System 3: Conversation as Intelligence

Michael Anton Dila
16 min readApr 5, 2023

We are used to thinking of intelligence as something that individuals exhibit. There are many reasons for this. One is the connection that many have insisted on between the brain and thinking. Another is a theory of mind that focuses narrowly on the identity of isolated individuals. “I think, therefore I am,” said Descartes. Another still is the tests that began to emerge in the late 19th and early 20th century from the practice of psychometrics: intelligence, in this context, became a mathematical quotient: intelligence = IQ.

These days we probably hear the word most often in the phrase, artificial intelligence. And, while the theory of artificial or machine intelligence goes back more than a hundred years, we most strongly associate its emergence with the pervasiveness and ubiquity of computing technology. In fact, our sense of the word technology, too, is strongly anchored in the field of computing. When we say tech, most people think of products, systems, networks that are powered by computing, even though very few people actually understand what computing, in this context, means and how it works.

Most of us who use computers of any kind understand that there is a distinction between software and hardware, for instance. We certainly know that apps and chips are different sorts of things, but most of us could probably say very little about how either of these things work or are made. We are not that interested, to be frank, because knowledge of these things is neither necessary nor particularly advantageous to being able to use an iPhone or an Android device.

Our ideas of intelligence are similarly black boxed. Most people do not really know what we mean by intelligence, but it is such a commonplace in our language that we all act as if we know precisely what people mean when they use the word. But as little as we might actually know, we have long since been primed for the arrival of machines that might think like us.

Alan Turing, the British mathematician, code breaker and designer of the eponymous Turing Test, authored a paper in 1950 which told us what to watch out for. He described an “imitation game” in which a human would interact both with another human s/he could not see and a machine. If the person playing that game could not tell the difference between the responses of the human and the machine, then we could presume the machine’s intelligence.

Turing’s test and the thinking behind it were nuanced and complex. He dealt explicitly with the question of what it is we call thinking, in the first place. We should notice two things about the Turing test: first, that the game at its center proposes an interaction, and, second, that this sort of interaction is absolutely paradigmatic for our experience of the interface that we know as ChatGPT. I call it an interface, because it is a program which gives us a way of interacting with the GPT (generative pre-trained transformer), which is (according to the Wikipedia) “a decoder-only transformer network with a 2048-token-long context and then-unprecedented size of 175 billion parameters, requiring 800GB to store.” Got it?

Chat is an interaction paradigm that has been around since the early days of the internet. If you know what IRC is, chances are you are old and nerdy. Most folks who use chat today have never heard of Internet Relay Chat. Our experience of chat forked from Internet messaging to mobile messaging with the advent of the telecommunications protocol called SMS, which most of us think of as texting. Of course, the medium of all forms of chat is that of text.

Within roughly the same time frame that ChatGPT made the scene, two quite different sort of AI systems were released into the world. DALL-E and Midjourney are systems that take descriptive input via text and produce images as output. There is a somewhat magical effect to this for which the word transformation seems apt. Very quickly people moved from imagining that they could produced creative output merely by descibing what they wanted to see, to creative people producing experiments which seemed to legitimately involve a creative partnership between humans and a machine.

From New York Times op-ed by Frank Pavich

In a New York Times op-ed, filmmaker Frank Pavich reflects on the experiments conducted by another filmmaker, Johnny Darrell, and the system called Midjourney. Pavich sees that there is legitimate creativity at work in this interaction. But while he sees the real and compelling creativity that arises from Darrells’s interaction with Midjourney, he never suggests that he thinks that Midjourney is, itself, creative. There’s something important I want us to notice in this example. Creativity is always a product of an interaction with another, even if that other is a machine. A machine, however, is not something that can, on its own account, be creative. Of course, what Midjourney and DALL-E have made possible is a new kind of access to creativity for those without training or skill working with visual tools. This advance in a certain kind of access is exciting and I have no doubt that it will lead to interesting new creative expression.

In another recent op-ed in The New York Times, Noam Chomsky and two co-authors offered a view of why ChatGPT falls short of the mark of an intelligent system: namely, its inability to provide explanations. Emerging machine learning systems can describe and predict, but they cannot offer us explanation of “what is not the case and what could and could not be the case. Those are the ingredients of explanation, the mark of true intelligence.” Chomsky, a linguist, believes that an innate ability to navigate/generate the underlying grammar of language is what sets human beings and human intelligence apart from the abilities attainable by machines. “[S]uch programs are stuck in a prehuman or nonhuman phase of cognitive evolution.”

Chat is an impoverished model of interaction. Joseph Weizenbaum created ELIZA in 1964 as a way to explore human/machine interaction. One of his hypotheses, based on the interaction paradigm between therapists and patients, was that human communication is actually more superficial than we think, and, therefore, perhaps not really so difficult to reliably simulate. Chatbots like ELIZA established the paradigmatic query/response model of interaction that still characterizes such systems today. These advances, in which Weizenbaum was a pioneer, have largely been focused on the ability of the system to parse ordinary language through natural language processing (NLP). No question, the technical advances in this domain have been considerable. And while such systems do not exhibit anything we might meaningfully call intelligence, human interactions with voice chat interface systems like Siri and Alexa have become commonplace and effective.

Chat based systems work. Not always perfectly. In fact, though there are notable and troubling examples of failure and dysfunction in such systems, many have achieved an impressive degree of reliability. Our expectations of them may be high, but our reliance on them is mostly limited to low stakes tasks, like activating playlists, dialing phone numbers, doing searches for the best hot dog in Chicago.

“It’s not you, it’s me.”

One of the most interesting exceptions in this recent history significantly pre-dates Kevin Roose’s interaction with the newly AI-enabled Bing search engine. When Microsoft was experimenting with chat and AI almost 10 years ago it developed a persona for its chatbot that had some very interesting consequences. In 2015, a reporter at Geek Wire claimed that 25% of users in China (10 million people) had said “I love you” to Xiaoice. Why users reacted so emotionally and confessionally to a Microsoft product is something that no doubt had researchers at the company hopeful that they had finally made a long sought after breakthrough. What was unexpected, I think (though perhaps hoped for), was that people might want to have a relationship with a machine “intelligence.”

Theodore enjoys a day on the water with his OS, Samantha in the film, Her

This unexpectedness is the premise of Spike Jonze’ film, Her. Theodore, played by Joaquin Phoenix, is a sensitive and sad writer of personal letters, who is recovering from his divorce from Catherine. Lured in by an ad campaign for a new kind of AI operating system, Theodore develops a relationship with the persona of the OS, Samantha. What is fascinating about this relationship is that it is conducted entirely through conversation. The relationship that develops between Theodore and Samantha is premised on conversation, not chat. The relationship between Theodore and Samantha has all of the hallmarks of the activation of what I call System 3, intimacy, immersion and intent. Under such conditions, love becomes possible, and perhaps out of this love can emerge a certain kind of intelligence.

One of the things that is evident in Michael Lewis’ book about Daniel Kahneman and Amos Tversky, The Undoing Project, is that there was a deep intimacy between them. This is never described as romantic love, but it occurs alongside Kahneman’s own claim that he and Tversky shared a mind. There’s a echo of this sense in the film Her; that to share intimately is to not only share a world, or share experience of the word, but to become one, in some real sense. Not one body, obviously. So, perhaps, one mind? This powerful form of relationship is, at least, a “place” in which divisions of all kinds break down in favor of connection. In this kind of intimate connection, people can become “more” than their individual selves and often become something other than they were. Many believe that such relationships can and do make us better people. What could that mean? Better in what sense?

Human intelligence runs on a platform called language. Our ability to think at all, as we understand thinking, is premised on having, and, as importantly, sharing a language. I have said elsewhere that there is a kind of thinking which is characterized by the connection of people in and through conversation. I call this way or mode of thinking, System 3. In order to understand how it works we need to be precise both about what we understand conversation to be and what we mean by intelligence. A conversation is a system of exchange and interaction through which people seek to understand and agree with one another. It is through the mechanism of conversation that we come to share a world with others, whether that world is that of a shared religion, culture or society. There are other kinds of conversations that are more circumscribed and purposeful. I’ll use three examples to describe how conversations function at different levels of scope and scale: making a film, doing experimental physics, and dismantling a system of White Supremacy.

A feature film production is a thing of variable extension in time, but, on average, from my limited experience working in the profession in my early years, they last about a year. Over that time, a person or person initiates a project, and almost immediately begins to enroll collaborators of a wide variety of kinds: people who will help finance the production, writers, directors, grips, electrics, costume and set designers, editors and more. The end goal, of course, is a completed film in market. Throughout the journey to get there, all of the participants in the production are involved in a conversation that we might simply call, “the Production.” What is important about this conversation for our purposes is that no one knows at the beginning of the production how the film will actually get made. As detailed as plans get, production schedules, call sheets and other production management artifacts, at best, provide a mere outline of the activities and decisions that constantly need to be discussed and decided before action is possible on any given day.

We have likely all seen a dramatized conversation between a Director and actors about a scene they are about to shoot. In such conversations a director might communicate specifics about the action of the scene that are note detailed in the script or production notes. These may include instructions about the movement of players on the set, the emotional tone the Director is trying to set, or notes about the goals of individual characters in a scene. Sometimes in such conversations the Director will introduce a change that has occurred to them. Often this leads to a conversation between Director, Actors and other Crew about adjustments that now need to be made to the original shooting plan. The Director of Photography may introduce a new approach to lighting the scene to better accomplish a given effect that is part of the Director’s intent, but not part of any explicit instruction.

There is a very real way in which the “the Production” simply is this conversation that takes place over the course of a year, which is occasionally punctuated with breaks in the conversation during which the actual work of filming takes place. When I spent a little over a year working on a feature film production, this was, in fact, exactly what it was like. What I want to call to attention to is that it is that conversation that not only makes the work of “the Production” possible logistically, but it is also how the collaborative art project of a film gets made. The conversation, as much as the film’s script, is a vital part of the artistic process. If we were diagram that conversation, we would see that while it has some linearity, it also describes a messy tangle of loops, from on set decisions about action in a scene to real-time change in the craft services plan for what time lunch will be made available and what will be served.

The quality of this conversation during “the Production” at every level will have an impact on the quality of the product that emerges from the process. Throughout this process there are a host of interdependencies, great and small, which not only affect outcomes, but provide ways in which the quality of inputs can be changed, improved and innovated through feedback. “The Production,” as a whole, is a kind of self-organizing intelligence that emerges from the conversations that happen from the bottom-up. It is this intelligence that produces the film, rather than any one or any particular group of people who are part of the system. It is in a very real sense the conversation which produces that intelligence or, at least, allows it to emerge.

Now, let’s think about how one of the world’s most ambition projects in fundamental physics operates in much the same way. Consider the project to run experiments at the Large Hadron Collider (LRC) at CERN, located just outside of Geneva, Switzerland. The LRC is one of the largest experimental apparatus that human beings have ever built. So large and some complex is the LRC one could easily mistake its construction and operation for the goal of project to bring it into existence. But, in fact, this massive piece of equipment was build so that could advance a conversation about the nature of the interactions of the fundamental particles that make up our physical world.

There is an excellent documentary film, Particle Fever, which beautifully documents the last couple of years leading up to the running of initial experiments at the LRC. At the center of the action is an experiment that is hoped will resolve the status of the hypothesized Higgs boson particle. The films does an amazing job of demonstrating the complex interdependencies of the project, from the building and testing of the apparatus, to the many sub-component conversations that were vital in advancing the several theoretical and experimental projects to be part of the first experimental run of the LRC. What is unmistakable is that without all of these conversations, not only would there be no sense in conducting these experiments at all, but neither would there be an international community of collaborators who advance the work of discovery and understanding in the science of the physical world.

This may seem to be trivially true, in the sense that we might find it obvious that conversations take place among those who are doing such work. My assertion is stronger that that, however. What I am saying is that the conversation is the locus of the intelligence at work in the system. To further explain this, let‘s consider an adjacent example. In a legendary incident, the Noble laureate physicist, Richard Feynman, was giving an interview to a journalist. The journalist noticed some notebooks in which Feynman had written, made drawings and scientific notations. The journalist remarked how impressive it was that Feynman kept such a detailed record of his thinking. Feynman corrected the journalist, saying: “That is the thinking.” The point of my comparison is that System 3 thinking is that thinking that becomes activated in conversation and that there is an emergent intelligence that is neither the sum of individuals, nor merely a product of the individual’s participation. Individual selves, to the degree that people become truly immersed in a conversation, become part of something that is other than self or a collection of selves. I am going to call the state that such people enter in such practice “entangled consciousness” and I will have much more to say about this concept in future posts.

Now to the most complex level of scale in my list of examples, the social scale of the system of White Supremacy. Here I want to use a historical example. I hope it will begin to illustrate the dynamics of a conversation at the social scale. Let’s call this the “Dominant conversation.” The part of history I want to call to mind is centered on the decade from 1954 to 1964. At the one end of that ten year period is the Supreme Court decision in the case of Brown v. Board of Education, which began the process of desegregating public schools that had been racially segregated under the so-called “separate, but equal” doctrine. On the other end, was the passage of the Civil Rights Act of 1964 which finally explicitly prohibited by statute discrimination on the basis of race, color, religion, sex or national origin. Both of these “events” are outcomes of a set of entangled conversations. I will only focus on three of the major ones: the social science conversation (in this case, social psychology), the legal conversation (involving three distinct types of legal conversation: legislative, judicial and executive), and the activist conversation (the collection of activist conversations from church groups to citizen groups, broadly referred to as the civil rights movement).

This set of conversations became entangled almost one hundred years after the abolition of slavery in 1865. In many ways the system of slavery had effectively persisted after its legal abolition. There is a great deal of detail in the history of the interaction of these conversations that it is too far beyond my scope and expertise to elaborate here. Let me take the risk of being simplistic and say that the activist conversation drew attention to the many ways that the legal regime of White Supremacy was very much in effect in the discriminatory practices of legally enforced racial segregation, including that of the public school system. The social science conversation, through the work, most notably of Kenneth and Mamie Clark, that sought to establish a evidentiary basis for the proposition that segregating Black children did them demonstrable harm, disadvantaging them educationally and therefore denying them the equal protection of the laws guaranteed by the Constitution. Finally, the legal conversation, initially the judicial one, by engaging the arguments of the activists and social scientists found a basis for providing a legal remedy to the persistent injustice of systemic racism at work in the racial segregation of public schools.

The entanglement of these conversations led to important and enduring, albeit incomplete, change in the workings of a political system of dominance. They made a different reality conceivable, imperative and necessary. Though these conversations did not change everything and certainly not all at once, I hope I have been persuasive in showing the role conversations can play in both the construction of reality and in changing it. It is an important addendum to this arguments that the intelligence of conversation does entail that the outcomes of conversations are always desirable on account of being the product of an intelligence. Intelligence is in this sense is a kind of power, and not necessarily a type of wisdom.

In a way that may be both encouraging and also somewhat destabilizing, I want us to realize the degree to which reality and our places in it are not merely effected by the dynamics of conversation, they are one and the same. Reality is conversations all the way down. We do not and cannot live outside of language. The great Austrian philosopher of language, Wittgenstein, said: “the limits of my language are the limits of my world.” We live inside a world that is itself alive. This is true, of course, of the natural world, but the life of conversations is, in it’s way, as dynamic and changeable and liable to birth, growth, as well as death and decay. What keeps worlds intact and vital are our constant renewal of agreements. System 3 is the thinking of worldmaking and it is through this thinking alone that we can access the capacity for change.

Why does this matter? At the moment we are entangled in two worlds, one that many of us hope may be dying, and another that is beginning to emerge, and, perhaps surprisingly, our ideas of intelligence are very much at the center of both of them.

White supremacy as culture, ideology and reality is very much a product of a theory of superiority. The cornerstone of that system of belief is the concept of race which is organized around a central pillar of a conception of intelligence that not only posits intelligence as individual, but also assets that levels of intelligence are biologically determined according to phenotypes. As much as it may appear that this so-called science has long since been rejected, there is strong evidence of its persistence in both scientific thinking and that it has pervaded the dominant culture in many places.

Artificial Intelligence is not only also anchored to our ideas about intelligence, but also caught up in troubling theories of superiority. As long as we have wondered about the possibility of creating a machine that could attain an intelligence equal to our own, we have also worried and, in some cases, hoped for a kind of machine whose intelligence would surpass our own. Here, too, it is unmistakeable that our ideas and models of intelligence are equated both with power and superiority. And just as some once asserted the inevitability of the stratifications of race, many now argue for the inevitability of the emergence of a form of intelligence that will either destroy us or make us inferior.

Understanding System 3 thinking is vital to ensuring that we are active participants in the worlds we live in and in ensuring that we do not accept the role of bystanders. System 3 thinking has always been part of movements of resistance and change. Our more globally connected, networked and contiguous world(s), however, means that the terrain has acquired new kinds of complexity. We need new conversations and new language. Now, more than ever.

--

--

Michael Anton Dila

Michael is a Design Insurgent and Chief Unhappiness Officer