A Systems Library, Vol. 7
In a nutshell, what is this book with a rather mysterious title about?
The task of an adaptive organism is to find the difference between an existing state and a desired state and then find the correlating process that will erase the difference. This is a means-end analysis. Simon’s ‘Sciences of the Artificial’ proposes a generic theory of ways in which humans conduct this means-end analysis. It is a theory of human problem solving. For Simon, problem solving is design, is tinkering with artifacts. The Sciences of the Artificial are therefore a meta-design theory.
We’ll go into somewhat more depth now. Let me assure you that this book, which presents itself rather innocuously as a popularizing excursion by one of the previous century’s most gifted polymaths, demands a lot from the reader. That friction is a result of the density, substance and organisation of the material that is collected in this book.
To start with the latter: the first edition was published in 1969 and included the three Karl Taylor Compton lectures delivered by Simon at MIT in the spring of the previous year. The book counted a mere 120 pages. The third edition, published in 1998, added a good hundred pages to the original. The reader cannot but notice that the book has grown by accretion, adding layers without substantive rewriting of the original material. And this does not make for the most clear or helpful line through the author’s argument.
Qua substance, the book by its very nature encompasses a very wide scope. If we understand cybernetics to be the science that seeks to understand the ‘adaptive brain’ (Pickering), then Simon’s ‘Sciences of the Artificial’ squarely fits into this tradition. But in doing so, it roams over a patchwork of disciplines, including organisational decision-making, economics, human cognition and artificial intelligence.
Finally, the book provides some sort of ‘summa’ of a very long career and hence is saturated with ideas. But the introductory nature of ‘Sciences’ correlates with a rather informal writing style that skirts some of the underlying subtleties. Hence, even the foundational notion of ‘the artificial’ remains ambiguous in the end. Eventually it becomes clear that human beings themselves belong to the realm of the artificial. Indeed, they are probably the most important class of artifacts given that they are able not only to create other artifacts but also to re-engineer themselves (i.e. ‘reconfigure the appreciative basis for their existence’, according to Geoffrey Vickers) to fit changing circumstances. But this statement is buried somewhere deep in the argument.
So we concur with Saras Sarasvathy, who was tutored by Simon in her PhD research on effectual entrepreneurship, when she sums up the merits of the book as follows: “Sciences of the Artificial is one of the most exciting pieces Simon has ever published. In an oeuvre of over a thousand publications, that is saying a lot. But it is also, in my considered opinion, one of the most irritating. It bursts at its seams with brilliant ideas and mouth-watering possibilities for scholarship and pedagogy, but does not develop many of these into something readers can sink their teeth into, especially in the domains of management and economics. One is left with a sense of the enormity of work to be done, but not quite sure where to begin.”
The book (Third Edition) is organised in eight chapters. However, in my mind I reorganised it in six parts to clarify the general flow of the argument. In Part I (Chapter 1), Simon introduces the scope of his project and the notion of ‘the artificial’. As I indicated, although fascinating enough, this curtain raiser leaves important questions in suspension. Simon then goes on to discuss how human beings use certain artifacts — notably markets and corporations — to solve complex allocation problems (Chapter 2, Part II). These artifacts are more intelligent devices for means-ends analysis that centralised planning authorities by virtue of their decentralized sense-making and decision-making. It is in this context that Simon introduces the pivotal notion of ‘bounded rationality’.
One of Simon’s key contributions, for which he was awarded a Nobel Prize, was to question the orthodox concept of economic rationality. Instead of an economic agent engaging in rational maximization he postulates the existence of fallible agents subject to bounded rationality. This notion expresses that human rationality is always limited by 1) the cognitive limitations of their minds, 2) the time available to make a decision, and 3) the complexity of the decision problem. As a result, we have to drop the illusion that we are in a position to choose an optimal course of action. Rather we have to find a way of assessing where a reasonable solution lies. Economic agents are not optimizers but satisficers.
In Part III (Chapter 3 and 4) the perspective shifts abruptly to human decision-making. Economic complexity is exchanged for the relative simplicity of short-range design challenges, i.e. well-defined, short-term, laboratory-style decision-making problems that reveal something of the inner workings of our own information-processing system. Based on experimental evidence, Simon hypothesizes a rather simple and crude artifact that suffers from stringent neurological limitations. Human beings’ external environment is complex, but their inner environment, the hardware, is straightforward. It consists of a system that is basically serial in its operation, that can process only a few symbols at a time and that is relatively slow to transfer information to long-term memory. Superimposed on this are sets of generic control and search-guided mechanisms, and memory-based learning and discovery mechanisms that permit the system to adapt with gradually increasing effectiveness to the particular environment in which it finds itself.
Chapter 5 and 6 (Part IV and V) extrapolate these findings to progressively more complex decision-making problems (say, mid-range and long-range design challenges). Optimization problems offer a basic structure to reason about problem solving strategies in the way they align external constraints, alternatives for action and our subjective assessment of the value of these alternatives. This leads Simon to posit three activities as core elements of human problem solving skills: the ability to conduct a heuristic search for alternatives, the ability to evaluate solutions, and the the ability to allocate resources for search.
Enter one of Simon’s key metaphors: the maze. ”Human problem solving involves nothing more than varying mixtures of trial and error and selectivity. The selectivity derives from various rules of thumb, or heuristics, that suggest which paths should be tried first and which leads are promising.” Understanding this process of ‘heuristic search’ in human affairs is at the core of Simon’s life work. Considered as such, his work is pendant to Charles Darwin’s theory of evolution of natural species.
Throughout the argument, Simon discusses how these ideas found implementation into early AI concepts and algorithms. However, this conception of AI — as representational and symbol-processing in nature — has nowadays been overriden with a distributed and neurological way of operationalizing artificial intelligence.
It was Herbert Simon’s key ambition to use these insights more generally as the basis for a design curriculum. Simon saw his design theory as a bridge to connect ‘epistemic communities’ that are usually disconnected. In essence, composers, medical professionals, engineers and managers are all doing the same thing. They are designing, i.e. they are ‘devising courses of action aimed at changing existing situations into preferred ones.’ Understanding the core underlying problem solving processes would enable these professionals to engage in meaningful conversation.
Chapter 7 offers a bridge into the argument of the final chapter (Part VI) where Simon discusses the properties of so-called ‘hierarchical systems’. These are systems that are composed of interrelated subsystems, each of them being in turn hierarchic in structure until some lowest level of elementary subsystem is reached (for instance, animals including organs including tissues including cells). This particular architecture offers significant advantages in dealing with external complexity. The existence of subsystems (or ‘intermediate stable forms’) leads these systems to evolve more rapidly and hence allows them to cumulate the benefits of learning over time. This is why these systems are present everywhere around us. Markets and organizations, for instance, are hierarchically constructed artifacts created by human beings to navigate in a parsimonious way through the maze, in never-ending search for local optima. Also the brain is a hierarchical system, both in its neurological structure and in the symbolic complexes it relies on to solve problems.
‘Sciences of the Artificial’ offers a broad vista on a fascinating body of work. I see it as providing a necessary bridge between a ‘hard’, positivist and a ‘soft’, constructivist approach to human problem solving. These are not only dispassionate theories about thinking machines. There is humanity and realism in Simon’s vision, as it is touchingly rendered by the book’s envoi at the end of Chapter 6:
“Our age is one in which people are not reluctant to express their pessimism and anxieties. It is true that humanity is faced with many problems. It always has been but perhaps not always with such keen awereness of them as we have today. We might be more optimistic if we recognized that we do not have to solve all of these problems. Our essential task — a big enough one to be sure — is simply to keep open the options for the future or perhaps even to broaden them a bit by creating new variety and new niches. Our grandchildren cannot ask more of us than that we offer to them the same chance for adventure, for the pursuit of new and interesting designs, than we have had.”
Check out my other Medium stories for more reviews of classic systems thinking books. Follow me on Goodreads to find out what other stuff I’m reading.