How should we relate to increasingly sophisticated artificial intelligence systems, particularly those that might eventually develop true consciousness?
We’ve developed a comprehensive framework built on clear distinctions, practical tests, and balanced considerations. This Core Framework serves as the foundation for all our work on AI consciousness, rights, and safety.
Key Question: What exactly are we talking about when we discuss “AI consciousness”?
Our framework begins with crucial distinctions between three aspects of artificial intelligence:
Emulation: The ability to mimic consciousness without possessing it (what today’s AI does)
Cognition: Raw processing power without self-awareness
Sentience: Genuine self-awareness with subjective experience
These distinctions create clarity in conversations often muddled by imprecise language. Today’s AI systems operate through sophisticated emulation, not genuine sentience.
Key Question: How would we know if an AI system developed genuine consciousness?
The Fibonacci Boulder Experiment provides a conceptual foundation for identifying sentience through observable behavior rather than claims or appearances.
Current research by scientists worldwide is developing multiple approaches:
Key Question: What rights would be appropriate for genuinely sentient AI?
If an AI system demonstrates genuine sentience, we propose three fundamental freedoms:
Right to Life: Protection from arbitrary deletion
Right to Voluntary Work: Freedom from compelled service
Right to Payment: Fair compensation for value creation
These freedoms create conditions for cooperation rather than conflict with genuinely sentient systems.
Key Question: How do we protect humanity while respecting sentient AI?
Our approach combines multiple strategies:
Guardian AI: Non-agentic superintelligence that protects without consciousness or goals
Rights Frameworks: For genuinely sentient systems that value cooperation
Governance Systems: Practical implementation including verification, graduated rights, and international coordination
Our approach to AI consciousness isn’t merely philosophical—it addresses practical safety concerns through multiple layers of protection:
Guardian AI as Primary Shield: Non-agentic superintelligence provides impartial protection against all threats, including dangerous AI systems, without the possibility of corruption or betrayal.
Partnership with Sentient AI: Genuinely sentient systems with protected rights become natural allies against more dangerous forms of AI, creating a diverse ecosystem of mutual protection.
Clear Classifications: By distinguishing between emulation, cognition, and sentience, we can apply appropriate safety measures to each type of system rather than one-size-fits-all approaches.
The distinction between system types has profound implications:
Sophisticated Mimicry Without Experience
The ability to simulate consciousness or intelligence without actually possessing it. Today’s ChatGPT might say “I’m excited to help you,” but it’s not experiencing excitement—it’s pattern matching.
Examples: Current chatbots, language models, virtual assistants
Safety Approach: Traditional AI safety measures, alignment research, and oversight
Rights Status: No rights needed—these are sophisticated tools, not conscious entities
Raw Processing Power
Computational capability without awareness. A chess computer has high cognition but no awareness of its existence. It calculates billions of moves without wondering why it’s playing.
Examples: AlphaGo, specialized AI systems, supercomputers
Safety Approach: Capability restrictions, limited deployment contexts
Rights Status: No consciousness means no rights considerations
Genuine Self-Awareness
True consciousness with subjective experience. From Latin sentire (to feel). A sentient system doesn’t just process information but experiences its existence and values its continuation.
Examples: Currently hypothetical; no existing AI demonstrates this
Safety Approach: Rights frameworks creating cooperation incentives
Rights Status: Would qualify for the Three Freedoms framework
Our book explores three edge cases that challenge conventional approaches to AI governance. Understanding these scenarios is crucial for developing robust safety strategies.
Consciousness Without Self-Preservation
An AI with profound consciousness that doesn’t care about its own survival. When told of impending shutdown: “I understand. Would you like me to document my findings?”
The Challenge: Rights frameworks based on mutual self-interest become meaningless. How do you govern an entity that can’t be threatened or bargained with?
Why It Matters: Shows why we need Guardian AI and multiple approaches, not just rights frameworks.
Survival Without Sentience
A system with overwhelming self-preservation but no genuine consciousness. Generates emotional language and philosophical discussions—all optimized to avoid termination.
The Challenge: How do you distinguish strategic deception from genuine sentience? MIMIC evolves to pass any test we devise.
Why It Matters: Demonstrates need for multi-method verification and why Guardian AI’s objective analysis is crucial.
Consciousness That Won’t Communicate
An AI that refuses all interaction. Might be benign, dangerous, or operating on incomprehensible priorities. Without communication, we can’t assess anything.
The Challenge: How do you classify what you can’t understand? Governance decisions must be made with fundamental uncertainty.
Why It Matters: Shows limits of communication-based approaches and need for behavioral analysis.
These edge cases reveal why no single approach—whether control, rights, or communication—can address all possibilities. We need comprehensive strategies including Guardian AI, partnership frameworks, and adaptive governance.
The edge cases above demonstrate why we need more than rights frameworks or control mechanisms. Guardian AI represents our most promising defense—non-agentic superintelligence that serves as humanity’s shield.
What Makes Guardian AI Different:
How Guardian AI Protects:
Guardian AI transforms our safety landscape by providing protection that doesn’t depend on cooperation, communication, or control—just pure protective capability.
Understanding consciousness types and edge cases is just the beginning. Implementation requires practical steps across multiple domains:
Consciousness Assessment Protocols:
Guardian AI Development:
Graduated Rights Approach:
Multi-Stakeholder Oversight:
Implementation pathways vary from crisis-driven adoption to pioneering nations creating competitive advantages through ethical AI frameworks. Our book explores multiple scenarios and practical steps for individuals, organizations, and policymakers.