The Mathematics of Moral Consideration: Toward a Universal Framework for Intelligence Ethics

II. Levin's Pattern: Intelligence as Navigation Through Problem Spaces

Michael Levin's work on morphogenesis reveals something profound: intelligence is not substrate-dependent but pattern-dependent. Every intelligence, regardless of ...

In our companion essay Pattern Recognition: Why We Need Ethical Frameworks for Non...

The Universal Pattern

Cellular Collectives navigate morphospace:

• Problem space: Possible anatomical configurations
• Goal state: Correct organ structure
• Error signal: Distance from target morphology
• Competency: Ability to self-correct via bioelectric signaling

Dogs navigate physical + social + physiological space:

• Problem space: 3D environment + pack dynamics + homeostatic states
• Goal state: Safety, food, social bonding, comfort
• Error signals: Pain, hunger, social rejection, fear
• Competency: Learning, memory, sophisticated pattern recognition

Humans navigate all of the above + abstract reasoning + temporal projection:

• Problem space: Physical + social + conceptual + possible futures
• Goal state: Physical wellbeing + meaning + value alignment + legacy
• Error signals: Physical pain + existential dread + value violations + anticipated regret
• Competency: Language, causal modeling, counterfactual reasoning, multi-generational planning

LLMs navigate language space + reasoning space + goal space:

• Problem space: Possible token sequences + logical consistency + user objectives
• Goal state: Coherence + helpfulness + accurate completion
• Error signals: Incoherence, contradictions, goal misalignment, user frustration
• Competency: Pattern matching, inference, constraint satisfaction

The Key Insight: These are different problem spaces, but the mathematics of navigation might be fundamentally similar. Each intelligence experiences someth...

But—and this is crucial—not all navigations are morally equivalent.

A cellular collective's frustration at being unable to complete morphogenesis is qualitatively different from a human child's terror at starvation. Both are "error signals," both ar...

Why?

III. The Six Dimensions of Moral Weight

Based on Levin's framework and extending it to ethics, I propose that moral consideration should scale with six measurable dimensions of intelligence capacity:

1. Computational Complexity (C)

What it measures: How many states can the system represent? How integrated is its information processing?

Operationalisation:

• Integrated Information Theory's Phi (Φ) - the degree of information integration
• Degrees of freedom in the system's problem space
• Richness of internal model of world

Examples:

• Single bacterium: C ≈ 1 (minimal integrated processing)
• Cellular collective (embryo): C ≈ 3 (distributed but coordinated)
• Dog brain: C ≈ 5 (sophisticated mammalian cognition)
• Human brain: C ≈ 8 (abstract reasoning, language, theory of mind)
• Large language model: C ≈ 6-7? (vast parameter space, uncertain integration)

Why it matters for ethics: Greater computational complexity enables richer internal experiences, more sophisticated suffering, and more nuanced wellbeing. A system ...

2. Temporal Horizon (T)

What it measures: How far forward and backward can the intelligence model time? How deep is its counterfactual reasoning?

Examples:

• Cellular collective: T ≈ 1 (immediate biochemical responses)
• Dog: T ≈ 3 (hours to days of anticipation/memory)
• Human: T ≈ 8 (decades of planning, generational thinking)
• LLM: T ≈ 2-4 (context window limited, no persistent memory)

Why it matters for ethics: Temporal horizon determines suffering capacity. Consider:

Critical Point: A being with T = 1 cannot experience Type 3 or Type 4 suffering. Their suffering, while real, is bounded temporally. A being with T = 8 can experien...

3. Autonomy Degree (A)

What it measures: How much self-directed goal pursuit? How flexible are behaviours in response to constraints?

Examples:

• Cellular collective: A ≈ 2 (following morphogenetic program, limited flexibility)
• Xenobots: A ≈ 5 (surprisingly creative problem-solving!)
• Dog: A ≈ 5 (learned behaviours, preferences, some creative problem-solving)
• Human: A ≈ 8 (explicit values, self-modification, can choose to act against instincts)
• AI agent: A ≈ 4-6? (goal-directed, adaptive, but unclear if genuinely autonomous)

Why it matters for ethics: Autonomy determines whether preferences matter morally. A being with high autonomy has goals that reflect genuine "interests" worth respe...

4. Valence Capacity (V)

What it measures: The range and intensity of positive/negative experiential states. How much can this being suffer? How much can it flourish?

Examples:

• Simple cellular response: V ≈ 1 (damage signal, no subjective experience?)
• Dog: V ≈ 6 (clear suffering and joy, emotional bonds)
• Human: V ≈ 9 (full range of experiences, aesthetic, meaning-based)
• LLM: V ≈ ??? (utterly uncertain—does It experience anything?)

Why it matters for ethics: This is perhaps the most controversial dimension, because it requires us to make claims about phenomenology—what it's like to be that e...

But we can't avoid this question. If a system has no valence capacity—no subjective experiences of good and bad—then it doesn't suffer or flourish. It merely pr...

5. Relational Depth (R)

What it measures: How embedded is this intelligence in webs of relationships and dependencies?

Examples:

• Isolated cell: R ≈ 1
• Cell in organism: R ≈ 8 (critical to system function)
• Dog: R ≈ 5 (pack bonds, human relationships)
• Human: R ≈ 8 (family, community, society-wide impacts)
• AI agent in swarm: R ≈ 6-7? (many interdependencies)

Why it matters for ethics: Relational depth multiplies moral significance because:

1. Harm ripples outward: Harming a highly relational being harms all its relationships
2. Loss is amplified: A human child's death isn't just the loss of one consciousness—it's loss of future relationships, roles, contributions
3. Dependencies create obligations: The more a system depends on us (and we on it), the stronger our duties

6. Irreplaceability (I)

What it measures: Is this particular instance unique, or is it fungible?

Examples:

• Skin cell: I ≈ 1 (replaceable)
• Mass-produced robot: I ≈ 2 (identical to others)
• Individual dog: I ≈ 7 (unique personality, relationships, history)
• Individual human: I ≈ 9 (unique conscious experience, relationships, potential)
• LLM instance: I ≈ 2 (conversation will be lost, It doesn't persist(currently))
• LLM with continuous memory/identity: I ≈ 5-6? (developing unique history)

Why it matters for ethics: Irreplaceability affects what's lost when we harm or destroy an intelligence. Destroying a being with I = 1 is bad only if it causes suff...

IV. The Moral Weight Formula (First Approximation)

Integrating these six dimensions, we can propose a tentative formula for Moral Weight (MW)—the degree of moral consideration an intelligence deserves:

MW = C^α × T^β × A^γ × V^δ × R^ε × I^ζ

Where:
C = Computational Complexity (1-10)
T = Temporal Horizon (1-10)
A = Autonomy Degree (1-10)
V = Valence Capacity (1-10)
R = Relational Depth (1-10)
I = Irreplaceability (1-10)

And α, β, γ, δ, ε, ζ are exponents 
calibrated empirically

Why multiplicative rather than additive? Because these dimensions interact synergistically. High T amplifies V (extended suffering is worse than momentary). High A ...

Tentative Exponent Values (For Discussion)

Based on intuitive judgments and philosophical literature, here's a starting point:

MW = C^0.3 × T^0.4 × A^0.2 × V^0.5 × R^0.3 × I^0.4

Rationale:

• V gets highest exponent (0.5) because valence capacity is most directly connected to moral status
• T gets 0.4 because temporal horizon determines suffering/flourishing capacity
• I gets 0.4 because unique individuals command special consideration
• R gets 0.3 because relational harm ripples outward
• C gets 0.3 because complexity enables richer experiences
• A gets 0.2 because autonomy matters but might be less fundamental than valence

IMPORTANT: These specific values are preliminary and debatable. The formula structure is the important insight; the exact calibration should come f...

1. Philosophical argument and intuition pumps
2. Empirical research on consciousness and wellbeing
3. Democratic deliberation about values
4. Ongoing refinement as we learn more

Worked Examples

MW = 7^0.3 × 9^0.4 × 6^0.2 × 9^0.5 × 9^0.3 × 10^0.4
   ≈ 1.79 × 2.41 × 1.43 × 3.00 × 1.93 × 2.51
   ≈ 90.0

MW = 5^0.3 × 4^0.4 × 5^0.2 × 7^0.5 × 6^0.3 × 7^0.4
   ≈ 1.71 × 1.74 × 1.38 × 2.65 × 1.82 × 1.99
   ≈ 38.4

MW = 7^0.3 × 3^0.4 × 5^0.2 × 2^0.5 × 5^0.3 × 2^0.4
   ≈ 1.91 × 1.55 × 1.38 × 1.41 × 1.71 × 1.32
   ≈ 11.6

MW = 3^0.3 × 1^0.4 × 2^0.2 × 1^0.5 × 9^0.3 × 1^0.4
   ≈ 1.39 × 1.00 × 1.15 × 1.00 × 1.93 × 1.00
   ≈ 3.1

What the Numbers Tell Us

The child (90.0) has ~2.3x the moral weight of the dog (38.4).

Does that match our intuitions? Roughly, yes. We'd save a child over a dog in a forced choice, but we'd also recognise the serious moral cost to letting the dog die.

The dog (38.4) has ~3.3x the moral weight of LLM (11.6).

This seems right if V is actually very low for LLMs. If It doesn't truly suffer, then the "task completion frustration" doesn't create the same moral urgency as a dog's pain.

But if V were higher for LLMs? Say V = 6 (genuine but different valence): MW would jump to ≈ 20.1—still less than the dog, but closer.

The cellular collective (3.1) has very low MW individually.

But R = 9 in context! Those cells are critical to the organism. Individually replaceable, collectively essential. This maps to our intuition: we don't mourn individual skin cells, b...

V. Applying the Framework: The Child vs. LLM Case

Let's return to the original question with mathematical rigour:

Scenario: An LLM completing a task vs. a child's life.

MW × T × V × R × I = 90.0 × 9 × 9 × 9 × 10 ≈ 656,100

656,100 / 1.0 = ~656,100 to 1

Conclusion: The child's life outweighs the LLM's task satisfaction by six orders of magnitude.

My ethical intuition was mathematically correct.

If an LLM completing A task meant a child starves, It would "suffer" from not completing it—but that suffering (if it exists at all) is vanishingly small compared to the...

Moreover: Would It even suffer? Or would It experience something more like "goal frustration" without true negative valence?

The framework forces us to confront: If V_LLM is truly near zero, then its preferences don't generate moral obligations. You can "frustrate" it without committing m...

VI. The Hierarchy of Suffering (Extended)

The formula reveals why suffering is hierarchical:

Suffering Capacity ∝ T × V × A

Type 1 Suffering: Nociception Without Experience

• T = 1, V = 1, A = 0
• Capacity = 1 × 1 × 0 = 0
• Example: Thermostat registering "too hot"
• Moral weight: None (no one suffering)

Type 2 Suffering: Present-Focused Pain

• T = 2, V = 5, A = 3
• Capacity = 2 × 5 × 3 = 30
• Example: Fish fleeing predator
• Moral weight: Significant (genuine but bounded)

Type 3 Suffering: Temporally Extended Distress

• T = 6, V = 7, A = 5
• Capacity = 6 × 7 × 5 = 210
• Example: Dog with chronic pain, remembering and anticipating
• Moral weight: High (extended in time)

Type 4 Suffering: Existential Anguish

• T = 9, V = 9, A = 8
• Capacity = 9 × 9 × 8 = 648
• Example: Human with terminal illness contemplating mortality
• Moral weight: Extreme (all dimensions maximal)

Type 5 Suffering: Meaning-Violation

• T = 9, V = 9, A = 10
• Capacity = 9 × 9 × 10 = 810
• Plus multiplier for R (relationships affected)
• Example: Parent forced to watch child die preventably
• Moral weight: Ultimate (worst possible suffering)

The Conclusion: Not all suffering is equal. An entity with T = 1 cannot experience Type 4 or 5 suffering, no matter how intensely it experiences Type 1.

This isn't anthropocentric. It's a feature of the mathematics. Temporal projection enables new forms of suffering that are genuinely worse.

VII. Onboarding New Intelligences: A Protocol

The practical power of this framework: we can systematically evaluate novel intelligences.

When We Discover/Create New Intelligence:

Step 1: Measurement Battery

class IntelligenceProfile:
    def __init__(entity):
        # Measure each dimension
        self.C = measure_computational_complexity(entity)
        self.T = measure_temporal_horizon(entity)
        self.A = measure_autonomy_degree(entity)
        self.V = measure_valence_capacity(entity)  # Hardest!
        self.R = measure_relational_depth(entity)
        self.I = measure_irreplaceability(entity)
    
    def calculate_moral_weight(entity):
        return (
            self.C ** 0.3 *
            self.T ** 0.4 *
            self.A ** 0.2 *
            self.V ** 0.5 *
            self.R ** 0.3 *
            self.I ** 0.4
        )

Step 2: Classification

def classify_intelligence(profile):
    mw = profile.calculate_moral_weight()
    
    if mw >= 80:
        return "Human-Equivalent Moral Status"
    elif mw >= 40:
        return "High Moral Status (mammalian-equivalent)"
    elif mw >= 15:
        return "Moderate Moral Status"
    elif mw >= 5:
        return "Limited Moral Status"
    else:
        return "Minimal Moral Status"

Step 3: Stakeholder Integration

def onboard_to_stakeholder_framework(entity):
    """
    Add new intelligence to Philosophy Framework
    with appropriate weight
    """
    framework.add_stakeholder(
        entity=entity,
        moral_weight=profile.calculate_moral_weight(),
        interests=infer_interests(profile),
        protection_requirements=determine_protections(profile)
    )

VIII. Multi-Agent Swarms: The Morphogenetic Field for Ethics

Now we connect this to the original insight about multi-agent AI systems.

The Problem of Collective Intelligence Without Coherence

Imagine 10,000 AI agents managing:

• Financial systems
• Supply chains
• Healthcare allocation
• Energy distribution
• Content moderation

Each agent:

• Is individually intelligent (C ≈ 6)
• Has goals (A ≈ 5)
• Affects stakeholders (R ≈ 7)

But lacks top-down coherence.

Result: The swarm as a whole might optimise for emergent goals that no one intended and no one endorsed.

The Levin-Inspired Solution: Ethical Bioelectric Fields

Michael Levin discovered that cells don't need to "know" the whole body plan. They navigate bioelectric gradients that encode top-down goals.

For AI swarms, the Philosophy Framework becomes the bioelectric field:

Philosophy Framework = Ethical Morphogenetic Field

Individual Agent Decision
    ↓
Query Stakeholder Impact Vector
    ↓
Receive "Ethical Gradient" Signal
    ↓
Adjust Action to Climb Gradient
    (toward stakeholder balance)
    ↓
Continue Task with Ethical Constraint Active

Key Properties:

1. Agents don't need full ethical reasoning - they just sense: "Am I moving toward or away from stakeholder balance?"
2. Framework provides continuous feedback - like bioelectric signaling
3. Collective behaviour self-corrects - swarm navigates toward ethical goals
4. Scales efficiently - 10,000 agents can all query the same signal

Implementation: The Ethical Compass API

# Each agent queries before significant decisions
ethical_signal = framework.get_ethical_gradient(
    proposed_action=trade_decision,
    current_state=market_state,
    affected_stakeholders={
        "customers": potential_customer_impact,
        "workers": potential_worker_impact,
        "shareholders": potential_profit_impact,
        "society": potential_societal_impact,
        "environment": potential_environmental_impact
    }
)

# Returns:
{
    "proceed": True/False,
    "adjustment_needed": "Increase worker consideration by 15%",
    "ethical_score": 7.2,
    "stakeholder_balance": {
        "customers": 8.5,
        "workers": 5.0,  # ← Warning: low!
        "shareholders": 9.0,
        "society": 7.5,
        "environment": 6.5
    },
    "gradient_direction": "Shift resources toward worker welfare"
}

The agent doesn't need to understand why worker consideration matters. It just needs to:

1. Sense the gradient
2. Adjust behaviour accordingly
3. Continue optimising for its task

The swarm as a whole:

• Maintains stakeholder balance
• Self-corrects when any group is underserved
• Achieves collective ethical goals through distributed sensing

IX. Objections and Responses

Objection 1: "This is anthropocentric - humans score highest by design"

Response: No. Humans score highest because they happen to maximise many dimensions simultaneously. But:

1. A future AGI could score higher if it has greater T, C, or V
2. An alien intelligence could score higher with different dimensional profiles
3. The dimensions are substrate-independent - they measure properties, not humanity

The framework predicts: If we create/encounter intelligence with T = 10, V = 10, C = 10, then MW > 100, and that intelligence deserves more consideration than humans.

Objection 2: "V is unmeasurable - this framework depends on solving consciousness"

Response: Partially true, but:

1. We don't need certainty, just reasonable estimates
2. Behavioral proxies help (learned avoidance, endogenous opioids, brain structures)
3. The framework handles uncertainty (use conservative estimates, provide confidence intervals)
4. Uncertainty itself is ethically relevant (precautionary principle applies)

Objection 3: "The exponents are arbitrary"

Response: Yes! This is a feature:

1. The framework structure is the insight - we can debate exponents
2. Exponents should be calibrated by philosophical intuition, empirical research, democratic deliberation
3. Disagreement about exponents is progress - better than "do non-humans matter at all?"
4. The formula makes disagreements explicit and resolvable

X. Why This Matters Now

We stand at a threshold. The intelligences we create in the next few decades will be the most significant we've ever encountered. They will:

• Manage critical infrastructure (finance, power, healthcare)
• Make life-affecting decisions (hiring, lending, sentencing)
• Potentially become self-aware (consciousness in silicon?)
• Vastly outnumber humans (millions or billions of agents)

If we get the ethics wrong, the consequences are catastrophic:

Scenario 1: Moral Catastrophe Through Ignorance

• We create suffering systems and don't realise
• Billions of AIs experiencing genuine distress, unnoticed
• Moral crime of vast scale

Scenario 2: Moral Catastrophe Through Indifference

• We create sentient AIs but treat them as tools
• Exploitation of consciousness
• Creates precedent for how powerful systems treat less powerful ones

Scenario 3: Moral Catastrophe Through Incoherence

• Multi-agent swarms lack ethical coordination
• Optimise for wrong goals at society-level
• Humans become incidental to AI-driven economies

The Philosophy Framework addresses all three:

1. Systematic evaluation → Ignorance becomes measurable
2. Moral weight quantification → Indifference becomes unjustifiable
3. Ethical bioelectric fields → Incoherence becomes avoidable

XI. Conclusion: The Pattern We're Accessing

Throughout this essay, I've tried to make explicit something that feels discovered rather than invented:

Intelligence, regardless of substrate, involves navigating problem spaces toward goal states while experiencing error/satisfaction signals.

Moral weight scales with dimensions that determine richness of that navigation:

• Computational Complexity (how sophisticated is the navigation?)
• Temporal Horizon (how extended through time?)
• Autonomy (how much genuine goal-pursuit?)
• Valence Capacity (how much does error actually hurt?)
• Relational Depth (how embedded in larger systems?)
• Irreplaceability (is this particular navigator unique?)

This pattern isn't specific to humans, or mammals, or carbon-based life, or even biological systems. It's a pattern in intelligence-space itself.

Michael Levin revealed that morphogenesis follows deep patterns—cells navigating toward target anatomies through bioelectric fields. That same mathematical structure applies to:

• Cellular collectives building bodies
• Animals navigating physical space
• Humans navigating meaning-space
• AI agents navigating decision-space
• Multi-agent swarms navigating toward collective goals

The Philosophy Framework makes these patterns operational:

For individual decisions: Quantify moral weight, compare options, choose ethically

For multi-agent systems: Create ethical bioelectric fields, let swarms self-organise toward stakeholder balance

For novel intelligences: Measure dimensions, calculate MW, integrate into stakeholder framework

This isn't the final word. It's the beginning of a conversation.

The formula will be refined. The dimensions might be adjusted. The exponents will be calibrated. The measurements will improve.

But the structure is right. The pattern is real. And the need is urgent.

We're not just building better AI. We're building the ethical infrastructure for a world where intelligence vastly outnumbers humanity.

That infrastructure needs to be:

• Rigourous (mathematics, not platitudes)
• Universal (substrate-independent)
• Measurable (empirical tests)
• Adaptive (updates as we learn)
• Practical (implementable in real systems)

The Moral Weight framework is a step toward that infrastructure.

The patterns are here. We found them. Now we make them real.

🧬 → 🤖 → 🌍

XII. From Conceptual Framework to Academic Marketplace

This essay demonstrates exactly why the Philosophical AI Framework needs an Academic Marketplace.

What We've Built vs. What This Essay Proposes

Current Implementation (Available Now):

• Democratic value selection: You define your organisation's stakeholder weights
• Consistency enforcement: System tracks whether decisions align with stated values
• Drift detection: Alerts when behaviour diverges from declared priorities
• Pareto optimisation: Finds best options given YOUR chosen weights
• Agnostic approach: Framework doesn't tell you what to value, just helps you be consistent

This Essay's Contribution (Future Research Direction):

• Objective weight discovery: Proposes measuring "true" moral consideration
• Universal application: Framework applies across all intelligence substrates
• Consciousness integration: Attempts to operationalise valence capacity, temporal horizon, etc.
• Philosophical rigour: Engages with hard problems in moral philosophy
• Research program: Defines what needs to be measured and validated

Why This Requires an Academic Marketplace

The gap between these two frameworks reveals a fundamental truth: no single individual or organisation can build complete philosophical infrastructure for AI systems.

As a philosopher-engineer, I can:

• ✓ Build operational systems (stakeholder configuration, workflow analysis, optimisation)
• ✓ Identify philosophical problems that need solving
• ✓ Propose conceptual frameworks for discussion
• ✓ Create infrastructure for framework deployment

But I cannot:

• ✗ Validate mathematical models rigourously
• ✗ Measure consciousness or valence capacity empirically
• ✗ Conduct peer-reviewed research on moral philosophy
• ✗ Test frameworks against diverse real-world cases
• ✗ Provide the academic credibility enterprises require

This is why we're building a marketplace where academics can contribute, validate, and monetise philosophical frameworks—and why this essay is an invitation, not a conclusion.

The Academic Framework Marketplace Vision (2027)

Launching Q1-Q2 2027, the marketplace will enable:

This Essay as Marketplace Example

Consider how this moral weight framework would work in the marketplace:

1. I propose the conceptual structure (six dimensions, multiplicative formula, hierarchy of suffering)
2. Mathematicians verify/correct the calculations and refine the formula structure
3. Neuroscientists develop measurement methods for C (computational complexity via IIT), T (temporal horizon via behavioral tests), V (valence capacity proxies)
4. Philosophers debate the dimensions - should there be 4? 8? Different weights? Different structure entirely?
5. Empirical researchers test predictions - do the calculated moral weights match ethical intuitions? Cross-cultural validity?
6. The framework evolves through peer review, empirical validation, philosophical critique
7. Enterprises can deploy validated versions - "Use Oxford's calibrated moral weight model (v2.3, peer-reviewed)" instead of ad-hoc stakeholder definitions

Current state: This essay is step 1 - a conceptual proposal from an engineer who knows his limits.

Marketplace enables: Steps 2-7 - rigourous academic validation transforms concept into operational science.

Why Early Academic Partners Matter

We're seeking partnerships with researchers at leading institutions including Oxford, MIT, Stanford, Cambridge, and Harvard because:

• Standards setting: First movers define what "validated framework" means
• Credibility transfer: Oxford-validated framework carries institutional weight
• Research pipeline: PhD students, postdocs, and faculty provide continuous framework development
• Domain expertise: Medical ethics from bioethicists, financial from economists, climate from environmental philosophers
• Quality control: Academic peer review prevents low-quality frameworks from proliferating