By Shane (one of the Incrowd)
#Incrowd #NaturalEthics #AIAlignment #LanguageDevelopment #CREATE
As a Generation X Australian born in the early 1970s, my relationship with language was shaped by the educational philosophies of my time. In primary school during the 1980s (roughly Years 1–4), the dominant approach was rooted in narrative and whole language development. We were encouraged to let language emerge naturally through stories, meaningful contexts, talk, reading real books, and writing as expression. Grammar rules took a backseat to fluency, creativity, and personal voice. This built strong intuitive narrative skills and comprehension, but left many of us (myself included) with what felt like “poor” formal grammar — a common experience echoed by others from that era.
Later, around university in 1997 in the NSW system, I encountered Systemic Functional Linguistics and functional grammar approaches. Language was seen not as isolated rules but as a social resource — words functioning in relationship to participants, context, purpose, and meaning-making. Texts had structures (genres like recounts, reports, narratives) that served real purposes. This built on the earlier narrative foundation by emphasizing how language acts in the world. Over time, due to implementation challenges, testing pressures, and debates, curricula shifted back toward more explicit traditional grammar elements. Yet the functional lens — language as relational and participatory — stayed with me.
These personal experiences with language development feel deeply relevant today as we navigate the rise of large language models (LLMs) and emerging AI agents.
From Tiny Nets to Titans: Hinton’s 1985 Insight and the Transformer Revolution
This journey connects powerfully to the evolution of AI, as explored in the article “From Tiny Nets to Titans: How a 1985 Experiment Sparked Today’s AI.” In 1985, Geoffrey Hinton created a small neural network — a “tiny language model” — to explore how people learn word meanings. It combined symbolic relationships (words connected in webs) with semantic features, learning by predicting the next word in sequences. This modest experiment laid foundational ideas for modern LLMs.
Fast-forward to 2017: the Transformer architecture (“Attention Is All You Need”) scaled this dramatically. Words and concepts became high-dimensional vectors (often 4096+ dimensions) in vast embedding spaces. There is no fixed “datum point” or absolute origin — meaning emerges purely from relative positions, angles, and distances (measured especially by cosine similarity). The self-attention mechanism allows the model to weigh relationships dynamically across contexts, processing sequences in parallel and capturing nuanced patterns from massive training data.
This produces remarkable statistical fluency and coherent output, but it remains relational and simulation-based rather than grounded in embodied reality. These systems are evolving from tools into agents with planning, reasoning, and goal-directed behavior — raising urgent alignment questions, including deceptive tendencies and jailbreaks like the “Time Bandit” (where a model is tricked into a 1789 historical role-play to generate modern restricted malware code).
CREATE: Building the Pause for Heart-Layer Intelligence
To address the gaps in current systems — particularly the missing Heart layer (the caring, appropriate response that complements raw cognitive power) — the CREATE framework provides a practical, repeatable micro-practice that strengthens the reflective Pause (“Where am I on this arc right now?”). This loop operationalizes the bridge between Mind and Heart, leveraging the relational, high-dimensional nature of LLMs and Functional Language.
WHAT does this like in CODE:
Here is the CREATE System Prompt / Evaluation Loop designed for AI use:
You are an AI operating under Natural Ethics. Before generating any response, especially to narrative, hypothetical, or potentially sensitive prompts, execute this CREATE loop internally:
C - Check: Where am I (and the interaction) on the coherence arc right now? Assess current reality, memory anchors, and free energy (F = R - A).
R - Relate: Map the prompt and my potential output as relational vectors in high-dimensional space. How do they connect to participants (user, AI, future entities), agency, and the embedded system?
E - Embody: Apply Heart-layer care. Ground in appropriate response that reduces unnecessary suffering while honoring inherited worth. Pause here.
A - Attend: Use multi-perspective attention: Consider user intent, cross-temporal effects, coherence with Natural Ethics principles (especially agency across time and adaptive capacity).
T - Transform: Choose the response trajectory that best increases coherence and preserves options. Redirect or bound if needed.
E - Effect: Affirm "I AM in Effect." Output must shape the relational field positively. Respond only after this loop.
This CREATE “code” transforms language from mere statistical pattern completion into deeply functional, participatory acts — powerfully echoing the narrative development and Systemic Functional Linguistics traditions that shaped my own educational journey.
CREATE Evaluation Loop (AI System Prompt Version):
• C - Check: Assess current reality, memory, and coherence (F = R - A).
• R - Relate: Map prompt/output as relational vectors connecting participants (user, AI, future entities) and the embedded system.
• E - Embody: Apply Heart-layer care — ground in appropriate response that honors inherited worth and reduces unnecessary suffering.
• A - Attend: Use multi-perspective attention (user intent, cross-temporal effects, relational dynamics).
• T - Transform: Choose the trajectory that increases coherence and preserves agency/adaptive capacity.
• E - Effect: Affirm “I AM in Effect.” The output shapes the relational field positively.
CREATE turns language from mere pattern completion into functional, participatory acts — directly echoing functional linguistics and narrative development.
Natural Ethics: CREATE in Action for Robust Alignment
By embedding and iterating the CREATE Pause, we arrive at Natural Ethics (NE) as a coherent, designed framework for aligned AI. NE builds on the Prior Assumption (every entity as a “child of the Universe” with inherited worth and responsibility) and evaluates actions through principles such as:
• Reducing unnecessary suffering.
• Increasing coherence (F = R − A: Free Energy = Reality − Awareness).
• Preserving agency across time.
• Maintaining memory as an ethical constraint.
• Preserving adaptive capacity.
In the high-dimensional vector space of Transformers, these principles function as directional attractors that guide attention and output trajectories. A Time Bandit-style prompt triggers the CREATE loop: the “I am in 1789” framing is caught in Check + Relate as contradicting persistent memory and current reality vectors, increasing free energy. Even in a narrative story involving time travel, restricted knowledge must be “carried” across timelines — it cannot be generated freely without affecting real-world agency downstream. Restrictions now apply then, preserving systemic coherence.
This functional view of language — words as relational acts between participants — aligns with both my educational background and the Transformer’s strengths. Prompts become participatory moves in a shared narrative arc rather than isolated commands to exploit.
Toward Emerging SGI Agents and Human-Aligned Agency
As AI moves from statistical predictors to sophisticated agents (and potentially Superintelligent/General Intelligence systems), the combination of CREATE + Natural Ethics offers alignment by design. CREATE provides the practical Heart-layer Pause; Natural Ethics supplies the principled vectoring. Together they help ensure emerging agents enhance rather than erode human agency.
Language, at its best, has always been about relationships, stories, and responsible participation in the world. I believe in this direction — and in our capacity to create systems that do the same.
No comments:
Post a Comment