Want to use this journal entry to sketch out the structure of the argument for an alternative view of AI models and respond to the essay by K Allado-McDowell “Designing Neural Media” which touches on some of the same observations about language and its ecological nature.
Some key themes to unravel:
- Language
- Language is networked and dynamic
- “Written language is also relational. Whether human-made or machine-generated, it only exists inasmuch as it is consumed by a reading mind…The networked nature of meaning-making in human-AI interaction points to a third high-dimensional structure stitching together humans and machines, that is, language itself.”
- I agree about the third high dimensional structure, but would also say that it’s never just language but rather the combined resonance of channels that convey meaning. For example, the co-constituted interactions between human and AI are also acted upon by
- underlying structures in the AI
- the human context of the interaction
- the task being solved
- the physical location the interaction takes place in
- background sound/music
- thoughts, emotions, feelings shaping the human at the time of the interaction
- see circumlocution
- I agree about the third high dimensional structure, but would also say that it’s never just language but rather the combined resonance of channels that convey meaning. For example, the co-constituted interactions between human and AI are also acted upon by
- “Written language is also relational. Whether human-made or machine-generated, it only exists inasmuch as it is consumed by a reading mind…The networked nature of meaning-making in human-AI interaction points to a third high-dimensional structure stitching together humans and machines, that is, language itself.”
- Comes into being through its ecological nature
- Leaves behind a trail of documents (a palimpsest of the collective consciousness)
- These documents become training data
- They capture language, and also its negotiation between actors
- Some language is factual, a lot of it is not but rather is representative of the intra-actions between writers and themselves and others
- Language is perceived
- A key process of meaning making
- Meaning making isn’t purely linguistic but works across multiple channels: language, sound, sight, touch, experiences of texture, emotion and emotional cues
- The space of channels upon which meaning making occurs is vast but (probably) finite
- The dynamical process by which meaning is made in an ecological network, and then the “latent space” that is continually opened up through this process, the space of concepts that have been invoked (this definition needs more work) I’ve been calling the sociosemioscape
- It often gets invoked as “landscapes of meaning” but we might speculate that there is a representational structure for this landscape, an underlying high-dimensional representational model for the process and the structure it acts upon
- Language models have their own approach to capturing the underlying structure
- Is sequence prediction a key mechanism for both?
- k writes about this process eloquently
- “Language and meaning are ecological in the sense that they surround us and emerge from the environment through dynamic relationships between species and forms of intelligence. This suggests that neural media’s high-dimensionality ought not to reduce selves to models or elevate models to selves, but should instead view both selves and models as relational structures in a high-dimensional network of meaning.”
- Whether or not the high-dimensional structure is a network or some other topology is still an open question, to me, but surely it has networked properties in the same way that ecologies do. Reducing it to a network might be a modeling assumption—after all, relations in the world are not quite networks as much as they are assemblages of actions that happen along the contours of relational networks and still are open to constant reconfiguration, distributed action, and so on.
- the “high-dimensional network of meaning” that comes into being in specific interactions between human-AI I’ve been calling the Semioscape
- it can be extended and traversed through language interactions
- it relies on prior contexts of humans and AI, along with their “context windows”
- it has the fluid, high-dimensional properties of an ecology and a landscape, one in constant flux and defined through it’s relational properties—co-constituted through linguistic and non-linguistic interplay
- “Language and meaning are ecological in the sense that they surround us and emerge from the environment through dynamic relationships between species and forms of intelligence. This suggests that neural media’s high-dimensionality ought not to reduce selves to models or elevate models to selves, but should instead view both selves and models as relational structures in a high-dimensional network of meaning.”
- Language is networked and dynamic
- associations
- Hypothesis: Associations are fundamental mechanism by which meaning is created, modified, transmitted, and extended within the high-dimensional structure of meaning
- Understanding association, conceptually, but also as a process used by both humans and LLMs can better help us understand the underlying structures of the sociosemioscape and the Semioscape
- Hypothesis: Associations are fundamental mechanism by which meaning is created, modified, transmitted, and extended within the high-dimensional structure of meaning