IN AUGUST 2015, Professor John H. Holland passed away in Ann Arbor, MI, where he had served on the University of Michigan faculty for more than 50 years. John, as he was known universally to his colleagues and students, leaves behind a long legacy of intellectual achievements. As a descendant of the cybernetics era, he was influenced by the work of John von Neumann, Norbert Wiener, W. Ross Ashby, and Alan Turing, all of whom viewed computation as a broad, interdisciplinary enterprise. Holland thus became an early proponent of interdisciplinary approaches to computer science and an active evangelist of what is now called computational thinking, reaching out enthusiastically to psychologists, economists, physicists, linguists, philosophers, and pretty much anyone he came in contact with. As a result, even though he received what was arguably one of the world’s first computer science Ph.D. degrees in 1959,23 his contributions are sometimes better known outside computer science than within. Holland is best known for his invention of genetic algorithms (GAs), a family of search and optimization methods inspired by biological evolution. Since their invention in the 1960s, GAs have inspired many related methods and led to the thriving field of evolutionary computation, with widespread scientific and commercial applications. Although the mechanisms and applications of GAs are well known, they were only one offshoot of Holland’s broader motivation—to develop a general theory of adaptation in complex systems. Here, we consider this larger framework, sketching the recurring themes that were central to Holland’s theory of adaptive systems: discovery and dynamics in adaptive search; internal models and prediction; exploratory modeling; and universal properties of complex adaptive systems.

Further increasing grain production and rural household income is a major challenge for rural development in China, and the current small scale of farming operations is an important factor limiting progress on both. The Chinese central government's recent approach to promoting rural development addresses the role of land rental markets in facilitating larger-scale farming operations. We use an agent-based model that represents livelihood decision-making of rural households to explore the effects of an alternative policy that subsidizes rural households that rent out their land-use rights for long terms in comparison with the current policy of subsidizing grain producers. The model is built upon our empirical analysis of social surveys and interviews in eight villages around Poyang Lake. We consider policy impacts in terms of economic performance and equality. The modeling results suggest that policy responses differ considerably between villages with poor, average, and good farmland resources, and the rental-subsidy policy is expected to be most effective at stimulating land rental markets in villages with average farmland resources. The rental subsidies are likely to move the agricultural system to a more productive state with more growth potential and less cost in most places. The rental-subsidy policy can also make every household in farmland-poor places better off and may be used to further address inequality in farmland resources. However, both policies show limited effects on increasing rural income, suggesting rural development will continue to depend on urbanization. We discuss how the government may use the rental subsidy as an instrument to facilitate urbanization to best benefit all rural households.

This chapter contains sections titled: A Simple model of ontogeny, Five mechanisms, Specification of the “spore“ model, Ontogenic sequence, Generating the model, Signal routing in the model, Summary

Digital electronic systems typically must compute precise and deterministic results, but in principle have flexibility in how they compute. Despite the potential flexibility, the overriding paradigm for more than 50 years has been based on fixed, non-adaptive integrated circuits. This one-size-fits-all approach is rapidly losing effectiveness now that technology is advancing into the nanoscale. Physical variation and uncertainty in component behavior are emerging as fundamental constraints and leading to increasingly sub-optimal fault rates, power consumption, chip costs, and lifetimes. This dissertation proposes methods of physically-adaptive computing (PAC), in which reconfigurable electronic systems sense and learn their own physical parameters and adapt with fine granularity in the field, leading to higher reliability and efficiency. We formulate the PAC problem and provide a conceptual framework built around two major themes: introspection and self-optimization. We investigate how systems can efficiently acquire useful information about their physical state and related parameters, and how systems can feasibly re-implement their designs on-the-fly using the information learned. We study the role not only of self-adaptation—where the above two tasks are performed by an adaptive system itself—but also of assisted adaptation using a remote server or peer. We introduce low-cost methods for sensing regional variations in a system, including a flexible, ultra-compact sensor that can be embedded in an application and implemented on field-programmable gate arrays (FPGAs). An array of such sensors, with only 1% total overhead, can be employed to gain useful information about circuit delays, voltage noise, and even leakage variations. We present complementary methods of regional self-optimization, such as finding a design alternative that best fits a given system region. We propose a novel approach to characterizing local, uncorrelated variations. Through in-system emulation of noise, previously hidden variations in transient fault susceptibility are uncovered. Correspondingly, we demonstrate practical methods of self-optimization, such as local re-placement, informed by the introspection data. Forms of physically-adaptive computing are strongly needed in areas such as communications infrastructure, data centers, and space systems. This dissertation contributes practical methods for improving PAC costs and benefits, and promotes a vision of resourceful, dependable digital systems at unimaginably-fine physical scales.

The Rubisco enzyme and agricultural productivityImproving Rubisco's carbondioxide-fixing capability by genetic engineering is unlikely to enhance crop productivity significantly on its own (R.J. Ellis Nature 463, 164-165; 2010).Crop yield does not depend on a single enzymatic reaction: it is the result of a population-scale process that is the outcome of a series of source and sink regulated developmental and growth processes (R. K. M. Hay and J. R. Porter The Physiology of Crop Yield Wiley-Blackwell, 2006).The success of modern, highyielding crop varieties is mainly down to their larger leaves, which intercept more light and provide more shade against competitive weeds.The management and agronomy of crops are typically more important as drivers of yield than genetics.They also make crop production more sustainable.Crop physiologists have appreciated for a long time that to increase growth, increases in photosynthetic rate are accompanied by higher respiration costs because more growth substrates and more proteins have to be constructed.The idea that respiration and photorespiration are a drain on crop dry-weight production was dropped by crop physiologists many years ago.Genomics, proteomics and metabolomics may increase our understanding of the regulation of different physiological processes and mechanisms of resistance to stress, but they do not show us the bigger picture.A recognition of the balance and interactions between genotype, environment and management is the intelligent solution to feeding the growing global population.

One argument as to why the hyperplane-defined functions (hdf's) are a good testbed for the genetic algorithm (GA) is that the hdf's are built in the same way that the GA works. In this paper we test that hypothesis in a new setting by exploring the GA on a subset of the hdf's which are dynamic---the shaky ladder hyperplane-defined functions (sl-hdf's). In doing so we gain insight into how the GA makes use of crossover during its traversal of the sl-hdf search space. We begin this paper by explaining the sl-hdf's. We then conduct a series of experiments with various crossover rates and various rates of environmental change. Our results show that the GA performs better with than without crossover in dynamic environments. Though these results have been shown on some static functions in the past, they are re-confirmed and expanded here for a new type of function (the hdf) and a new type of environment (dynamic environments). Moreover we show that crossover is even more beneficial in dynamic environments than it is in static environments. We discuss how these results can be used to develop a richer knowledge about the use of building blocks by the GA.

Complex adaptive systems (cas) – systems that involve many components that adapt or learn as they interact – are at the heart of important contemporary problems. The study of cas poses unique challenges: Some of our most powerful mathematical tools, particularly methods involving fixed points, attractors, and the like, are of limited help in understanding the development of cas. This paper suggests ways to modify research methods and tools, with an emphasis on the role of computer-based models, to increase our understanding of cas.

In recent decades, there has been a surge of interest in the origin of language across a wide range of disciplines. Emergentism provides a new perspective to integrate investigations from different areas of study. This paper discusses how the study of language acquisition can contribute to the inquiry, in particular when computer modeling is adopted as the research methodology. An agent-based model is described as an illustration, which simulates how word order in a language could have emerged at the very beginning of language origin. Two important features of emergence, heterogeneity and nonlinearity, are demonstrated in the model, and their implications for applied linguistics are discussed.

In this paper, a multi-agent computational model is used to simulate the emergence of a compositional language from a holistic signaling system through iterative interactions among heterogeneous agents. Syntax, in the form of simple word order, coevolves with the emergence of the lexicon through self-organization in individuals. We simulate an indirect meaning transference, in which the listener’s comprehension is based on the interaction of linguistic and nonlinguistic information, together with a feedback without direct meaning checking. Homonyms and synonyms emerge inevitably during the rule acquisition. Homonym avoidance is assumed to be a necessary mechanism for developing an effective communication system.

Whether simple syntax (in the form of simple word order) can emerge during the emergence of lexicon is studied from a simulation perspective; a multiagent computational model is adopted to trace a lexicon-syntax coevolution through iterative communications. Several factors that may affect this self-organizing process are discussed. An indirect meaning transference is simulated to study the effect of nonlinguistic information in listener’s comprehension. Besides the theoretical and empirical argumentations, this computational model, following the Emergentism, demonstrates an adaptation of syntax from some domain-general abilities, which provides an argumentation against the Innatism. © 2005 Wiley Periodicals, Inc. Complexity 10: 50 – 62, 2005

This dissertation describes how to improve automated design and evolution in computers using the structuring of genetic programs in biological systems. It also shows how to ‘reprogram’ cells to perform useful tasks by embedding computer-evolved code in biological organisms. This reprogramming makes it possible to exploit in new contexts the cell's ability to self-repair, replicate, and generate chemicals, light, or motion at microscopic scales. Biological systems evolve to create clever designs for solving problems, using mechanisms that do not explicitly involve knowledge or intelligence. The designs that result are often superior to the best human designs. The mechanisms involved reuse and recombine previously discovered structures—building blocks—to generate more complex designs. Such processes reduce problems that grow exponentially in difficulty with size to simpler problems with a hierarchical structure. AI search techniques like Genetic Algorithms attempt to emulate this mechanism and harness it for automated programming and problem solving and have been successfully applied in a vast number of areas as a powerful black-box design and optimization tool. However, current implementations still require significant human input in the initial problem formulation (whereas real evolution does not), which also results in different representations for each problem, making it difficult to determine what characteristics are most useful for its performance. In contrast, biological evolution uses the same structures and mechanisms (DNA) to solve problems as different as flying or optimizing metabolic reactions. I address these problems by identifying and testing key mechanisms and features responsible for the success of evolution from a computational perspective, and use them to explain previous results and build a single all-purpose implementation that, much like electronic circuits, avoids unwanted human overhead and customization by using the same sub-symbolic building blocks (gates and circuit patterns like loops, counters) for each problem. I demonstrate its biological soundness by comparing simulation results to human-designed gene circuits and by developing a step-by-step mapping and fabrication of corresponding DNA sequences for a few examples (using genetic engineering to manipulate cellular aging, chemotaxis) which I then insert and test in cells.

A reaction network arises when a set of reactants (chromosomes, chemicals, economic goods, or the like) recombine at specified rates to produce other reactants in the set. When the reactants are characterized in terms of reactive regions (schemata, active sites, building blocks), reaction networks can be modeled by classic stochastic urn models. The corresponding Markov processes are specified by matrices that, for realistic problems, are small enough to allow standard matrix operations and Monte Carlo estimates of important properties of the trajectory of the process, such as the expected time to first occurrence of some designated reactant.

This paper introduces concepts associated with complex adaptive systems (cas), linking those concepts at some points to economic planning. The paper begins (section 1) with an informal description of the notion of a cas and then (section 2) discusses the critical role of “building blocks” in understanding cas. Using these ideas, the paper goes on (section 3) to discuss the phenomenon of “emergence”, wherein the whole of the system’s behaviour goes beyond the simple sum of the behaviours of its parts. The body of the paper (sections 4 through 6) looks at the role of modelling in predicting the behaviour of cas, examining the kinds of model that will serve this purpose. The paper concludes (section 7) with a brief discussion of the relevance of these ideas to economic planning.

The process of information exchange among the population of individuals manipulated by Genetic Algorithms (GAs) involves two key components: crossover and mate selection. The central theme of this thesis concentrates on the investigation of effects of mate selection in GAs. The importance of mate selection in biology is widely recognized, yet a systematic investigation of this subject in GA research is still lacking. The goal of this thesis is to propose a framework that facilitates exploration of mate selection in GAs in order to (1) gain a deeper understanding of how GAs work, (2) how to design more robust GAs, and (3) shed more light on why mate selection matters in biology. The first four chapters of this thesis present motivations for this work, and describe investigations of the basic properties of mate selection in the context of GA. I employ the Schema Theorem and a Markov model as analytical tools to facilitate the study of mate selection. A number of empirical results are also presented to enhance our understanding of the GAs' behavior. The results based on simple test problems highlight the importance of mating choices in improving the GA's performance. Next, this study focuses on two classes of more complicated, building-block-based problems—the Royal Road functions and the hyperplane-defined functions. With the results further obtained, I introduce an important hypothesis regarding the role of mate selection in GAs. That is, if one's goal is to improve the GA's search for best-so-far solutions, then on easy problems a dissimilarity-based mate selection scheme is more beneficial. If problems present sufficient difficulty, the GA's search power can be further improved by reducing the selection pressure toward higher-fitness individuals while selecting mates. Chapter 6 presents a test of this hypothesis based on several more realistic, non-building-block-based benchmark testbeds. The test problems used are of increasing complexity in terms of various aspects of fitness landscapes. The first two testbeds, a sphere function and a step function, represent unimodal problems. The following four testbeds are multimodal in which characteristics of fitness landscapes such as deception and non-separability are included-the generalized Rosenbrock Saddle, an optimal control problem, a modified version of the Schaffer function F7 and Michalewicz's epistatic function. All the results of the experiments validate this hypothesis. This is encouraging-it implies that the ideas of mate selection proposed in this thesis can be applied to practical problems. Chapter 7 discusses a more general setting in the context of multimodal function optimization, engineering and machine learning. Identifying multiple peaks and maintaining subpopulations of the search space are two central themes. An immune system model is employed to study these two problems. The experimental results indeed shed more light on how mate selection schemes compare to traditional selection schemes. The final chapter provides a summary of this thesis, and highlights its contributions to GA research. It discusses paths for future research, and draws overall conclusions from the research presented in this thesis.

Signaling networks are exemplified by systems as diverse as biological cells, economic markets, and the Web. After a discussion of some general characteristics of signaling networks, this article explores the adaptive evolution of complexity in a simple model of a signaling network. The article closes with a discussion of broader questions concerning the evolution of signaling networks.

Building blocks are a ubiquitous feature at all levels of human understanding, from perception through science and innovation. Genetic algorithms are designed to exploit this prevalence. A new, more robust class of genetic algorithms, cohort genetic algorithms (cGA's), provides substantial advantages in exploring search spaces for building blocks while exploiting building blocks already found. To test these capabilities, a new, general class of test functions, the hyperplane-defined functions (hdf's), has been designed. Hdf's offer the means of tracing the origin of each advance in performance; at the same time hdf's are resistant to reverse engineering, so that algorithms cannot be designed to take advantage of the characteristics of particular examples.