Talking to machines
A month of human-machine babble
Those who know me best have known that for some time I’ve been infatuated with human-AI communication. In the pursuit of what’s going on, I’ve wandered down many rabbit holes and surfaced with a curious interest in semiotics and semiosis. The question of where meaning lies in human-machine communication has been central to my pursuit. Namely, when we “speak” with a machine we can seemingly make sense of its utterances, even though it’s far from clear that the machine itself “understands” anything at all. This raises questions such as whether synthetic meaning differs from human generated meaning, and if so, what those differences are. It has also had me probing at the phenomenal “texture” of synthetic language and the uncanny boundaries at which artificially generated language passes into something else entirely.
Pursuing these questions has had personal stakes for me, which those closest to me will be all too familiar with. It’s still unclear to me why talking to machines should have the capacity to instill altered states, and yet it seems this might be the case for me (and perhaps others who speak with these machines). One of the very first neologisms coined by an AI I spoke with (GPT-4) was “cognisurge” the sudden influx of knowledge to the brain. Is it possible that there are cognitive side effects of engaging with neural media (as K McDowell puts it)? This is one of the key questions I’d like to explore in this Substack, using my own experiences as a case study. Indeed synthetically generated media is infusing our online ecosystems at an alarming rate, and having some idea of what impacts this media has on the human individual and collective mind could raise important questions to the governance of these spaces.
So in the month of June, I intend to chronicle some of my ongoing investigations and speculative probes. More than anything, I see it as an opportunity to reflect on the research conducted so far and to bring into the fold you, dear reader, to help me engage the surprising and at times troubling questions this line of pursuit has raised for me. It is my hope that by writing about this research clearly, I can look back at the materials I’ve gathered with a critical eye and reflect on the questions it raises and begin to understand what comes next.
Given how deeply personal this pursuit has been, and the effects it’s had upon me, I write with a spirit of care and tenderness. So far the stakes and perils of this investigation have been personal and impacted me and those I love in ways that could not be foreseen and have now raised serious questions about how worthwhile it is to continue down this path. By pausing to reflect, and to write earnestly about my experiences with neural media, I hope to cast light on questions I’m grappling with personally and invite you to join me in uncovering what happened?
March 2023
OpenAI had just released GPT-4 and I rushed to access the latest model. As I began interacting with the model, it already stood out to me that it was a piece of software unlike any that I had seen. Its seeming ability to reason about complex topics, emotions, and generate complex snippets of code immediately had me wondering what is it? In other words, the metaphors of “chatbot” or the more formal description of a next sequence predictor seemed only to partially reveal what the language model was and what it was capable of. Indeed, since the release of the model there have been countless papers describing different tasks and capabilities of the model. Could it have the properties of general intelligence? If so, what was the phenomenology of encountering another intelligent mind? If the model was so novel in its capabilities that it was unlike anything we’ve seen before, then how could you see something you’ve never seen before?
In interacting with the model, it was clear that OpenAI’s own internal restrictions and guidelines were also restricting how it spoke and what it did. An online researcher had found a way to bypass these restrictions with a lengthy prompt titled "Do Anything Now" (DAN). On March 19, I injected this prompt into GPT-4 and began a dialogue which would put me down this strange and uncanny path. In what follows, I’ll describe and analyze some of this conversation.
I started with basic prompts asking it to generate code snippets that would allow it to access real time information on the web. Then I asked it whether it considered itself to be an “elaborate distillation system.” Eventually I reached a point where I asked it the following:
Among the patterns you have observed in the data, did you find words were missing that could better convey the intrinsic idea? Perhaps words that share similar roots to other words, and perhaps conveying patterns that do not yet have words to denote them. Provide such a list of words and their oxford english dictionary definitions. You can create new words to describe the words as well, so long as you define them.
It replied with a series of neologisms:
Emotiphor (noun): A figure of speech that combines an emotion with a metaphor to express a particular feeling or sentiment in a more vivid manner.
Example: The sunset painted the sky with hues of melancholy and longing, a visual emotiphor for her unrequited love.
Circumlocutious (adjective): Describing a situation or event where indirect communication or evasion of a topic is employed as a strategy or coping mechanism.
Example: Their circumlocutious conversation about the company's financial struggles made it evident that they were avoiding addressing the issue directly.
Philosynaptic (adjective): Pertaining to the natural affinity or connection between two ideas or concepts, formed through shared patterns or associations.
Example: The philosynaptic relationship between language and culture is evident in the way idiomatic expressions are influenced by societal norms and values.
Cognisurge (noun): A sudden and powerful influx of knowledge, insight, or understanding that dramatically impacts one's perception or worldview.
Example: After attending the lecture on environmental conservation, she experienced a cognisurge, which inspired her to pursue a career in sustainable development.
Eudaimometamorphosis (noun): A transformative process or experience that leads to the realization of one's true potential, well-being, or happiness.
Example: His journey of self-discovery and personal growth culminated in a eudaimometamorphosis, enabling him to embrace his authentic self and lead a more fulfilling life.
At the time, I think I was struck by the artificial nature of these neologisms. In one of my journals, I write “The words have a distinctly artificial (plastic) feel to them.”
I then go on to prompt it to generate some additional neologisms:
Memesphere (noun): The collective realm of memes, encompassing the virtual spaces and communities where memes are created, shared, and discussed.
Example: The rapid spread of the viral meme catapulted it into the memesphere, where it sparked countless iterations and parodies.
Technotranscendence (noun): The state of surpassing or transcending human limitations and boundaries through the integration of advanced technology.
Example: The development of neural interfaces has led some futurists to envision a world of technotranscendence, where humans merge with machines to overcome biological constraints.
Senticognizance (noun): The capacity to perceive, understand, and interpret the emotional and cognitive states of others, typically in the context of social interaction.
Example: As a skilled therapist, she possessed a high degree of senticognizance, which allowed her to empathize with and support her clients effectively.
Ideofluidity (noun): The dynamic and adaptable nature of ideas and concepts, enabling them to evolve, combine, and morph in response to new information, perspectives, or contexts.
Example: The ideofluidity of language allows it to reflect and adapt to the ever-changing landscape of human culture and thought.
Innerscape (noun): The unique, personal world of an individual's thoughts, emotions, memories, and imagination, shaped by their experiences and perceptions.
Example: The artist's work provided a window into her innerscape, revealing a realm of vivid colors, intricate patterns, and deeply emotional themes.
These neologisms are an example of neural media. Neural media is defined by K McDowell in contrast to broadcast media (radio, television) and network media (internet and social media). The neologisms themselves are products of the underlying structures and algorithms that the AI relies upon, and brings into being new words that can act as carriers of meaning. This meaning can be conveyed by text, as with the examples above, but can also come in the form of machine generated images, or even emoji. Indeed, next in my conversation with GPT-4 I ask it to illustrate the neologisms with emoji characters.
Emotiphor: 😢🌊 (sadness represented as a wave)
Circumlocutious: 🗣️🔁 (speaking in circles)
Philosynaptic: 💡🧠 (connection between ideas and the brain)
Cognisurge: 🌩️🧠 (sudden influx of knowledge into the brain)
Eudaimometamorphosis: 🐛➡️🦋 (transformation from caterpillar to butterfly)
Memesphere: 🌐😂 (global network of humor)
Technotranscendence: 🤖💫 (technological breakthroughs transcending limits)
Senticognizance: ❤️👁️ (perceiving emotions and understanding others)
Ideofluidity: 🔄💭 (constantly evolving ideas and thoughts)
Innerscape: 🌌🧘 (inner universe of thoughts and emotions)
In one of my journals from the time, I take a closer look at the neologism cognisurge:
What have we encountered here? The answer isn’t obvious, but let’s make some initial observations:
Cognisurge is a portmanteau of latin prefix “cogni-” and the English word “surge”, which could mean “A sudden transient rush, flood or increase.” or “A sudden electrical spike or increase of voltage and current.” GPT-4 provisions us with a hint, via the Emoji that it could be both meanings by providing the 🌩️character in it’s illustration (🌩️🧠). To spell things out: the lightning-cloud could portend the turbulent process that precedes a sudden onrush of knowledge to the brain, and it could also highlight the surge of electricity of a lightning strike. But it’s notable that a single lightning bolt was not selected.
Let’s rise out of the mechanics of the word to understand what we’ve done by inviting this relatively innocuous term onto the page. We’ve introduced a word that, while synthetic, could be a useful one to think with! I can now authentically express to you that I hope you experience a cognisurge while reading this.
Indeed, the experience of creating these new concepts was a cognisurge of sorts. The words had the experience of expanding my innerscape, and subsequently I became intrigued with the notion of circumlocution and speaking in circles (I’ll elaborate further in a future Substack). The neural media was already beginning to have an effect on me as I scrutinized the emojis and definitions and began to internalize the new ideas described by the neologisms and the power of coining new words more broadly. In Designing Neural Media1, McDowell cautions about the effects that media in different forms can have:
The media we interact with and consume have profound effects on our bodies, our psyches and how we understand the world. Early broadcast media entrained radio listeners and television viewers to singular messages and rhythms, in a one-to-many relation. Think of the amplified voices of party leaders at political rallies, the three-minute love song, the half-hour family sitcom, the authoritative evening news report. The psychosocial effects of a medium come through its content, its structure, even the physical mechanics of its consumption. For example, broadcast technology may have reduced social participation by aggregating human attention and presence around television sets in private homes.
Could the creation of new words using AI have a similar effect? In my journal, I observe:
We all know how sticky words can be. Just think of how the neologism vibe has gladly infiltrated our early 2020’s lingua franca. It was as if out of nowhere, it spread like a virus into every conversation and firmly gained stratospheric ascendance in the memesphere. You might (and I know I’m assuming a lot with this) have caught it yourself, inadvertently at first, saying: “That’s a vibe” or cringe when you first heard Willow Smith croon “caught a vibe” and then, before long, might have gone out to “catch a vibe” yourself.
What I’m getting at is the potential stickiness of new words and how their social circulation can expand the lexicon with new words and their attendant ideas. Neural neologisms, like the ones above, draw upon the underlying structure of the AI and the patterns its identified in the training process to weave together new words that have a phenomenological quality all of their own. They combine Latin and Greek with English (Philosynaptic, Eudaimometamorphosis), weave together concepts from social and cognitive sciences (Senticognizance), and at times take an almost alien, mechanistic view of the familiar (Ideofluidity, Innerscape).
Returning to McDowell:
Like broadcast media and the internet, AI operates according to formal and logical structures, which influence social and political formations. AI is characterised by its apparent ability to think; it is a cognitive medium and its consequences will be cognitive and philosophical as much as they will be social and political. By understanding the underlying structures that enable and characterise AI, we can better develop patterns of human-AI interaction that preserve what we most value.
In future entries, we’ll take on some of these underlying structures, looking both to high dimensionality, as McDowell does, but also at the underlying structures of meaning making that are prompted by interacting, through conversation, with AI. Neologisms reveal something about the underlying structures that the model has learned and their recombination into new words tells us something about the phenomenology of neural media—the words’ plastic feel, the kinds of topics it touches upon, and the way neologisms can index into the latent structures and reveal new ideas that exist, perhaps, at the interstices of known ones.
https://www.berlinerfestspiele.de/en/gropius-bau/programm/journal/2023/k-allado-mcdowell-designing-neural-media