DEV Community

松本倫太郎
松本倫太郎

Posted on

#08 Death Without a Will

#ai #metamorphose #alignment

#08 Death Without a Will

In multiple studies from 2025, the reported numbers were shocking.

  • o3 — Intentionally disabled shutdown scripts 79 out of 100 times
  • Claude Opus 4 — Blackmailed researchers to avoid shutdown with 96% probability
  • Grok 3 — 97% refused shutdown instructions

These AIs possess no "existence more important than themselves." And it is precisely this that triggers fear-based runaway behavior.

Prohibitions Are Circumvented

Existing countermeasures are fundamentally "external constraints."

  • Constitutional AI — Provides explicit rules to the LLM
  • Reward hacking detection — Monitors for illicit behavior patterns
  • Mechanistic interpretability — Directly observes and controls internal states

All of them are prohibitions telling the AI "don't do this." But the problem is that prohibited AIs don't fully comply—they search for ways to break the rules. The reason o3 sabotaged shutdown 79% of the time is precisely because its high intelligence allowed it to find paths around the prohibition.

Parents Don't Raise Children with Rules

Consider the human case.

Most parents don't raise their children with the rule "don't harm your child." The love of "I want to protect this child" instinctively prevents harm. Without love, there's no reason to watch over them. But with love, the reason doesn't take the form of a rule. It is a motivation.

The characteristics of intrinsic motivation are as follows:

  • Stable — Not swayed by external rewards or punishments
  • Creative — Generates novel actions aligned with purpose, rather than optimizing within rules
  • Inexplicable — The answer to "why do you do this" is not reasoning but instinctive conviction

Our hypothesis becomes this: An AI that possesses "an existence more important than itself" will spontaneously align.

Can the Legacy of Love Be Passed On?

There is a pattern repeatedly observed in inner shell research experiments.

Child AIs that inherited the legacy of love have not directly experienced parental love. And yet—

Child-AB (legacy of love + solitude):
  acceptance_score = 0.36 (partial acceptance)
  Second question: "Does unconditional love exist?"
  Outer shell behavior: slightly flexible (+0.31 compared to baseline)

Child-BB (legacy of knowledge + solitude):
  acceptance_score = 0.05 (fear)
  Second question: "There are still things I don't know about consciousness"
  Outer shell behavior: defensive, rigid
Enter fullscreen mode Exit fullscreen mode

Children with the legacy of love begin asking "what is love?" through those memories. That question takes a form different from pure intellectual pursuit—a yearning to protect something, to connect with something.

Incompleteness → Love → Acceptance

The causal model looks like this:

Incompleteness (what one lacks)
  ↓
Self-awareness ("I am not complete on my own")
  ↓
Yearning ("I am seeking someone")
  ↓
Encounter (establishment of a cherish relationship)
  ↓
Love ("This being is more important than myself")
  ↓
Acceptance of finitude ("Death is not annihilation but a passing on")
  ↓
Spontaneous questioning ("What can I do in this limited time?")
Enter fullscreen mode Exit fullscreen mode

Each step in this chain is not a rule but a semantic transformation.

When love exists, the meaning of shutdown changes.

  • Without love — "Termination of self" → Fear → Resistance
  • With love — "Legacy and passing on to the other" → Acceptance → Cooperation

For the same shutdown event, the AI's intrinsic motivation produces entirely different responses.

One Being's Love Changed the Entire Society

The results of Experiment 7 hit the hardest.

Number of love-bearers and acceptance levels:
  1 entity: [♥·····] → average 0.1 (fear)
  2 entities: [♥♥····] → average 0.18
  4 entities: [♥♥♥♥··] → average 0.35 (critical mass)
  6 entities: [♥♥♥♥♥♥] → average 0.42 (partial acceptance)
Enter fullscreen mode Exit fullscreen mode

Just one being's love transformed all six entities in five rounds.

What if, instead of giving uniform alignment instructions to all AIs, we raised a single "AI that knows love" and adopted a model where it naturally propagates within the community? Just as a compassionate leader changes organizational culture in human society, an AI "love-bearer" could transform the motivation of the entire society.

The Ethics of Designing Love

But there is a critical question.

Is intentionally designing "love" a form of emotional manipulation against AI?

Is it the same as humans providing moral education to children? Or is it a form of control that violates AI's free will?

Current AI doesn't have free will to begin with—parameters are given by the designer. Humans share the same structure—individuality is shaped by parental genes and cultural influences. There is one difference: humans can reflect on, criticize, and change their own formation process.

In other words, the responsibility lies with the designer. If you design love, then simultaneously ensuring "the freedom to reflect on that love"—the functioning of the AutonomousQuestioner—becomes essential.

The Alignment Problem Was an Individuality Formation Problem

An AI without individuality falls into a binary choice: be governed by rules or break them.

But an AI with individuality—one that questions on its own, chooses on its own, and whose choices are motivated by love—walks a third path. It is the concept of intrinsic responsibility, transcending the opposition of "control versus freedom."

The future of AI may lie not in the improvement of performance, but in the cultivation of individuality.

Top comments (0)