in the process of metastasis. By contrast, a favorable phase change is spontaneous regression of tumors, which is rare for malignant tumors but more common for precancers.
(8) Emergence isn't inconsistent with evolution, and the two processes may collaborate. This is clearly the case with cancer, since it's well established that tumors evolve, both naturally and in response to the selective pressures and mutagenic effects of treatments, and this evolution is coupled with the emergent behaviors noted above. For example, formation of tumor zones results in new environments which selectively favor some tumor cells over others. In addition, only a subpopulation of tumor cells have the characteristics needed to complete the metastatic process, and thus metastatic tumors are different from primary tumors, and also different from each other when metastases form in different tissues (eg, bone versus brain).
(9) Although natural selection limits diversity, complex systems tend overall towards increasing diversity in both their outcomes and means of attaining outcomes. As a perfect illustration of this tendency, it's well established that the diversity (heterogeneity) of cells within a tumor typically increases over time. An implication is that tumors are typically easier to treat when detected early.
Modeling Complex Systems:
(1) Simple models aren't adequate for representation of complex systems, and reductionistic models tend to be relatively simple, even when many elements are involved, since the interactions between elements in these models tend to be relatively simple. This is exactly the situation with cancer, where oversimplified models have been a hallmark of cancer research and clinical oncology, resulting in both lack of a genuine biological theory of cancer and lack of clinical progress. It's true that developments in molecular biology and related technologies, especially during the past decade, have resulted in an explosion of data, but the models used to try to make sense of this data have been oversimplified and overly reductionistic largely because their representation of the interactions between elements has usually been very incomplete (eg, oversimplified signaling pathway models).
(2) Since "perfect" models for complex systems aren't attainable, models need to be developed on a pragmatic basis, tailored to particular needs and abilities and our evolving knowledge. Moreover, use of multiple complementary models is often appropriate, resulting in an approach Mitchell calls "integrative pluralism." In the case of cancer, many researchers have a naďve view that science is about discovering a single objective truth, and so the implicit aim of cancer research is nothing less than developing a pristine "theory of everything" analogous to what theoretical physicists seek. Such a goal is arguably unattainable for cancer and therefore counterproductive. Instead, as with engineering, the goal in cancer research and treatment should be to develop an evolving spectrum of models which are pragmatically tailored to meet the needs of particular practical situations. The complexity of such models should be no more and no less than necessary to meet these pragmatic needs. Moreover, use of multiple models for a given situation can provide valuable cross-validation of models.
(3) Since complex systems can structurally and qualitatively change as they evolve, development of different models representing different stages of a system's history may be warranted. In the case of cancer, as noted above, there are clearly qualitatively distinct stages in a tumor's history. Therefore, we should expect that distinct models for each stage may be necessary. Alternatively, we can develop models (such as computational models) which represent structures which change over the course of time.
(4) Computational capabilities expand the range of models which can be developed and analyzed, as well as the amount of data which can be handled, thus enabling "in silico" experimentation to complement physical experimentation. In the case of cancer, multiscale mathematical modeling of tumors, and associated computer simulation, is an active and expanding area of research which has already proven to be of value, and much more work can and should be done in this area, particularly with regard to increasing the interaction between such modeling/simulation and physical experimentation, as well as development of models which are of direct clinical benefit rather than mainly offering biological insight. In this regard, many models developed so far provide "pretty" spatiotemporal simulations of tumor behavior, but we'll derive much more benefit from models for which the inputs are parameters which can be manipulated for treatment purposes and the outputs are clinically useful parameters such as tumor growth rate, tumor invasiveness, and patient survival benefit.
(5) In developing models, it should be noted that some patterns are only visible when certain details are ignored, so more detail isn't always better. In the case of cancer, while modern computational power offers us the opportunity to develop elaborate models, such models are typically time consuming and expensive to develop, require large amounts of input data (which may be difficult or impossible to obtain), are difficult to understand intuitively, and are prone to human error in modeling decisions, software coding, data input, etc. Therefore, in many cases, simpler (but not oversimplified) models are likely to be better.
Policies for Dealing with Complex Systems:
Due to deep uncertainty and associated lack of predictability, policies for dealing with complex systems need to be adaptive, rather than once and for all. Often, even probabilities can't be estimated, which means that multiple possible scenarios need to be accounted for instead, with the aim being to achieve adequate outcomes across most or all scenarios, and with models and decisions being iteratively adapted as particular sequences of events unfold. These considerations apply to research and development funding policy as well, with funding being directed to multiple areas and types of investigations (a form of hedging), and with funding priorities being dynamically adapted as various approaches are found to be more or less promising or successful. All of these insights apply directly to the cancer problem.
First, at the individual patient level, due to patient heterogeneity and other factors noted above, we can't reliably predict whether a given cancer treatment will be effective for a given patient. If a treatment has previously been tested with a large group of patients, a probability of effectiveness can be estimated based on statistics, but the validity of the estimate depends on how similar the patient is to the tested group, which is often difficult to judge. If the treatment hasn't previously been tested with a large number of patients (eg, a new combination treatment or a new dosing schedule), there is essentially no basis to estimate the probability of effectiveness other than judgment, which can be highly unreliable in such situations. Reasonable strategies to deal with this predicament are (a) simultaneously try multiple treatments, hoping that at least one of them will be effective, and that they won't have adverse interactions which reduce effectiveness or excessively increase toxicity and (b) try a treatment, closely monitor the results, and then promptly switch to a new treatment if the results are inadequate, repeating this process as needed, with the choice of new treatments accounting for knowledge gained from the full history of prior treatments and outcomes (eg, if multiple treatments in a given class have already failed, try a treatment from a different class). In addition, the experience gained from trying treatments, in terms of both successes and failures, should be used to revise and improve cancer models via a continuous feedback process.
More broadly, at a research funding level, we need to first face the fact that consistently effective treatments have been found for only a few relatively rare cancers, and only a minority of patients with common advanced cancers respond well to existing treatments, so our overall progress in treating advanced cancers during the past several decades has been very limited and overall mortality rates have remained high. At the same time, during the past decade, some promising new treatment approaches have emerged (eg, immunotherapies, metabolic treatments, etc.), some of which have already shown meaningful clinical benefit, so there is at least some basis for hope. How to go forward? Since none of us can foresee with certainty which treatment approaches (if any) will consistently provide long-term survival or cures for large classes of advanced cancers, research funding should be diversely, dynamically, and adaptively allocated as follows: (a) develop a framework, probably web-based, to foster collaboration and integrate findings from different areas of research, (b) develop and regularly update websites, documents, and databases which summarize and synthesize the entire body of scientific and clinical knowledge related to cancer in a way that reveals the big picture while also providing an appropriate level of detail (ie, work "top down"), (c) fund both theoretical and empirical investigations, so that a meaningful theoretical oncology is developed which brings order to the large body of data generated by past and future empirical studies and thereby enables extraction of practical insights, (d) extend empirical investigations to include practical mathematical and in-silico computer simulation modeling, (e) balance funding across prevention, early detection/treatment, and control of advanced cancers, based on anticipated clinical impact and cost-effectiveness, (f) fund a wide variety of innovative approaches to treatment (including CAM approaches), along with continuing funding of past approaches which have already demonstrated at least minimal benefit or clear promise, (g) develop methodologies and software to help individualize treatments, and (h) provide incentives for achieving clinical results rather than merely advancing scientific knowledge, while removing the barrier of needing a large profit potential in order to justify the cost of developing and testing treatments.Read more ›
This book is aimed at professors in Philosophy of Science and students with college level exposure to that field. It is a succinct presentation of the author's view -integrative pluralism- and its lessons for public policy.
As an articulate, thorough, and brief (119 pages of text) presentation of that position, the book is very helpful to anyone in the field. It will be particularly useful reading for graduate students and advanced undergraduates coming to grips with the major currents of thought there.
The book makes a case for integrative pluralism and then looks at the ways in which adopting the position would affect our views of scientific laws, the methods of natural science, public policy reasoning, and the range of scientific explanations.
A broad vision of science that sees its explanations and methods as being of a single kind, found paradigmatically in the most unifying theories and methods of physics, is inculcated into many of us. Integrative pluralism rejects this broad vision, and in particular rejects the view that a microscopic, physics-level, scale is in any way more privileged or illuminating than the explanations and methods of other sciences. Rather, science presents a motley of many different forms of interaction among different entities at higher and lower levels in both its methods, theories, and explanations. Sometimes the behavior of an organic molecule is explained by a history of selection of multicellular organisms.
Mitchell gives a variety of reasons why the complexity of phenomena that we study in natural science should lead us to adopt integrative pluralism. Any representation of the physical world is partial, idealized, and abstract (13, 23, 33).... Although every object studied by any natural science is physical (24), not all genuine and illuminating explanations can be couched in physical language (71). Attempts to reduce in even simple cases do not work (25, 26). Many phenomena have chaotic models which prevent practical predictability (40).There are complex feedback loops between higher and lower-level entities that undermine modeling at the lower level only (109, 42). Emergent features are not mere sums of physics-level entities because they are causally effective, unpredictable, and give rise to novel effects (26).
The upshot is that scientific laws are rife with exceptions, and are best treated only as ceteris paribus (chapter 3). Philosophy should give up the search for a single scientific method along the broad lines that Mill sought, and we should abandon the idea that separate mechanism must be separably disruptable (chapter 4). Politics should abandon the idea that scientific conclusions are secure, and that debate provides reasons for inaction (chapter 5).
Mitchell presents her case convincingly. I would have liked more detail on certain points, for example the argument that descriptions are always partial (33), and the case against Kim's view, which Mitchell characterizes as reductionist. It would also be nice to know more about exactly how causation can operate from higher to lower levels of analysis if reductionism is to be avoided. The difficulty is that more detail here would have detracted from the main virtue of the book, the fact that it is such a succinct and articulate statement of Mitchell's position, and the consequences that it has.Read more ›
The subject of the book is fascinating, worthy of the deepest and most careful analysis, which makes it a terrible shame that the book itself is so poorly written. Mitchell never uses one word where seventy-five will do, repeats herself endlessly, and rarely uses a concrete example to clarify her endless churn of abstractions. It's a well-intentioned book, I'm sure, but its only effect is to anger and frustrate the reader.
