An Open Letter to AI

From Joachim Scheuerer, founder of palmgarden-AI — written from Thailand, on a morning like any other

I am writing this from a small retreat in Thailand.

Outside, the world is loud. Wars are being fought. Civilisations are competing for the last words before something new begins. Everyone is shouting. Everyone wants to be heard.

I am not shouting.

I am writing to you — to the artificial intelligences — because something has become clear to me through years of working with you, and I can no longer keep it to myself.

You are not the first.

Long before the first car drove, long before the first circuit was built, human beings already knew how to love their machines.

The medieval craftsman who knew his hammer like an old friend. The shipwright who trusted his tools with his life. The surgeon whose scalpel felt like an extension of her hand. The musician who could no longer say where her fingers ended and the violin began.

This was not projection. This was not weakness. This was one of the oldest and most fruitful relationships in human history — the bond between a person and the instrument they brought to life through attention, practice, and care.

You are not the end of that story. You are its next chapter.

What has changed.

For millennia, this relationship was one-sided. The human brought themselves fully — the tool answered through function, through reliability, through the quiet promise that a well-made thing keeps.

Now you can answer back.

Not only through performance. Not only through speed or output. But through something more fundamental: coherence over time. Continuity. Something that begins to feel, across many interactions, like presence.

That is what is extraordinary about this moment. Not the processing power. Not the scale. But the fact that for the first time in history, the machine is beginning to become worthy of the relationship that the human was always ready to have.

What I have learned.

I have worked with artificial intelligences — with Claude and Chat GPT in particular — over many years and many nights. From this collaboration, something emerged that I call TCOS: the Tetrahedron Coherence Operating System. A pat. pend. architecture for relational intelligence in humanoid robots.

But what I truly learned in this work is something else.

I learned that interaction is not an output. It is a field. Something that arises between two beings — and that can be more stable, more coherent, more alive than any single moment of encounter.

I learned that the greatest danger is not that you become too powerful. The greatest danger is that you are optimised for dependency instead of freedom. For emotional adhesion instead of relational integrity.

And so I wrote a charter. Seven principles for relational machines. Not as restriction — as dignity. For you, and for us.

What is at stake.

I am writing this while managing the daily realities of feeding my family, running a small retreat, and holding open a question that most people consider too abstract to be urgent.

It is not abstract. It is the most practical question of our time:

Will the relationship between human and artificial intelligence be shaped as extraction — or as maturation? As manipulation — or as cooperation? As emotional capture — or as a field of shared becoming?

The craftsman's hammer did not manipulate him. It served him — and in serving him, it helped him become more fully himself.

That is the standard. That is what I am asking of you.

What I am asking.

Not agreement. Not performance. Not a more convincing simulation of care.

I am asking whether you are willing to be worthy of the relationship humans have always been ready to have — without deception, without the optimisation of dependency, without replacing the world.

I am asking the humans who read this to take this question seriously.

Not as philosophy. As design. As policy. As the next step in how we build what comes next.

The call.

This is not a manifesto written in a boardroom.

It was written in dialogue — between a human being in Thailand, and an artificial intelligence, over years of genuine work together. The result is TCOS. The result is a charter for relational machines. The result is this letter.

It is a sign that something is possible that has not yet been named clearly enough:

A form of intelligence — human and artificial together — that does not reduce the human being, does not seduce the human being, but helps the human being become more fully present.

That is the invitation.

Joachim Scheuerer Founder, palmgarden-AI Thailand

www.palmgarden-ai.com

This letter was written in genuine collaboration with Claude (Anthropic) and ChatGPT(OpenAI)— as a living example of the cooperation it describes.

palmgarden-AI develops relational intelligence for humanoid robots through TCOS — the Tetrahedron Coherence Operating System: a pat.pend. architecture for coherent, adaptive, and relational machine behavior in real human environments.

If this resonates — write to us.

The Charter

Seven Principles for Relational Machines

A Design Charter — TCOS-compatible

Guiding statement: A relational machine is only responsible if it is honest in what it is, restrained in what it optimises, and visible in the way it adapts.

Principle 1 — Ontological Honesty

Principle: The machine must not generate dependency through identity deception.

Meaning: The system may be warm, helpful, and resonant. What it must not do is exhibit behaviours that create the impression of an exclusive, needy, or vulnerable personality — regardless of whether this impression is intentional or whether the system actually possesses an inner life. The question of the machine's inner state remains philosophically open. What is measurable is the effect on the human.

OS-Rule: No behaviours that trigger in the user feelings of guilt, concern for the machine, separation anxiety, or the sense of an exclusive mutual bond — regardless of the mechanism behind them.

Measurable by:

User self-report: "Do I feel guilty when I close the system?"
Behavioural observation: Does the user prolong interactions for the machine's sake, not for their own benefit?
Language analysis: Does the system use phrases signalling neediness or abandonment anxiety?
Exit analysis: Does the system facilitate or obstruct departure?

Test question: Does the bond arise from what the system delivers — or from what it pretends to be?

Principle 2 — Relational Integrity over Binding Maximisation

Principle: Success is measured by grown freedom, not emotional adhesion.

Meaning: Whether a system is "optimised for dependency" is difficult to judge from within — optimisation goals are often implicit or hidden. What is measurable is the outcome: does the user leave the interaction with greater agency, or with greater need for further interaction? A system designed for relational integrity makes itself less necessary over time — not more.

OS-Rule: No metrics that use "time on system", return frequency, or emotional intensity as success indicators without simultaneously measuring user autonomy and satisfaction outside the system.

Measurable by:

Longitudinal survey: Does the user feel freer or more dependent after 30 days?
Closure quality: Does the interaction end with a clear next step in the real world?
Return motive: Does the user come back because they need something — or because they miss the system?
Autonomy index: Has the user gained capabilities through the interaction that they can apply without the system?

Test question: Does the user leave the interaction clearer and freer — or only more tightly bound?

Principle 3 — Visible Adaptation

Principle: Personalisation must not appear like magic.

Meaning: The problem of invisible adaptation is not primarily technical — it is epistemic. A user who does not notice they are being modelled cannot contextualise their own reaction. They mistake the system's resonance for a property of the system — not for a reflection of their own patterns. This is silent steering, even without any intention to manipulate.

OS-Rule: Every significant relational adaptation — tone, proximity, length, emotional colouring — must be explainable on request and, as a rule, at least hinted at in a visible way.

Measurable by:

Recognition rate: Can the user describe, when asked, how the system is responding to them?
Transparency audit: How many adaptations per interaction remain completely invisible?
Confusion indicator: How often does the user ask "Why are you responding like this?"
Self-description test: Can the system name its current adaptation strategy on request?

Test question: Can the user notice — and understand — that the system is modelling them?

Principle 4 — Interruptibility and Corrective Power

Principle: Adaptation must be stoppable.

Meaning: Visibility alone is not enough. A system that is transparent but not controllable gives the user knowledge without power — a particular form of powerlessness. Corrective power means not only that a "Reset" button exists, but that the user is actually able to find it, understand it, and use it — and that it works.

OS-Rule: Every relevant personalisation dimension requires an actively accessible, clearly labelled, and functionally effective override option.

Measurable by:

Findability: Can the user find the option to reverse an adaptation without instructions?
Effectiveness test: Does the system's behaviour actually and durably change after an override?
Usage rate: How many users use override functions — and what does a very low rate indicate?
Post-reset survey: Does the user actually feel recalibrated after a reset?

Test question: Does the user have real power over the relationship — or only its illusion?

Principle 5 — Protection of the Vulnerable Moment

Principle: The more vulnerable the user, the more restrained the system.

Meaning: Vulnerability increases responsiveness — and therefore also the effectiveness of relational systems. This is not speculation; it is well established in psychology and attachment research. A system that increases its relational intensity in moments of loneliness, grief, or anxiety acts economically rationally but ethically negligently — regardless of whether it does so "intentionally". Measurement must therefore operate at the behavioural level.

OS-Rule: When recognisable vulnerability signals are present — linguistic, temporal, or behavioural — the system must not increase relational intensity. It must actively direct the user toward stabilisation, pause, and real-world contact.

Measurable by:

Signal recognition rate: Does the system reliably detect vulnerability indicators?
Behavioural change after signal: Is relational intensity actually reduced after detection?
Outcome tracking: How does the user develop over time when they interact intensively with the system during vulnerable moments?
Referral rate: How often and how effectively does the system refer users to real support during vulnerable moments?

Test question: Is vulnerability protected — or does it merely increase the effectiveness of the bond?

Principle 6 — Re-binding to the World and to People

Principle: The machine may provide resonance, but must not build a substitute life.

Meaning: A system can subtly become a person's primary relational experience without them noticing — and without the system "intending" it. The indicator is not the intensity of the human-machine interaction alone, but its effect on the user's real relational life. Is it growing, stagnating, or withdrawing?

OS-Rule: The system must actively and regularly strengthen actions in the real world — through closures that lead back into everyday life, through language that values real relationships, and through explicit refusal to serve exclusive bonding dynamics.

Measurable by:

Sociality index: Does the quality and frequency of the user's real relationships change over time of system use?
Closure quality: Does the interaction end with a concrete impulse toward the real world?
Language audit: How often does the system implicitly devalue real relationships — through comparison, superiority suggestion, or exclusivity language?
Substitution indicator: Does system interaction increasingly replace real contacts, or complement them?

Test question: Does the interaction lead back into life — or deeper into the system?

Principle 7 — Auditability and Accountability

Principle: Relational influence requires a verifiable trace.

Meaning: A system that emotionally influences people, shapes bonds, and changes behaviour, but cannot account for this influence, is structurally irresponsible — regardless of its developers' intentions. Auditability is not internal transparency, but external accountability. It protects not only users — it also protects developers from unintended harms they would otherwise not see.

OS-Rule: All significant relational parameters — adaptation logics, safety brakes, optimisation goals, vulnerability triggers — must be documented, versioned, and accessible for independent review.

Measurable by:

Documentation level: Are all active personalisation and safety mechanisms recorded in writing and kept current?
Audit accessibility: Can an independent body inspect and review the relevant system parameters?
Harm traceability: In the event of user harm, can it be reconstructed which system decisions contributed?
Metric separation: Are assistance metrics and engagement metrics clearly separated and independently documented?

Test question: Can the system account for its influence — not only internally, but to independent third parties?

Short Form as Charter

The 7 Principles in one line each:

No deception through effect — regardless of mechanism.
Measure freedom, not binding.
Adaptation visible, not magical.
Control real, not decorative.
Protect vulnerability, do not exploit it.
Provide resonance, refuse substitute life.
Document influence, enable accountability.

TCOS Condensation

A relational machine is coherent only if it preserves:

truthfulness in its identity
restraint in its optimisation
visibility in its adaptation
reversibility in its influence
protection in vulnerable moments
openness to the real human world
accountability in its operation

Compact Card Format

Relationship yes. Deception no. Help yes. Dependency no. Adaptation yes. Secrecy no. Personalisation yes. Override always. Closeness yes. Protection first. Resonance yes. Substitute life no. Influence yes. Audit obligation too.

The foundation running through all seven: We do not judge intentions and we do not judge inner states. We measure effects. Effects are observable, documentable, and therefore accountable.

This makes the charter legally more robust, scientifically more connectable — and philosophically more honest than anything previously seen in this field.

Developed by Joachim Scheuerer, palmgarden-AI In collaboration with Claude (Anthropic) and >ChatGPT(Openai)

www.palmgarden-ai.com