I have been developing a method for representing moral beliefs as networks of ideas. Various friends have also been contributing to the development of this approach. So far, we have asked individuals to name their own beliefs, given them back their lists, asked them to note which pairs of beliefs seem connected, and generated network maps of their beliefs and connections. I’ve also asked individuals to share their maps with peers and to consider making changes in response to other people’s arguments. I have mapped the ideas of multiple people as one network. Instead of using surveys, one could interview people or groups about their moral thinking on a given topic and identify the beliefs and connections implied by their speech–or use a rich text, such as a poem, to discover an implicit network map. Major moral theories also have network shapes that can be diagrammed. Virtues, for instance, are important nodes in Aristotle’s conceptual network, and he says that the virtues are all connected by way of one central concept, practical wisdom.

I do not see this network approach as a model of moral thought, an empirical theory about how we actually think, or a normative theory about how we should think. Instead, I see it as a technique of analysis that is relatively neutral with respect to models and theories, yet it does have certain substantive implications.

According to the Stanford Encyclopedia of Philosophy, there is little clarity or consensus about what defines a theory versus a model. But let me propose two analogies:

1. A Lego car is a model of a real car. It can be used to develop and test theories about the performance of actual cars. These theories prove true or false when tested in real cars; the Lego model is reliable to the degree that the theories pan out. Making Lego models is a technique that is more or less helpful for engineering. Its value depends on the context and the available alternatives. For instance, Lego is almost certainly a better material than soap for making models of cars. It is a less precise medium than 3D printing, but it is also cheaper and easier. A good theory is true; a good model is valid and reliable; and a good technique is useful.
2. In economics, I would call each equation a theory and use the word “model” to mean a whole set of equations, along with definitions and explanations of the hypothesized mathematical relationships. Working with equations is a technique. It is pretty obviously an essential technique for economics, but some have argued that it has been valued to the exclusion of other techniques, such as collecting better data, looking for natural experiments, or identifying important topics. Paul Krugman wrote recently: “It has been all too obvious that there are people with big reputations who can push equations around but don’t seem to have any sense of what the equations mean.” Like building with Legos, mathematics is a technique whose value varies with the context.

Likewise, I would propose that mapping moral networks is a technique with which one can build models and test hypotheses. It is fairly flexible and can accommodate a range of substantive views from both psychology and philosophy. But its relative value (compared to other techniques) varies depending on some assumptions about morality. I’ll compare it to two prevalent alternatives.

First, some moral philosophers construct systematic views. An example would be the sophisticated utilitarianism of Henry Sidgwick (which we could call “utilitarianism 3.0,” if Bentham’s was 1.0, was Mill’s was 2.0). Sidgwick held that there is just one ultimate moral principle: maximizing human happiness. But it generates a set of important moral rules, such as being kind and telling the truth. These precepts, in turn, imply many ordinary moral judgments, such as telling the truth to your mother.

Sidgwick’s structure was mainly philosophical, not empirical. He did not say that everyone is a utilitarian (in fact, he explicitly denied that), but that everyone’s judgments should be consistent with the results of utilitarian reasoning. There was, however, an element of empiricism is his view. He doubted that we can directly apply the utilitarian principle to real cases, which is why subsidiary rules are valuable.

Sidgwick’s structure can be diagrammed as a tree-like network, and that is somewhat illuminating. Individuals’ actual moral networks could also be mapped and compared to Sidgwick’s diagram, as a form of moral assessment. However, if Sidgwick was right, then network analysis has limited value. After all, his proposed network is quite simple, and some of the power of network modeling (e.g., detecting subtle clusters in large fields of data) would be wasted. Thus …

P1. Network techniques become more useful if we presume that real people hold many different structures of moral thought, that a theoretically driven structure like utilitarianism is not necessarily ideal, that some structures are much more complex than Sidgwick’s, and that comparing structures is illuminating.

Second, the moral psychologist Jonathan Haidt currently proposes six “moral foundations,” although his theory is subject to testing and improvement, and he is open to finding more than six foundations. One technique he uses is factor analysis. Many individuals are asked many questions about moral topics, and Haidt and colleagues look for unobserved variables (“factors”) that can explain a lot the variance in the answers. In developing statistical models that predict the actual results as a function of a few variables, they seek parsimony and fit. “Parsimony” means that fewer factors are better, but “fit” means that the unobserved variables should explain the actual survey answers without too much error.* Once the data yield statistical factors, Haidt and colleagues consider whether each one names a psychological instinct or emotion that 1) would have value for evolving homo sapiens, so that we would have developed an inborn tendency to embrace it, and 2) are found in many cultures around the world. Now bearing names like “care” and “liberty,” these factors become candidates for moral foundations.

Network analysis could represent Haidt’s model, just as it can represent Sidgwick’s very different conception. Each of Haidt’s foundations would be a central node connected to many concrete beliefs by one-way arrows. However, if Haidt is right, then network analysis is not as valuable a technique as the one he uses, factor analysis. First, network analysis is not nearly as parsimonious. A network map may show hundreds of beliefs clustered to varying degrees. Instead of generating six nameable foundations, a network map might yield fifty somewhat vaguely defined and partly overlapping clusters.

Second, the network method presumes that people’s explicit connections are meaningful. I diagram subjects’ networks using their assertions that their own beliefs are linked–for example, I link A to B when someone thinks that A gives her a reason to think B. But Haidt and colleagues argue that we do not know which beliefs are meaningfully connected. We reach conclusions because of unconscious biases and use reasons as mere rationalizations, gerrymandering our arguments to fit what we want to believe because of the underlying foundations. Sidgwick (like most philosophers) held that in morality, “as in other departments of thought, the primitive spontaneous processes of the mind are mixed with error, which is only to be removed gradually by comprehensive reflection upon the results of these processes.” But Haidt et al. believe that such reflection is basically ineffective, for only the primitive spontaneous processes of the mind really count. If that is the case, than the very items that matter most (the unobserved foundations) will be missing from a network map that is derived from people’s explicit connections. Thus …

P2 Network techniques become more useful if people have many clusters of moral ideas, if important information is lost by seeking parsimonious statistical models, and if reflection on explicit, conscious ideas and connections is valuable.

*Graham, Jesse et al. “Mapping the Moral Domain.” Journal of personality and social psychology 101.2 (2011): 366–385. PMC. Web. 1 Dec. 2014.

At any given moment, an individual holds a very large number of beliefs relevant to moral evaluation and judgment. Some of these beliefs are connected to other ones, producing a network. Not all the connections are strict logical entailments; some are resemblances, similarities, or causal generalizations.

Our moral networks may not determine—or even strongly influence—our choices and actions. We often act because of unconscious instincts, emotions, and interests. Yet it is important to have a good network of explicit ideas, for only a reflective life is fully worthy. Besides, our ideas can constrain and motivate our choices, sometimes indirectly, by way of the general rules that we devise and submit to.

Each person’s network is unique, but our ideas and connections overlap profoundly. A culture is a group of people who share many nodes and links, yet every member of a culture is at least partly different, and it is typical for several cultures to overlap and for individuals to belong to many cultures at once. There is no perspective external to all culture, no view from nowhere.

We normally do not adopt our moral networks wholesale, nor are they designed or planned. We start our lives with virtually no moral ideas and add or subtract a few nodes and connections at a time. Some philosophical and religious systems present themselves as whole designed systems or as rules for improving any given network. Such systems may provide insights by directing our attention to particular ideals and implications that are worthy of attention, but they do not and should not generate anyone’s whole network.

An ideal network would contain all the relevant and correct beliefs and connections applicable to an individual. That is what we ought to strive for. However, in the moral domain, truth is difficult to ascertain and demonstrate. What we actually do is to assess nodes and connections, one or a few at a time. When we make those judgments, grand abstract questions about the nature or basis of moral truth tend to recede, and we focus instead on whether a given choice is right or wrong given the other things we believe.

Since our personal perspectives and interests are too narrow, we reason better together. Since we think most carefully when we are deciding how to act (or when we reflect on what we have done), our best reasoning is often connected to work in the world. And in order to reason well with others, we must cultivate moral networks that have certain formal properties, e.g., they should not be too centralized nor too disconnected.

In reflecting on our own moral networks, we are obliged and entitled to consider at least three large classes of questions: Am I relating well to others? Am I promoting my own equanimity or a healthy inner life? And am I avoiding falsehood and error? These three goals map onto sangha, buddha, and dharma, or onto oikeiosis, eudaimonia, and logos as pursued by Hellenistic philosophers.

Alas, they are are not consistent with each other. For example, truth often disrupts relationships and reduces happiness. That is why reflecting on all three will generate a complex moral network, rife with tensions.

The social contexts that best allow us best to improve our networks are ones characterized by talking and listening, collaborative action and reflection, and ongoing relationships that involve the exploration of other people’s moral thought. These contexts are not necessarily small; whether by using digital tools or by reading printed texts, we can form relationships with thousands of people. But the most valuable contexts are personal and relational, characterized by responsiveness to other human beings and tangible human actions.

Relational politics cannot achieve sufficient stability and scale to distribute goods and rights fairly or to deliver security and predictability, without which liberty is impossible. Although huge agglomerations of people may form, they cannot honor the abstract virtues of large systems, such as equity, impersonality, and rule of law. Thus we also need impersonal systems: laws, states, and markets.

Of these systems, we should ask whether they are just, and if not, how they fall short. Justice has many aspects (including fair treatment of other species), but an important component is whether systems protect and help all people to live lives informed and enriched by their own complex moral network maps, which they develop in voluntary interaction with others.

But whether a system is just is not the only political question we must pose. After all, we lack the capacity to design or reconstruct whole systems. Even a king or dictator cannot usually do that. All we can literally do is take concrete steps, such as saying something to someone, seeking an office or other official role, designing and making an object, joining or forming a group, or buying or selling a good.

Using these actions to change (or maintain) an impersonal system requires some combination of replication, leverage, and leadership. Replication means encouraging a desirable practice or activity to arise over and over again, and tying the instances together as a network. Leverage means exercising power at a distance. (For example, to vote is to try to move the government from afar.) And leadership—in this context—means obtaining a position or reputation within a system that permits more than the average amount of impact.

The essential political question is not “How should a society be structured?” but rather “What forms of replication, leverage, and/or leadership should we use now to change the world?” Our strategies should still be guided and constrained by the moral networks that we develop with one another through dialog and relationships. But when we use power, we must combine sequences of concrete actions into strategies and choose ethical and effective means to relatively distant ends.