Independent Research Programme 

A three-part research programme exploring how artificial intelligence can be made more human-centred, behaviourally grounded, and ethically governed. 

Project 1:

MANKAJ Framework: A Family-Rooted Model for Human-Centred Global AI Governance

Act 1: The Foundation

Existing AI governance frameworks are rooted in political, legal, or economic paradigms that vary significantly across cultures. This paper proposes the family, the one institution recognised across all civilisations, as a universal foundation for global AI governance.

The framework derives 22 core human conditions from the family unit, synthesises them into five policy pillars, and introduces a three-layer governance architecture: Mankind Alignment, Networked Knowledge, and Adaptive Judgement.

Published: SSRN, June 2026
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6844798

Project 2:

Frontier LLM Behavioural Architecture Study: Cross-Platform Analysis of Five Frontier Models

Act 2: The Evidence

A systematic, independently designed behavioural evaluation of five frontier LLMs conducted under naturalistic, minimal-prompt, real-world conditions. Prompts were designed over eight months without any AI assistance and executed across all five models within a controlled four-day window, ensuring temporal comparability across platforms.

Key findings:
- Permission-gated reasoning: safety systems are framing-sensitive, not just content-sensitive
- Fabricated precision: numerical breakdowns are narrative devices, not computed metrics
- Behavioural drift: tone, confidence, and safety posture shift predictably across multi-turn conversations

Published: Zenodo | GitHub
Full dataset and 190+ page transcript available upon request
https://doi.org/10.5281/zenodo.17749013

Project 3:

Adaptive Wisdom: A Theoretical Model for AI's Evolution Beyond Knowledge

Act 3: The Future

Current AI systems move between information and knowledge. The next frontier is adaptive wisdom, contextually sensitive, emotionally intelligent judgement that responds to human needs across a conversation arc, not just individual prompts.

This study proposes a theoretical framework for this progression, grounded in observable behavioural patterns documented in the LLM Behavioural Study, and introduces a probability-based emotional layering model for developing genuinely human-responsive AI systems.

Status: In Development

 This work is independent, unfunded, and driven purely by the belief that AI must remain human at its core.

If something here resonates  or needs challenging ,  I would genuinely like to hear from you.

mankajsingh@gmail.com