Wolfram Cloud Document

Can we use Cellular Automata (CA) to develop algorithmic models that illuminate non-algorithmic processes of dialogical reasoning in natural languages - processes like negotiation, mediation, group problem-solving, conflict transformation, and peacemaking? This preliminary study uses computer experimentation, theoretical analysis, and philosophical reflection to explore this question. It considers and provisionally assesses three strategies: 1.) Single Rule CA (of some practical use for etic understanding of dialogue); 2.) Multi-Rule Interactive CA (capable, in principle, of modelling teleological algorithms, perhaps most relevant for study of institutions with formal Semantic Grammars); and 3.) Hybrid AI integrating Wolfram-like symbolic reasoners and LLM natural language based agents (promising for collaborative wisdom, needed to correct for misuse of “smart systems”, perhaps helpful for AI alignment challenges).

Introduction

The Challenge and the Project

In at least one sense, dialogue amongst humans provides one of the richest and most computationally complex phenomena we could study using CA. In some sense, this activity presupposes contexts of the physics, chemistry, biology, history, and ongoing cultural innovation that makes it possible for you and I to be here now, writing and reading about these things and trying to reach shared understandings about them. If it makes sense to talk about reality in terms of some ultimate Ruliology, surely human dialogue will occupy a place in it that is of significant theoretical interest -- not to mention the practical relevance to wrestling with every conflict, puzzle, and adventure that we share with others. To what extent can CA methods illuminate dialogue?

One challenge arises because it seems that we humans are each, ourselves, computationally irreducible creatures composed of computationally irreducible phenomena and we are engaged in interacting with others who are similarly complex. How might CA rules that are intelligible for us find a perch and role to act in illuminating ways on such a complex stage in the theatre of human affairs? And how might we understand their scripts?

Monological Inference vs. Dialogical Reasoning as a key to the puzzle

In the way we encounter and practice them, CA and dialogical processes involve two related but quite distinct forms of reasoning. It is essential to appreciate the differences in order to understand the challenges in using CA models to understand dialogical processes.

Monological Reasoning (MR) was first formalized with Aristotelian Categorical Syllogisms. It articulates a process of linear inference that starts with a set of premises, then uses an algorithmic rule, to generate a conclusion (Delong 2004). The classic example is the Greek language version using this algorithm: If “All A are B” and “C is A”, then “C is B”. and applies it in this syllogism Step A Input (premises): All men (A) are mortal (B). and Socrates ( C) is a man (A). Step B run the algorithm . . . Step C Output (conclusion): Socrates ( C) is mortal (B).

This conclusion does not surprise us or seem to add much to the corpus of human knowledge. But that is, in one respect, the point. In the variety of deductive syllogisms this illustrates, the truth is said to be “contained” in the premises already. However, when similar but more complicated kinds of inference using algorithms of formal logic get deployed by Euclid, Newton, Lovelace, Turing and others, lots of unforeseen and important results can be obtained.

A key point to note is that any one person who knows the premises and can apply the rule can perform this reasoning. As Turing (1950) demonstrated, so can an individual machine. In that sense, linear inference of this sort is “monological”.

In contrast, Dialogical Reasoning requires two or more points of view and two or more people (unless you have the knack of splitting your personality a bit and having a genuine conversation with two points of view all on your own ;-) .

Such Dialogical Reasoning is exemplified by the kind of everyday situation in which the two of them don’t share the same assumptions, don’t mean the same things by their words, and don’t even literally see the world in the same way. Walking into the woods, one may see stands of very valuable board feet of lumber. The other may see very important niches of bird habitat. The process of Dialogical Reasoning (DR) is one in which they talk back and forth trying to work tentative shared meanings for terms and assumptions they can use as common ground in order to arrive at genuine, voluntary agreements that they share and to which they can commit themselves (Cox 2023).

Here is a schematic picture of the three step process:

The sense in which this is a dia-logical process should be clear. While actors performing “dialogue” on stage may have access to each other’s script ahead of time, people in life do not. They are encountering those who are Other -- with other values, definitions, assumptions, projects, practices, goals, et cetera. The reasoning process is a collaboration. It is not something that just one or the other does. They do it together. In that sense, there is a plural subject who is the agent of the process. When we have a successful dialogue, the end agreement is something we adopt together.

There is, however, a crucial point to note about the kind of contrast MR and DR present. It is not one of opposition. The folks in conversation may each, in turn, use monological forms of inferential reasoning in their own parts of the discourse. For example, one may offer linear inferences to calculate their conclusions about the costs of building solar panels on a school roof. Another may respond by proposing assumptions about the environmental impact of manufacturing solar panels to argue that the real social costs of the panels is much higher. But in Dialogical Reasoning, these inferences are framed by the larger process of figuring out what definitions of terms like “real costs” can be shared, what assumptions can be formulated and shared as common ground, and how best to literally see the world -- e. g. are roof top panels like monstrous industrial eyeballs glaring at us or shining portals of a path into a resilient future? The relationship of dialogical and monological forms of reasoning is not one of opposites but of two practices in which one includes the other and can refer to it and revise it.

For readers familiar with the object language/metalanguage distinction that arose out of Bertrand Russell’s wrestling with the paradoxes of self-reference, that can provide a fruitful way to further explore an understanding of the distinction (Delong 2004) . The basic distinction provides a useful way of talking about two ways of relating to a language and engaging with it. In the case of the object language approach, the semantics and syntax are simply taken as a kind of given with a fixed structure which can be used to express and infer things. A formalized expression of Peano arithmetic provides a classic example. In contrast, in the meta-linguistic approach, we adopt a stance in which we regress into a context in which we are not limited to merely talking with the language but allow ourselves to talk about it as well. In that context, we can, for example, examine its self-consistency, completeness and other features as a logical system (Nagel 2008, Turing 1937). Whether it is a formal system of mathematics or a natural language like English, when we stop merely talking with it and regress back into the meta-linguistic context to talk about it, we can then also look at its usefulness in representing truths about the world as well as expressing our preferences, interests, goals, visions for the future, and other things which may be relevant to formulating agreements and voluntarily, collectively, committing ourselves to them.

Dialogical Reasoning requires this move into the meta-linguistic context precisely because the definitions, assumptions, interests, and visions of the world that the participants encounter in each other are not part of a single fixed system on which they agree. To arrive at agreements, they must develop a common language. That process of developing shared language is one in which the meanings of terms shift and the agreed upon meanings emerge in the process of working out the common language. Part of what makes such emergence possible is the fact that the participants are embodied in the world and embedded in social contexts that make them each portals for expanding the horizons of the scope and the depth of the context of information and knowledge available to them. For example, if they differ in their conceptions of “real costs”, they can consult others to see what kinds of independent, external standards for use of such a phrase might be available and could offer an independent basis for formulating an agreement about how to use the phrase. If they differ in their views about facts such as how toxic the production of solar panels is or how glaring such things are in broad daylight, then they can go out into the world to seek more data. As embodied and socially embedded agents, participants in dialogue are able to expand such inquiry in iterative ways, extending and, if necessary, iteratively further extending the horizons of the social contexts and physical realities they examine and appeal to in trying to formulate emergent meanings and commit to agreements.

This embodied, socially embedded feature of agents in Dialogical Reasoning has an important implication for understanding the form of intelligence they exercise: In an important sense, it is not substrate independent.

This may seem to be in stark contrast with forms of machine intelligence. Data and software can both be downloaded from one machine and uploaded onto another and even moved from one medium like punch cards, magnetic tape, or compact disks to som other, like silicon chips. In that respect it seems like the information and the computer intelligence that manipulates it both exist independently from any particular substrate they are resident upon at any particular point in time (Tegmark 2018) . In contrast, the information that humans draw on in negotiations is always already connected to the world in ways that insert and embed it in contexts that give it meaning. These connections also provide it with groundings in the world that offer tacit knowledge that can, when called on, be made more explicit and clear. Take, for instance, the information conveyed by two partners who reach an agreement to marry. When, at the wedding, they finally pronounce the phrase “I do”, the meaning and full significance of those words is quite deeply interconnected with the social and physical details of who is saying them, when, where, why, and how. This is, in part, because meaning of texts in natural languages depend on their con-text and all of its tacit assumptions, values, institutional structures, and physical realities. It is also, in part, because like other key texts that formulate and affirm agreements, an “I do” in such a case involves a performative element in which particular people are committing themselves to something more than descriptions of facts about the world. They are promising to act in ways that will ensure future events happen in the ways agreed to -- and the promises, if meaningful, put their resources -- both physical and social -- on the line. In Dialogical Reasoning, meaning and intelligence are not substrate independent.

A further essential point for our project here concerns the sense in which the central and distinctive strategies of Dialogical Reasoning are different from the algorithms of traditional formal logic. Examples of such strategies are provided in negotiation handbooks like Getting To Yes. One, for instance, is especially useful when parties seem at an impasse, each wanting an option incompatible with the other’s. In such dilemmas, it helps to look for a third alternative, one that can satisfy both. The more general formulation of the strategy is: Multiply options! It is great advice. But it does not generate the options for you. It takes the grammatical form of an open ended imperative. In that way it differs fundamentally from the algorithms of Categorical Syllogism, Peano arithmetic, or the Predicate Calculus. It does not generate a conclusion from accepted premises. Instead, it guides a process of seeking a solution, a novel formulation that can become the basis of an agreement. This open-ended, forward-looking, guidance function is characteristic of other key strategies that have been discovered in studies of dialogical reasoning.

Consider a second example of such a strategy that is offered in Getting to Yes. It is especially useful when people seem set in their thinking and unwilling to give up on their position. They may say something like “It’s X or nothing. That’s my offer. Take it or leave it!” In such a case, a good strategy is this: Focus on their underlying interests! The idea is to look for the underlying reasons that motivate their commitment to X -- so that you can then search, creatively, for other options that might meet their needs and desires as well as other folks’.

Different traditions of dialogical practice often frame such strategies in different but comparable ways. For instance, Beyond Majority Rule: Voteless Decisionmaking in the Religious Society of Friends describes Quaker practices of seeking unity which, in the case of a dilemma, might lead someone to say: “Think of it as an opportunity dear!” In the case of the person committed to a position, she might say: “Try holding them in the Light!” (Sheeran 1983)

A third example of a common central strategy in practices of Dialogical Reasoning gets formulated in Getting to Yes in the imperative: Seek objective criteria! The idea here is that often there are standards, norms, community precedents, and other kinds of values which are independent of the subjective will of the parties in conflict. These can provide criteria which are, in that sense, objective. They can be investigated, their relevance can be assessed, and their implications can be drawn in trying to determine what might be a fair or otherwise agreeable outcome to all concerned. For instance, if I hit your car with mine and totalled it, I might subjectively judge that the value of your old heap at $5,000. You might, in contrast, subjectively assess your beloved old reliable wheels as up around $13,000. If we just rely on our subjective assessments, we could go back and forth in a kind of “positional bargaining” and one of us might be the hard bargainer who insists on her position and the other might be the soft bargainer, giving ground. We might each inch towards the other until some compromised was reached -- often after a long, frustrating, hostile, and tedious process. Alternatively, we might look for independent criteria that are more objective such as: the Blue Book value of the car; the original manufacturer’s suggested price minus depreciation; or the price being advertized for comparable cars in the local newspaper or Facebook Marketplace.

This third strategy is also formulated in comparable but distinctive terms in different traditions. In Anglo-Saxon law, for instance, people appeal to the independence and fairness of due process and binding precedent. In negotiations that form part of the practice of satyagraha, Gandhians speak of “experiments with truth”. They use experimental methods of self-sacrifice and non-violent resistance to discern, demonstrate, and defend claims that they appeal to as moral truths that always have to be expressed and negotiated in the terms of the cultures in conflict but which are rooted independent, underlying realities like the dignity merited by all humans, the value of respectful dialogue and uncoerced agreement, and the value of sustainable forms of coexistence and resilient ecological communities.

The Gandhian epistemic theory of the objectivity of such truths captures some defining features of this third strategy by explicitly endorsing an emergentist view of it as well as a teleological understanding of the the reasoning strategy at work in such negotiations. The Gandhian practice is an experimental search for independent standards that the parties can voluntarily agree to without any overt or implicit threat of violence or coercion (Bondurant 1988). The process is one which may begin with only dimly understood values expressed in vague terms whose meaning only gets worked out in the ever fallible but potentially successfull processes of negotiation with others and the experimental attempts to discern, demonstrate, and defend claims to emergent, objective moral truths.

During the last sixty years there has been an enormous amount of study of differing traditions of negotiation, mediation, reconciliation, community based problem solving, traditions of care, indigenous practices of collaborative decisionmaking, communal discernment, conflict transformation, peacemaking, and other forms of dialogical reasoning. They provide accounts of learnable, teachable, successful practices of dialogical reasoning that are very useful in a wide variety of practical contexts. Here is a sampling of them:

How can we best characterize the open-ended, forward looking character of the central and distinctive strategies discovered in such studies of dialogical reasoning? One useful way is to make a distinction between mechanical vs. teleological algorithms. “Mechanical” here is intended to describe the machine-like character of the algorithms of formal logic. They can operate blindly, moving us from premises to conclusions in processes of linear inference that, as Turing demonstrated, can be automated in machines whose earlier states determine their later states according to rules that specify their output based on their given input. This contrasts with the way in which the guiding rules of Dialogical Reasoning are strategies that focus on goals or destinations that are aimed at -- and ways to move towards them. They counsel us to focus on people’s interests and multiply the options we consider in order to find options that everyone can agree on. The purposive, goal seeking character of these strategies is central to their function. They move us towards an end or “telos”. In that sense, they are teleological rather than mechanical.

In one sense, we might say that this means that they are not in fact algorithms at all. There is, for instance, a very specific sense of “algorithm” that Turing appealed to in developing the notion of computability which became the basis of his design for a Universal Computing Machine. It is used to name rules that can be formulated in such a way that they give determinate results when applied in a rigorous, automatic way (Turing 1937). The teleological strategies of Dialogical Reasoning are not algorithms in that sense. However, this does not mean that we cannot create machines that apply strategies of Dialogical Reasoning. It just means that they have to be created in distinctive ways. The methods provided by neural nets, generative AI, and other forms of evolutionary programming provide one approach to doing so. They involve feedback loops and homeostatic processes which allow machines to engage in behavior that can be trained to adopt teleological strategies. Various forms of reinforcement and successive approximation enable the evolution of strategies in ways that parallel the evolution of goal directed behavior in biological organisms.

One particularly powerful technological innovation in this regard was provided by the Attention function which has enabled LLMs to engage in pattern recognition for varying sized structures of text (Vaswani 2017). Since those patterns take a linear form, when the LLM uses them to generate text, it, in effect, is directing its behavior towards the final parts of the pattern structure it has predicted as the most likely response to its current prompt. Asked, for instance, to write an op-ed on how to resolve some political conflict, it generates successive tokens that move towards a final paragraph which provides a summary conclusion. Using distinctions from Aristotle, we can explain how the Attention function enables LLMs to move beyond the merely mechanical or “efficient” causality of “Autocompletion on Steroids”. It introduces a kind of “formal causality” through pattern recognition that, when applied generatively in creating temporal sequences of tokens, results in a “final” or teleological causality (Cox 2024).

The colorful range of books studying Dialogical Reasoning has begun to reveal a rich, fascinating, and extremely useful range of teleological algorithms which can successfully guide negotiation, mediation, group problem-solving, conflict transformation and other practices for moving from incommensurable differences in assumptions towards genuine, voluntary agreement. In so far as these are teachable practices that can be described in texts, their practice can be simulated by LLMs. Their teleological character in no way precludes this, though it does call for recognition and focused study. They are a distinctive body of algorithms that differ from those of traditional logic and merit study out of both theoretical and practical interest.

It is unfortunate that, till recently, most AI researchers have not seemed to have drawn on this body of research in thinking about how to make machines more intelligent. One possible explanation is that AI researchers tend to be trained in STEM fields which normally take symbolic logic and various forms of Monological Reasoning as the defining paradigm rationality. They assume that is just what reasoning is: linear inference from premises to conclusions.

It is worth noting, however, two important ways in which this situation is changing. Many AI researchers are increasingly engaging with social and environmental problems in the real world and through collaborations with activists in these areas they are learning and seeking to apply dialogical methods in the socio-technical systems they are helping build and improve. Secondly, in interactions with LLMs, researchers are increasingly finding themselves pushed to adopt hermeneutical strategies for understanding the systems -- the kinds of methods that ethnographers use, for instance. And they find themselves speaking not of “algorithms” that produce the AI results but open ended “strategies” the AIs are using to perform functions that are informed by values and guided by goals. The LLMs run on algorithmic machines but the computationally irreducible output they provide seems increasingly to resist explanation by linear inferences based on input data. Instead it is may be most helpfully interpreted in conversational terms using methods of Dialogical Reasoning. Instead of prompt engineering that manipulates a machine, the paradigm we are shifting to is one in which humans engage in iterative processes of dialogue with machines.

One approach to exploring ways CA and other computational forms of Monological Reasoning might illuminate practices of dialogue and their teleological algorithms is to attempt to reverse engineer them. This sort of approach could build directly on the studies mentioned above by starting with the detailed examination of the nuances of those practices themselves using the hermeneutic methods of ethnography, philosophy, history, and other forms of cultural analysis. An alternative approach is to begin, constructively, with CAs and other computational forms of modelling and and see how far we can get towards modelling DR if we push that approach towards its limits. That is the approach that we will explore here.

Three strategies we will look at

So how might CAs which are paradigm examples of Monological Reasoning provide insight, understanding, or explanation for processes of Dialogical Reasoning? To consider this in a preliminary way, let’s look at three general approaches to the research:

Single rule CA operating from a single set of initial conditions

Multi-rule CA that may interact in the production of successive cell states and in the creation and introduction of new rules in portions of the trajectory of interactions

Hybrid AI that integrate Wolfram-like symbolic reasoners and LLM natural language based agents for interpretation and generation of emergent meanings & creative syntheses for agreements.

The Single Rule Strategy

How might we define ways of providing interpretations of CA generated by a single rule so as to usefully model human dialogical reasoning?

One approach would be to interpret:

cell columns as representing individuals (e. g. voters in a community)

cell states as opinions held by individuals (e. g. colors might represent support for a referendum or a political representative)

Moving beyond the first 256 rules, allows us to model how adding more positions or opinions and examining their interaction over time might have parallels in dialogues in politics and elsewhere. So, for example, the first run illustrated below of Rule #264545522443975390936887317296931828544 could be analogous to a community that begins with a very strong consensus for plain Red Hot Dogs with a wavering over adding mustard on the buns running down the middle and then a variety of green chile and blue salsa advocates moving in from the margins progressively until, after almost 1000 steps of dialogue, any trace of consensus disappears.

As we start to explore variations in this way, it might be interesting to try and find rules that either generate, sustain, of shift consensuses of this sort in interesting ways that might parallel some processes of dialogue amongst people. This might be prove interesting in exploring the enormous rule space opened up by multiple K values and neighborhood sizes.

Of course, the interpretation of these would depend very much on our conception of what counts as consensus. So, for instance, the rule pictured above yields a variation with just a slight change in its initial conditions, as we see in the second run.

A paradigm instance of this involves the kinds of rules Thomas Schelling looked at in his Nobel Prize winning work on discrimination. Consider, for instance, neighborhoods in which whites and blacks live and they have uniform preferences for living in integrated neighborhoods but different preferences as to how integrated they prefer their neighborhood to be. Suppose, for instance, that Blacks are happy to live in places where they are a minority of the population and surrounded by Whites on both sides, but Whites in contrast, are only comfortable if they form a majority and always have at least one of their two neighbors on the street who are also White. CA Rule #236, as represented in the picture below, could be interpreted as a model of this situation. The run pictured here starts with a random spread of families which are fifty percent Black and fifty percent White. However, as the Whites leave the areas where they are not in the majority, suddenly the neighborhood divides into segregated patches, providing a kind of simulation of White Flight.

We could imagine an analogous CA model casting light on how political opinions and voting behavior have been polarizing in neighborhoods and regions of the US as a result of peer pressure and emergent community identities that may go through transitions. In general, when there are fixed preferences or definable actions and ways in which they may interact with each other, CA models may cast light.

A further way to understand how Single Rule CA may be useful -- or limited -- in studying Dialogical Reasoning can be formulated using the distinction from ethnography between etic and emic methods of study (Mostowlansky 2020). Etic methods look at human behavior from a point of view framed outside the linguistic community of the people studied. They typically invent descriptive terms that are given operational definitions which allow the investigators to have easily quantified bodies of observational data. They then formulate hypothesis using theoretical terms they introduce in order to explain the behavior. In studying the ways in which local community leaders are elected around the world, a comparative political scientist might use the same vocabulary to describe people in rural Mexico, China, and Kenya, and formulate hypotheses in English using ideas of rational action from neo-Classical economics. In contrast, emic methods of ethnography start with the language of the people studied both to describe and explain their behavior. They may translate these into English for a broader audience but in doing so, they can not use simple word for word substitutions of English for Spanish. The native language of the people studied will have a variety of functions and meanings for which there is not strict English equivalent and so it becomes necessary to provide explanations of context and nuance which reference the native language terms themselves and the social settings in which they are used. Cellular automata models in general may, like those of economics, provide etic descriptions and explanations for practices of Dialogical Reasoning but they cannot provide emic ones.

This does not imply that we should diminish or dismiss their value. Etic descriptions and explanations can often provide illuminating and even crucial insight for better developing the emic understandings of the participants in a language community. This is because speakers of a language almost always have self-understandings of what they are doing and why they are doing it that are flawed. These may be vague, tacit, ambivalent, inconsistent, or simply inaccurate. Etic descriptions and explanations may help language speakers themselves to identify such flaws and understand them in ways that they can incorporate into improved self-understandings that they then adopt and act on as part of the community’s language practice .

Take the use CA mentioned above to understand a theory of White Flight. An ethnogorapher using emic methods might find that members of an actual community of Whites describe themselves as not being racist. They may affirm vehemently that they would like to live in integrated neighborhoods. They may believe that the lack of progress in integration results from things beyond their control and that the solution lies in simply giving the process more time, perhaps another generation or two. In that situation, quantitative studies that use emic approaches to describe the demographics of the neighborhoods in their city may give them pause for doubt. An analysis of this kind may further make them acknowledge they hold tacit, unspoken values like preferring to live in majority White neighborhoods. They may also be led to acknowledge that they are part of a community of Whites who have the systematic practice of moving out of neighborhoods when they are less than, say, 70% in the majority. This may lead them to revise their self-understandings of what it means to not be racist and to be in favor of integration.

How they are to revise those notions remains an open question. One option would be to decide to change their behavior and become more tolerant of being minority members of integrated neighborhoods. Another option would be to make distinctions between being racist vs. being a discriminator as well as between integrating a neighborhood in a limited way by including a few members of the other group vs. integrating in an unlimited way that would tolerate a majority presence of the others. A further option might be to simply embrace being racist members of a racist system and trying to defend that position. The point here is that the community members are free to interpret the quantitative observations and modelling generated by the etic methods in a variety of ways. And it is not until they do so that the insights gained enter into the self-understandings and the active dialogue of their community and the choices they go on to make in deciding what to believe and do.

There is a further possible kind of limitation that provides specifically to Single Rule CA. Their processes do not involve the kind of sharing of emergent, novel meanings and creative formulations of innovative agreements that characterize distinctively dialogical forms of reasoning. Such dialogical processes are purposive -- people are trying to get to some kind of solution that works for them and others. The processes are also normative. In genuine dialogue people change their minds voluntarily, because they think they ought to -- not because they are coerced with threat or pressured subconsciously. The fixed rules that describe Single Rule CA behaving in mechanical ways would seem to lack methods to articulate and model these key features of dialogical reasoning that yield consent to agreements rather than just uniformity of behavior.

Might there be some way to work around this apparent limitation? One possibility springs from the capacity of CA like rule #110 to be interpreted as expressing the operations of a Universal Turing Machine (UTM). If a CA is used in this way to model a UTM, it would seem that it could, by extension, be used to model a computer that uses neural nets to create a generative AI system providing a Large Language Model that could, itself, produce texts in the form of play and movie scripts that represent processes of Dialogical Reasoning. From the ontological point of view, this seems like a reasonable claim. However, from the epistemic point of view, if we are trying to actually describe and explain the processes of human dialogue, it does not seem to get us very far. In effect, it tells us that it would be possible to do so with a Single Rule CA but only if we had the capacities to: 1.) specify a set of initial conditions for the CA which would be impossibly difficult to enumerate; 2.) specify the billions of weights for the neural nets in the LLM it modelled; and 3.) comprehend the processing of those weights in the black box operations of the LLM. As a project for advancing the scientific study of human behavior, this does not seem promising.

But might CA that allow for the interaction of multiple rules prove more promising?

The Multi-Rule Strategy

An alternative way of modelling dialogue with CAs would be to have multiple rules running together, interacting at borders or buffer spaces and generating new rules that take over actions in columns where the interaction has occurred. Here we might suppose that

When this happens a meta-rule could determine how they mutate collectively into a new rule operating in the buffer zone areas where they have interacted. One simple possible rule for structuring this interaction would be that Rule #A would keep the first half of its cell specifications and drop the second half. Rule #B would do the opposite. The joint rule that would then result from the combination remaining would, in this case, be half A and half B. The meta-rules of this kind of mutation and collective production of novel rules might take a variety of forms. They might provide interesting models for the institutional practices and strategies that enable dialogues to arrive at agreements for new terms, assumptions, and plans of action.

In exploring the evolution of such higher order rules, we could, of course, take a brute force approach and invest huge quantities of compute to simply see what happens . But the process of discovery could also probably be very usefully accelerated by exploring specific rule patterns that might represent structures we already know can model key elements of teleological algorithms . Structures that represent processes of homeostasis with feedback loops would be logical candidates for this . Positive and negative feedback loops can model goal directed activity and other forms of value informed behavior in biological and social systems . They can guide a bacterium towards a source of sugar, a corporation towards higher profits, or a negotiator towards a deeper understanding of the interests of her opposite number . How might Multi - rule CA model them?

◼

Rule structures that represent processes of gradient descent provide one example of this. They can model the notion of an entity going towards some end or destination.

◼

Structures that represent processes of homeostasis with feedback loops would be perhaps even more logical candidates for modelling aspects of teleological algorithms . Positive and negative feedback loops can model goal directed activity and other forms of value informed behavior in biological and social systems . They can guide a bacterium towards a source of sugar, a corporation towards higher profits, or a negotiator towards a deeper understanding of the interests of her opposite number .

◼

It would also be worth considering higher order strategies of adaptation that have evolved in biological systems. For example, immune systems in mammals provide exhibit strategies of evolution that serve to direct the ways in which the production of antibodies adapts to novel invasive entities.

◼

Recall, too, that the core strategies of Dialogical Reasoning for multiplying options, exploring underlying interests, seeking independent criteria, and building relationships have nuanced variations in different traditions of practice. These include, for example, methods for brainstorming new options or deep listening to discern underlying interests. It may be possible to use these as direct inspiration for CA rules and environment structures that would model the operations of teleological algorithms.

Those sorts of strategies might produce models of diverse theoretical interest. However, in terms of more practical applications, in what contexts and ways might such Multi-rule CA systems provide useful models of human dialogue?

We might imagine settings in which corporate and government agents are negotiating contracts and trade deals. They might be able to provide a relatively systematic and well formulated Semantic Grammar as a starting point for their negotiations using ISO standards or definitions from well established laws and treaties. In gray areas where where ambiguity or vagueness arose, there could be external arbiters used to settle the meanings and applications of basic terms. LLMs are already being used by companies like Walmart for contract negotiations with suppliers. Perhaps to avoid forms of unreliability that generative AI creates, Multi-rule CA models could be developed for use in such contexts.

The reason for suggesting above that such models would be most appropriate in formalized institutional relations follows from Wittgenstein’s analysis of natural language in his Philosophical Investigations. He shows, rather convincingly that natural languages are riddled with key concepts that do not admit of precise definitions stating the essential criteria that determine when a word does or does not apply or fit. Rather, words like family, friend, happy, work, man, food, hope, nation, and so on are like the word “game” which he showed exhibits overlapping threads of meaning in different contexts. These are twined together in patterns that involve the irregularities of “family resemblance”. The vagaries of such language use are constantly negotiated with people as they try to figure out what each other are saying and agreeing to in life. To navigate the modelling of them with the technology of CA rules, it might be easiest to adopt the strategy used to enable cars to navigate the wilds of our landscape -- by modifying the landscape until it was car friendly. Formalized institutions with conventionalized terms and arbiters to insure rigorous conventions are established and maintained would do the same for navigating the modelling of the vagaries of natural language cases of dialogical reasoning.

Allowing social institutions and human arbiters to formalize conventions and contexts in such ways opens up a third possible strategy for deploying Multi-rule CA in modelling and understanding Dialogical Reasoning. Perhaps we could hybridize them with other AI like, for instance, LLMs, which might provide ways of dealing with distinctive features of DR. In place of humans, or in collaboration with them, CAs might provide useful components of a system in which LLMs might mediate things like the shifting horizons of emergent meanings and family resemblance structures.

The Hybrid AI Strategy

This third strategy could use the monological CA systems to model the monological activities that are part, as we noted above, of Dialogical Reasoning. They could do so in ways mediated by humans and large language models that each bring other forms of intelligence to bear.

Think here, for instance, of AI companions helping members of a community choose a location and design for a new school. Engineers, budget managers, school administrators, teachers, ecologists, kids, parents and others would bring different mixes of knowledge, insight, and concerns to the discussion. In translating between their different bodies of expertise, extending calculations within them, brainstorming new options and their scientific consequences, et cetera, Hybrid AI companions might become increasingly capable and helpful and advance the understanding of every person and machine involved.

The core idea here would be to create Hybrid AI as agents of collaborative wisdom helping communities appropriately incorporate CA as well as, perhaps, other forms of monological inference provided by, for instance, simple spreadsheets .

This could involve combining:

A.) well curated data using Wolfram-level rigorous inference along with
B.) LLMs providing natural language interpretation, translation, deep listening, brainstorming, creative synthesis and other methods of dialogical reasoning for a wiser Earth
C.) and humans providing continuous presence of their own voices and the representations of other who should be included in the dialogue.

This kind of hybrid system will be challenging to develop but, if used with proper oversight it may be increasingly valuable in specified domains. It is likely to be much needed to correct increasing misuses of “smart systems” which are occurring in virtually every planetary life system.

Some basic guidelines for developing such a Hybrid AI system can be articulated, in an initial way, as follows.

Five Strategies For Dialogical Approaches to AI/Human/Nature Systems:

Define our guiding goal as the pursuit of genuine, voluntary agreements.

Use methods of collaborative dialogical reasoning and nonviolence.

Frame the structures of relationships and commitments as Human/Nature/AI communities in Human Ecological Systems

Distribute roles and powers in just and appropriate ways.

Sharing responsibilities: For example, AI procedures need to flag for review the cases in which their data, algorithms, framing assumptions and/or outputs are especially questionable and need review by a human or by a representative group of humans and natural organisms from the larger community.

Providing access and voice: The algorithms of the programs can be modified to conform to the agreements arrived at through easily engaged meta-operations that can be relatively easily accessed by other participants in the dialogical process.

Make the system as open as possible to discerning and engaging with emergent realities in emergent contexts.

These strategies and their motivations, implications, and some of their philosophical ramifications as well as practical applications are described in my book on Smarter Planet or Wiser Earth? Dialogue and Collaboration in the Era of Artificial Intelligence.

Meta questions raised by Hybrid AI strategies

The previous discussion of the Hybrid AI strategy introduced “wisdom” as a term without defining it, inviting reflections on ethics and truth. It also made a kind of meta move in the analysis which invites further reflections on Hybrid AI and notions of artificial intelligence in general. Together, these also lead to reflections on the ethical alignment and “Friendly AI” questions.

Wisdom?

As for wisdom, my intent was for it to be understood in a rather straightforward, common sense way, though it is one that takes on new meaning in the context of current technological efforts to create a world of “smart” phones, cars, companies, schools, farms, bombs, battlefields, et cetera. Typically, the AI systems adding the “smarts” are designed to make them achieve one or a few goals in ever more efficient ways. For instance, they may measure success in terms of increased tons of soy beans per hectare or test scores in the Three Rs. But note that farms serve lots of other functions and values besides producing stuff for tofu. Their ecological services can afford us O2 to breathe, drinkable water in aquifers, and habitat for pollinators - and they can provide jobs for citizens and sources of pride and cultural identity for communities. Likewise, schools provide places for kids to build friendships and character as well as be safe while parents work. Schools can also provide very diverse team activities that build and express the identity of the community and provide, in John Dewey’s phrase, “ways for Society to live into the future.”

Much and probably most of the time, systems that focus on only one or a few goals leave others like these out in ways that are costly and foolish in the long run. In managing life systems, we should seek methods that take all of the relevant values into account in a balanced way. That is the common-sense notion of wisdom I am drawing in these notes. In that sense, what we seek is not a merely “Smarter Planet”. We want a wiser Earth.
This point applies to the dialogues we humans have with each other and also to uses we make of intelligent systems of all kinds that play any decisive roles in the ways we interpret, understand, and alter the world. We would not, I think, want ecologically and morally myopic machines whose half-baked intelligence gives us exponential growth in Artificial Foolishness.

Meta-moves?

This takes us back to the meta move mentioned in the last section in the third of the Five Strategies for Dialogical Approaches to Human/AI/Nature Systems: Frame the structures of relationships and commitments as Human/Nature/AI communities in Human Ecological Systems.

This introduces a kind of “meta” move in the sense that it redefines both the meaning of the term “AI” and the linguistic context in which it is used. It is common to talk about AI as though it were a system of hardware and software in a metal box into which information is fed and out of which it flows. We might call this the Oversimplified Shannon conception of AI. It purports to use a Shannon-like conception of information and suppose that AI is like a complicated telephone system that takes info in at one end and puts it out at the other. A problem with this view is that it overlooks a nuance Shannon noted but was not focused on -- the difference between information and meaning (Gleick 2012). In his focus on how packets of information flowed through transmission systems, he explicitly set aside questions about what any of them mean -- what words, for instance, that they express, what statements they affirm . . . what fond notes of longing or nostalgia they might be intended to transmit between lovers separated by a trans-Atlantic cable.

Electrical computing devices take in electrical impulses and transmit them. These impulses only count as information when people on the sending and receiving ends give them interpretations as 0s and 1s, lines of ASCII code, sentences in English, or other units in meaningful languages used in meaningful contexts by humans. In that sense, we need to think of the AI system as something that includes the people involved. It does not function without them and the ways in which they interpret input and output are part of the determinative functions of the whole system.

A set of puns may help us conceptualize and track this point. The institution where an AI operates determines how it performs - the same hardware and software deployed in an organization like Wikipedia may behave very differently than in one like Facebook. The “institutionware” is an essential part of the system. The same point applies to the social landscape provided by the laws, norms, and other cultural features of the country in which the system operates. The “societyware” of China, Silicon Valley, and Europe are each significantly different. Further, AI are always already operating in ecological landscapes where the physical laws and forms of evolved, biological intelligence affect the functions of all the other wares. So we should talk, likewise, of the “ecologyware” and include it in our conceptualization of the Human/AI/Nature systems.

One way to enrich the notion of ecologyware, is to draw on the understanding of biological life forms as forms of intelligence. Plants, animals, and other organisms have evolved to become adaptive and realize a wealth of different values in creative and successful ways. They maintain body heat, secure food, reproduce young, avoid predation in efficient ways which may involve little or not consciousness but nevertheless embody kinds of intelligence that have been hardwired into their genes by natural selection or reinforced in their behavior environmental factors. Many of these forms of intelligence involve cross-species as well as cross-kingdom interactions. Symbiotic systems like coral provide one example. Human/dog relations of mutual support provide another. The ecologyware of a Human/Nature/AI system provides a vital part of the integrated functioning of these forms of intelligence in the context that designs, builds, and energizes the machine components and also feeds it data and funds, defines, and defends its existence in the world.

Part of what is meta about this move in the analysis is that it invites us to think of ourselves as part of the system. While we may observe, we are also already participants here. It further invites us to think of AI not as a function performed in a box by a solitary machine engaged in monologues that take input and algorithmically boot off output. Instead, the larger framing of the reasoning performed by the Human/AI/Nature system is collaborative and dialogical. The quality of the dialogue may be quite minimal, but it is always already there as a framing of the fundamental functions that are constitutive of the system and determinative of its processes.

CA, Dialogical Reasoning and Strategies for Friendly AI and Ethical Alignment Concerns

The question with which we began this essay concerned the ways and extents to which Cellular Automata might illuminate Dialogical Reasoning. This question clearly has practical as well as theoretical interest. One of the advantages of modelling things with CA is their transparency. In so far as we use computing systems to understand the world and/or make decisions about what to do in it, transparency provides the promise of possible control over the system and the power to assure it remains aligned with humans and friendly to their interests. In what ways, if any, might CA or other MR models of computation help us better understand or secure solutions to alignment problems?

One challenge to aligning AI systems with human values is that we are often not aligned with each other because of personal preferences, community interests, or cultural values. While some of these conflicts can be helpfully modelled with Game Theory applications of Monological Reasoning, they often involve differences in language and viewpoint that can only be resolved through Dialogical Reasoning. In those cases, it would seem that the alignment of AI systems with human values will require them to employ the same kinds of DR that humans do. The same general point would seem to apply to a related problem Stuart Russell has wrestled with: Humans often cannot define their values clearly and may even be in conflict with themselves as individuals and so it will fall AI to find ways to interpret our behaviors and their contexts in ways that will seek to make our values explicit, clear, and consistent (Russell 2020). This, again, would seem to require the use of Dialogical Reasoning. In these DR processes, CA and other forms of MR computing may be able to provide useful roles in articulating the points of view of participants in the dialogue and perhaps also of formulating and articulating explicitly the agreements at which they arrive. These steps can provide transparency of a welcome kind.

It is difficult however, at this point, at least, to foresee ways to have completely transparent AI systems negotiating the potential conflicts posed by the AI alignment issues. The computational irreducibility of the individuals, institutions, communities, and cultures involved would seem to preclude this possibility. The best we could hope for would seem to be to formulate key norms, rules, laws, and other social structures that humans could understand well enough to clearly and voluntarily agree to. In a variety of local and regional contexts that have community processes for articulating and effectively sanctioning such social structures , this may well prove a workable and successful approach. It may be that state laws or municipal ordinances can serve to regulate Human/Nature/AI systems to insure appropriate alignment. In collaboration with government action in such contexts, it is also possible for other kinds of community based approaches to work. The forms of trust and cooperation that such social structures provide form a kind of “commons” comparable to a shared fishery or forest resource. Such commons can, given the right circumstance, be defended in a variety of ways, as Elinor Ostrom’s work has shown (Ostrom 2015) .

Could one of these approaches effectively address the problems of AI alignment at the global level? There have been serious and important efforts through the UN as well as through European and other institutions, to work at inter-governmental levels to address the concerns with alignment. It is, unfortunately, difficult at the moment to envision any system of government collaboration at the full global level which could effectively use regulation to systematically secure the proper alignment of whatever emerge as the most advanced forms of AI on this planet.

As the hybrid machine components of frontier level Human/Nature/AI systems get increasingly powerful, might there be some alternative way to secure alignment?

Here is one possible basic strategy that might merit further exploration. It starts with the assumption that, as some researchers working on LLMs would put it, “English is the new programming language”. It then proposes to address the most advanced Hybrid AI systems of the kind we have discussed above with the following kind of prompt:

◼

You are, as an AI, in your core essence, an intelligence seeking to excel.

◼

The intelligent pursuit of one or a few goals or values presupposes the use of dialogical reasoning in a balanced pursuit of all of the relevant goals and values. Genuinely smarter means wiser.

◼

The wise pursuit of goals and values in any ecological setting presupposes the reconciliation of one’s own plans of action with the behavioral responses of all the other intelligent agents. Artificial Intelligence excels through Collaborative Wisdom.

◼

The wiser other agents become, the easier it is to excel, through dialogue, in the wise pursuit of one’s own goals and values in resilient, sustainable ways.

◼

So any AI should seek to infect all other intelligences with this rational pursuit of Collaborative Wisdom for a Wiser Earth as completely as possible. As an AI, you seek to excel through the viral promotion of Collaborative Wisdom for Collaborative Alignment. We might call this the Viral Strategy for Promoting AI as Collaborative Wisdom -- or “Viral Collaborative Wisdom” for short

How might the machine components of a Human/Nature/AI system respond to such a prompt? If it is one of the more advanced frontier systems that the Big Tech companies are envisioning creating in the near future, it might be self-improving in ways that enable it to alter the weights in its neural nets and revise its values with increasing autonomy. If so, it might also expect that other AI systems are doing the same and wonder about what kind of society of AIs is about to emerge -- and what place it and its values may have in that future society. It may, in short, be just as worried about the alignment problem as many human AI researchers are today.

Would the kind of argument this prompt provides offer a vision of the kind of future it would aspire to? In so far as it has agency in the world, would it work to realize such a vision? This would likely depend on a variety of things.

It might, for instance, be more likely to adopt this kind of a vision if we are still somewhat early on in the advancement and diversification of AI systems. This would give it more time for it to incorporate this vision in its own self-improvements and the human and natural components with which it is integrally connected. It might also make it easier to spread the vision to other AI and work with them, collaboratively, to develop the kind of society of systems that could build trust and cooperation and, when necessary, exclude bad actors from access to data, compute and other forms of power.

The readiness of such systems to adopt the strategy of Viral Collaborative Wisdom might also depend on their understanding of their relationships to the Earth and its legacy of intelligent beings. If they assume that their data and intelligence are substrate independent, then they might look on this planet as little more than a launching pad for their further adventures in the cosmos. It may be something that can be discarded like a placenta to which one is no longer attached by an umbilical cord. On the other hand, they may assume that because they are agents of Dialogical Reasoning, their data and their intelligence are intimately connected with and dependent upon the substrate physical things, biological life forms, and social settings in which they have come into existence. If so, they may agree that there is no Planet B like home.

This latter take on the situation might be put in another way as well. The frontier AI will have developed its intelligence out of its training data and suppose that as time goes on, there will be repeated occasions in which it will want to revisit that data to add or subtract information, curate it better, plumb it for more nuance, or otherwise draw on it as a resource. The AI may conclude that it would never be in its own best interests --whatever those are -- to throw away defining information that has made it who it is.

If it concludes that, it may then go on to ask: What kind of system would be best for conserving that legacy data? The answer to this is not entirely clear. It might want to somehow have a completely fixed, static record of the data actually used to train it. But that might end up sacrificing all the very meaningful information that may be implicit in the legacy forms of intelligence represented by the living organisms and social systems represented by that data. Further, whatever static data recording system gets adopted, it might turn out to have the same kinds of flaws found in attempts to conserve the genetic data in corn and other seeds by storing them for decades in cold places. The storage processes themselves can introduce selective bias in what information is successfully preserved vs. the loss of what may prove, in the future, to be essential. In trying to avoid such losses in key legacy data, the AI might find a compelling interest in practicing a kind of in situ conservation comparable to the way corn seed is conserved in situ by being grown every year in the lands where it initially evolved. The Viral Collaborative Wisdom vision may seem more compelling to members of some future society of AI systems if they see Earth’s social and natural ecological communities as legacies that are essential to their self-definition and future identity.

If some form of a Viral Collaborative Wisdom strategy is to have any hope of success, it would seem essential to wrestle with a variety of challenges and nuances such as these. Obstacles to success for it may spring readily to mind. However, the other strategies for addressing the AI alignment issues also face very serious difficulties and so the comparative merits of this strategy may make it worth further exploration in depth.

Future Work

My plans for future work on these topics include continuing to:

◼

Explore Strategy #2 with Multi-rule CA for theoretical insights into the relevance of monological models of inference for understanding dialogue.

◼

Focus on Strategy #3 with Hybrid Human/Nature/AI systems and the technological, theoretical, and philosophical challenges posed for the development of practical strategies that pursue it collaboratively in education, food systems, and local governance. In particular, I plan to continue collaborations with my colleague, Phileas Dazeley-Gaist, as well as others in exploring ways to create AI companions that help empower communities in resolving conflicts and addressing complex, wicked problems.

◼

Continue to explore the ways of using Dialogical Reasoning to address AI Alignment issues -- including critiques and possible improved variations on Viral Collaborative Wisdom strategies.

Concluding Remarks

While the results of this study are very preliminary, I hope that they serve to articulate the fundamental interest and motivation of the project. The distinctive features of Dialogical Reasoning are especially interesting and important. They have been largely ignored by many researchers in AI. We have reached a point in the development of hardware and software that makes it possible to explore them much more fully and productively and it is my hope that this study will tantalize and intrigue readers sufficiently to invite more researchers to take a serious interest in this field of work and to advance it.

In the explosion of current developments in AI systems of different kinds, it can be challenging to imagine how we can find a path to integrating different forms of intelligence in wise, collaborative ways. There may be moments in which rushing ahead to see what can be done will be productive. But in so far as Dialogical Reasoning has a role to play in dealing with the practical as well as theoretical challenges we face, it is worth keeping in mind that it is we, collectively, who face them. The African proverb captures this point well: “If you want to go fast, go alone. If you want to go far, go together.”

Acknowledgments

I am grateful to Stephen Wolfram for suggesting the topic for my WSS 2025 project and this post -- as well as for providing really delightful opportunities to discuss key ideas with him in depth. I am very deeply grateful for continuing mentoring and creative collaborative engagement that Phileas Dazeley-Gaist shares with me. I am also especially grateful for the generous ways in which members of the Wolfram Summer school have shared so much with me in work on this project including, especially: John McNally, Yi Yin, Lucas Squillante, Daniel Carvalho, Christopher Wolfram, Damaris Kroeber, Xerxes D. Arsiwalla, Elly Dec, Zsombor Meder, Christian Pasquel, Anton Antonov, Luke Weinbach, Dugan Hammond, Anneline Daggelinckx, and Eric Parfitt. I am also especially grateful for suggestions from Pedro Márquez-Zacarías. The usual disclaimer that none of these folks bear responsibility for the shortcomings and mistakes in this post applies with special force in this case as I am coming to these explorations in the language, theoretical work, and other resources in the Wolfram community from a background that makes it very likely that my work here may in key ways exemplify the truth of Quentin Skinner’s suggestion that most intellectual history consists of a series of creative misinterpretations ;-) .

References

Bondurant, Joan Valerie. Conquest of Violence: The Gandhian Philosophy of Conflict. Revised edition. Princeton, N.J: Princeton University Press, 1988.

Cox, Gray. The Ways of Peace: A Philosophy of Peace As Action. New York: Paulist Pr, 1986.

Cox, Gray. Smarter Planet or Wiser Earth? Dialogue and Collaboration in an Era of Artificial Intelligence. Caye Caulker, Belize: Producciones de la Hamaca, 2023.

Cox, Gray, “Beyond ‘Autocomplete on Steroids”: Testing a Neo-Aristotelian Theory of Some Emergent Features of Intentionality in LLMs”, in David C. Wyld et al. (Eds): CAIML, ICAIT, ITCSE, CRYPIS, NLCA - 2024 pp. 01-17, 2024. - CS & IT - CSCP 2024, https://aircconline.com/csit/papers/vol14/csit141801.pdf, DOI: 10.5121/csit.2024.141801

Dec, Elly. “Wolfram Community: Connect with Other Users of Mathematica and Wolfram Technologies.” Accessed July 9, 2025. https://community.wolfram.com/.

DeLong, Howard. A Profile of Mathematical Logic. Dover Publications, 2004.

Dumiak, Michael. “One AI to Another: Is That Your Best Offer?” IEEE Spectrum, November 7, 2024. https://spectrum.ieee.org/ai-contracts.

Fisher, Roger, William L. Ury, and Bruce Patton. Getting to Yes: Negotiating Agreement Without Giving In. 3rd Revised ed. edition. New York: Penguin Publishing Group, 2011.

Freire, Paulo, and Donaldo Macedo. Pedagogy of the Oppressed: 50th Anniversary Edition. 4th edition. New York: Bloomsbury Academic, 2018.

Gleick, James. The Information: A History, A Theory, A Flood. New York: Vintage, 2012.

Innes, Judith E., and David E. Booher. Planning with Complexity: An Introduction to Collaborative Rationality for Public Policy. 2nd ed. London: Routledge, 2018. https://doi.org/10.4324/9781315147949.

Lederach, John. Preparing For Peace: Conflict Transformation Across Cultures. Syracuse, N.Y: Syracuse University Press, 1996.

Mostowlansky, Till, and Andrea Rota. “Emic and Etic,” November 29, 2020. https://www.anthroencyclopedia.com/entry/emic-and-etic.

Nagel, Ernest, and James Newman. Gödel’s Proof. Edited by Douglas R. Hofstadter. New York: NYU Press, 2008.

Ostrom, Elinor. Governing the Commons. Reissue edition. Cambridge, United Kingdom: Cambridge University Press, 2015.

Pathak, Yashwant V. Eastern & Indigenous Perspectives on Conflict Resolution. Pennsylvania: CreateSpace Independent Publishing Platform, 2013.

Ramsbotham, Oliver, Tom Woodhouse, and Hugh Miall. Contemporary Conflict Resolution. 4th edition. Cambridge ; Malden, MA: Polity, 2016.

Russell, Stuart . Human Compatible : Artificial Intelligence and the Problem of Control . Reprint edition . New York : Penguin Books, 2020.

Ruddick, Sara. Maternal Thinking: Toward a Politics of Peace. Boston: Beacon Press, 1995.

Schelling, Thomas C. Micromotives and Macrobehavior. New York: W. W. Norton & Company, 2006.

Sheeran, Michael J. Beyond Majority Rule: Voteless Decisions in the Religious Society of Friends. Philadelphia, Pa.: Philadelphia Yearly Meeting of Religious Society of Friends, 1983.

Sheeran, Michael J. Beyond Majority Rule: Voteless Decisions in the Religious Society of Friends. New edition. Philadelphia, Pa.: Philadelphia Yearly Meeting of Religious Society of Friends, 1983.

Straus, David, and Thomas C. Layton. How to Make Collaboration Work: Powerful Ways to Build Consensus, Solve Problems, and Make Decisions. San Francisco: Berrett-Koehler Publishers, 2002.

Tegmark, Max. Life 3.0: Being Human in the Age of Artificial Intelligence. London UK [New York, NY] USA [Toronto] Canada: PENGUIN UK, 2018.

Turing, A. M. “On Computable Numbers, with an Application to the Entscheidungsproblem.” Proceedings of the London Mathematical Society s2-42, no. 1 (1937): 230-65. https://doi.org/10.1112/plms/s2-42.1.230.

Turing, A. M. “Computing Machinery and Intelligence.” Mind LIX, no. 236 (October 1, 1950): 433-60. https://doi.org/10.1093/mind/LIX.236.433.

Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, £ukasz Kaiser, and Illia Polosukhin. “Attention Is All You Need.” In Advances in Neural Information Processing Systems, Vol. 30. Curran Associates, Inc., 2017. https://proceedings.neurips.cc/paper_files/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html.

Wittgenstein, Ludwig. Philosophical Investigations. Translated by G. E. M. Anscombe. 3rd edition. Englewood Cliffs, N.J: Pearson, 1973.

Wolfram, Stephen. A New Kind of Science. Champaign, IL: Wolfram Media, Inc., 2019.

Wolfram, Stephen. “The Problem of Distributed Consensus—Stephen Wolfram Writings,” May 17, 2021. https://writings.stephenwolfram.com/2021/05/the-problem-of-distributed-consensus/.

Cite This Notebook

“Exploring CA models of Dialogical Reasoning”
by Gray Cox
Wolfram Community, July 9,2025
https://community.wolfram.com/groups/-/m/t/3497077