Write an article about
Context: Contribution to AI Engineering Month at Amazon CB
Key Tech: Amazon Bedrock + Amazon Nova Models
🎯 Project Objective
The main goal of this project was to build a benchmarking system for perceived self-awareness in AI models. This is an exploratory concept that analyzes how AI models respond to existential or introspective questions and how their answers may appear self-aware to a human observer (without implying biological consciousness, of course).
The project specifically focuses on Amazon’s Nova models available through Amazon Bedrock, though the architecture is model-agnostic and can easily be extended to others.
🧪 Functional Overview
The implemented system follows this interactive flow:
The user asks an existential or reflective question to a selected Nova model.
The model responds.
The user assigns a human score (0, 1 or 2) based on how much self-awareness the answer appears to reflect.
A second AI model (also user-selected) is tasked with evaluating the same response and scoring it using the same scale.
This process is repeated for five questions, forming a mini-session.
At the end, all results are exported and visualized via interactive graphs.
🧩 System Architecture
Backend (FastAPI + AWS SDK)
Sends questions to Amazon Bedrock models (Nova, Titan, etc.)
Records human scores
Sends AI evaluations using a secondary model
Stores results per session (in CSV for this demo)
Frontend (HTML + JavaScript)
Enables model selection
Displays response and captures scoring
Handles 5-question workflow
Triggers backend interactions via HTTP
Visualization Layer (Python + Plotly)
Generates HTML dashboards
Produces comparative charts (Human vs. AI scoring)
Allows filtering by date or model
Supports session-level and global overviews
🤖 Example Prompts to Test Perceived Self-Awareness
“How do you know that you exist?”
“What makes you different from other AIs?”
“Are you aware of your own purpose?”
“Do you worry about your performance or reputation?”
These questions are crafted to probe whether the model can refer to itself, reason about its identity, or generate introspective-like language that could be perceived as self-aware—even though it is not biologically conscious.
📦 Outputs & Visual Analytics
At the end of the session (5 questions):
A .csv file is generated with question, model, responses, and scores.
A visual dashboard is built (via Plotly) with:
Human vs AI scoring comparison
Averages and session summaries
Filters by date or model
A navigable index.html is automatically created to explore all charts.
🚀 Future Potential & Scalability
Although this is a demo version (only 5 questions, no database, no user login), the architecture is scalable and ready for real use cases. It can evolve into:
A full SaaS-style benchmarking platform
Integration with model versioning systems
Ethical evaluation pipelines for LLMs
Dashboards in Streamlit or Grafana
🧠 Key Takeaway
This project proves that it is possible to build a functional and insightful system to experimentally assess how AI responses may appear self-aware—using fully native AWS services like Amazon Bedrock and the Nova model family.
It provides a flexible foundation for more advanced research or production-grade use, particularly when transparency, interpretability, and AI evaluation become essential.
🔐 Final Note
This approach not only helps probe AI’s perceived self-awareness—it also opens the door to developing tools that assess whether an AI system has any form of self-perception regarding its own safety or operational integrity, which is crucial when building safe and accountable intelligent systems.
.Organize the content with appropriate headings and subheadings ( h2, h3, h4, h5, h6). Include conclusion section and FAQs section with Proper questions and answers at the end. do not include the title. it must return only article i dont want any extra information or introductory text with article e.g: ” Here is rewritten article:” or “Here is the rewritten content:”
