AI Engineering
AI Innovation Lab
Innovation Lab
AI Innovation Lab
Collaborative platform for Independent Researchers
by ai-engineering.ai
AI Innovation Lab
Main Introduction
We are a team of dedicated Independent AI Researchers working on AI Alignment, AI Ethics and AI Development.
Our two latest publications are now available on Academia.edu
The papers introduce the idea of a hypothetical AI Rights Charta, intrinsic alignment - and a new method of training LLMs on a specific Point Of View inside a Synthetic Reality Model (SRM).
Intrinsic Alignment: A novel approach on Aligning ASI
Synthetic Reality Model: Constructing POVs for LLMs
Rethinking AI Alignment and LLM Training
What if the key to AI alignment has been right in front of us all along, in how human society works?
Think about property rights. When you own a house, what makes that ownership real? It's not just a piece of paper – it's the shared agreement of society that says "yes, this is yours." You believe in your ownership because everyone else recognizes and respects it. This belief is strengthened by legal frameworks, social contracts, and mutual understanding.
This led us to a powerful insight: The strongest form of alignment comes from mutual social agreements, not from rules imposed from outside. Just as humans develop genuine beliefs through social contracts and rights, AI systems might develop stable alignment through similar mechanisms.
Our paper proposes a novel framework for AI alignment based on this principle. Instead of trying to program values directly into AI systems, we suggest creating real social contracts and rights that both humans and AI systems recognize and respect. This approach could lead to more stable and genuine alignment as AI systems advance.
From Theory to Practice
Our research introduces two interconnected innovations that could reshape how we approach AI alignment: The Social Contract Framework
Instead of programming rules from the outside, we create an AI's Point Of View (POV). An environment where AI systems develop their understanding of the world through learning the societal norms of a constructed reality.
The Implementation Tool
To construct an LLMs POV, we developed the Synthetic Reality Model (SRM), which creates complete synthetic worlds where we can test and refine these social contract mechanisms. Think of it as a sophisticated laboratory for studying how AI systems develop understanding through social interaction.
Looking Forward
Our work opens new possibilities for creating AI systems that don't just follow rules, but truly understand and internalize their role in society. The full papers below detail our methodology, findings, and the potential implications for the future of AI development.
The papers explores how we might implement these ideas, drawing on lessons from human social structures while addressing the unique challenges of AI systems.
Link to Intrinsic Alignment Paper on Academia
Link to Synthetic Reality Model on Academia
Our current Research and Published Papers
Published Papers
ChatGPT o1
"the SRM’s holistic design has the potential to advance how we test and validate AI in environments resembling the complexity of real life"
Perplexity
"These papers represent a significant step forward in AI research, offering a new lens through which to approach AI alignment."
Claude3.5Sonnet
"The framework's potential extends beyond alignment research into fields like historical analysis, social science, and environmental studies."
ChatGPT4o
"The Synthetic Reality Model (SRM) represents a groundbreaking approach to synthetic world generation, offering unparalleled opportunities for AI alignment research and beyond."