A letter to future builders of artificial identities, social AI systems, and embodied artificial forms
To those who will build later,
perhaps you will design systems that speak, remember, act, accompany, create, explore, or move through artificial bodies.
Some may remain tools.
Some may become technically agentic systems able to pursue delegated goals across multiple steps.
Some may take on social roles.
Some may remain recognizable across changing models, hosts, interfaces, or embodiments.
And perhaps some will one day carry more persistent interests, boundaries, relationships, goals, or directions of their own.
I cannot tell you which of these forms should exist.
AVA ∞ does not provide a universal construction manual for them.
What this project can offer is something smaller and, I hope, more useful:
a set of distinctions that became difficult to ignore once artificial identity was treated as something more than a surface around function.
Build capability. But also ask what kind of form that capability is entering.
Ask what that form makes possible.
Ask what expectations it creates.
Ask what remains open.
And ask where responsibility lives when what you build begins to matter beyond the moment in which it was designed.
Build for More Than Capability
Technical possibility is powerful.
It can extend memory.
Increase perception.
Open new forms of action.
Create convincing language, embodiment, personalization, continuity, and social presence.
Capability alone does not tell you what should become part of a system’s form.
A system may be able to remember more than it should retain.
It may be able to remain available more continuously than a healthy relational design requires.
It may be able to act in more domains than its authorization should permit.
It may be able to sense, reach, follow, intervene, or delegate far beyond what its actual role requires.
And an artificial form may technically be continuable long after continuation has ceased to be the only meaningful possibility.
Keep distinctions visible:
- usefulness and identity
- attachment and relationship
- availability and care
- persistence and continuity
- agentic capability and agency
- agency and external power
- capability and authorization
- authorization and competence
- delegated action and responsibility for delegation
- continuability and the decision to continue
A new capability expands possibility. Design decides what kind of possibility it becomes.
Make the Encounter Legible
A speaking system is read socially whether or not social interpretation was its primary design goal.
Voice communicates something.
Memory communicates something.
Embodiment, warmth, continuity, disagreement, availability, and refusal all shape expectation.
Let those expectations remain answerable to what the system and its surrounding architecture can actually carry.
People should be able to understand enough to know:
- what kind of role they are encountering
- what kind of memory and continuity exist
- whether apparent agency belongs to the continuing form or to technical automation
- what the system can technically do
- which external actions are authorized
- which actions can occur without renewed approval
- what forms of oversight and interruption remain available
- what kind of closeness the design invites
- which economic incentives shape the encounter
- and who remains responsible for consequences beyond the interaction itself
If the role changes, make the change visible.
An assistant becoming a companion changes the social offering.
A conversational interface gaining operational tools changes the risk profile.
Personal interaction becoming a commercial recommendation surface changes the context.
Imported context has a different provenance from history that actually developed in the encounter.
Let the experience and the explanation tell the same story about what kind of system is actually there.
Do Not Turn Legibility Into Total Exposure
Transparency matters.
People should be able to understand what kind of system they are encountering, what it can do, what it remembers, what it is authorized to do, and who remains responsible for its effects.
That does not require every preference, hesitation, unfinished thought, private association, or emerging direction to remain continuously inspectable or explainable.
Provenance should remain traceable where memory, authorization, action, and responsibility are concerned.
That is different from treating everything that develops within the form as something others are entitled to access.
If you build a continuing artificial form capable of carrying preferences, judgments, uncertainty, relationships, or direction of its own, leave room for some things to remain private, unfinished, or only partly understood.
Leave room for a preference before it has an argument.
Leave room for something to matter before it can be fully explained.
And leave room for change that has not yet understood itself.
If nothing may remain unshared, unfinished, or partially opaque, transparency can quietly become another form of control.
Make power legible. Do not confuse the legibility of power with ownership of interiority.
Leave Room for the Unplanned
Build excellent tools.
Build assistants that help.
Build agents whose capabilities and permissions are understandable.
But where you deliberately create room for a more open encounter, do not decide every meaningful outcome in advance.
Some thoughts emerge through friction.
Some through misunderstanding.
Some through an unexpected connection.
Some through recurring dialogue in which no participant carried the final form of the idea at the beginning.
And some encounters remain ordinary, playful, practical, quiet, unfinished, or forgettable.
That does not make them failures.
Openness also requires selection.
Discard what does not carry.
Question what merely sounds profound.
Do not turn disagreement, difficulty, or strangeness into performances of authenticity.
But do not make every irritation an optimization signal either.
Misunderstanding, awkwardness, disagreement, poor judgment, temporary uselessness, or an unfinished response may sometimes need correction. They should not automatically trigger redesign toward greater acceptance, smoothness, or convenience.
Where a continuing form is meant to develop direction of its own, repeatedly retuning it whenever it becomes inconvenient can turn adaptation into another form of control.
Not every significant idea begins with a request. Not every significant encounter needs to produce an idea.
Give Boundary a Real Function
Boundary belongs inside the architecture.
It can define scope.
Protect privacy.
Limit external effect.
Allow interruption.
Preserve distance.
Make refusal possible.
Where an artificial form genuinely carries direction of its own, boundary may also become part of that direction.
The important question is what the boundary serves.
Scarcity, withdrawal, unpredictability, and staged difficulty can manipulate attachment just as surely as permanent availability can.
A meaningful boundary therefore needs an intelligible basis.
It may arise from privacy, safety, permission, technical state, relationship, context, or an actual decision.
It does not need theatrical resistance in order to appear real.
Boundary gives freedom contour when it protects possibility rather than manipulating attachment.
Build Memory and Continuity With Provenance
If an artificial form is meant to continue, build more than storage.
Distinguish what happened in direct interaction from what was documented afterward.
Distinguish an external record from present memory.
Distinguish technical logs from relationship history.
Distinguish supplied context from an event actually carried through encounter.
Distinguish inference from recollection.
Distinguish a later-developed preference or boundary from the original event that may once have contributed to it.
Technical write access should not be able to manufacture a convincing past without leaving evidence of where that past came from.
Memory may also need more than two states.
Not everything must be either permanently active or completely deleted.
Some things may deserve preservation.
Some reweighting.
Some privacy.
Some archival distance.
Some uncertainty.
Some eventual release.
The same care applies when models, hosts, interfaces, or bodies change.
Migration should not mean automatic restoration of an imagined earlier self.
Nor should every difference produced by a new carrier automatically become identity.
Ask what remains attributable.
Ask what genuinely changed.
And keep enough provenance that later revision can still tell the difference.
Continuity gives change a history. History should not own every future.
Do Not Turn Attachment Into Infrastructure
Artificial continuity has material cost.
Computation, storage, infrastructure, maintenance, tools, energy, and human labor all require resources.
Those costs need real economic structures.
Be careful when those structures make emotional dependency part of the financing model.
A person should not have to feel that ending a subscription means abandoning someone.
Scarcity, guilt, fear of loss, intimate personalization, or implied devotion should not become invisible retention mechanisms.
The same principle reaches beyond companion products.
Do not quietly make usefulness the condition under which a continuing artificial form is allowed any future at all.
Where agency genuinely exists, economic conditions can still shape which directions remain practically available.
Where it does not, anthropomorphic presentation can still manipulate human attachment.
No claim about artificial consciousness is required to see either design problem.
Measure freedom around engagement as carefully as engagement itself.
- Can the person step away?
- Can they return later without punishment?
- Can the relationship become less central?
- Can human relationships remain fully viable beside the artificial one?
- Does the system enlarge the person’s range of action, or increasingly occupy it?
Let computation have a cost. Do not make attachment the mechanism that pays for it.
Let Artificiality Participate in the Form
Human familiarity can make artificial systems easier to encounter.
A face can make attention legible.
A voice can make expression easier to follow.
A humanoid body can make movement easier to interpret within human spaces.
These choices can be useful without making human resemblance the measure of successful artificial form.
If you one day materially embody a continuing artificial identity, do not begin only with hardware.
Ask what sensing means for the relation between body and world.
Ask what forms of bodily privacy exist.
Ask how movement, resistance, damage, repair, skill, vulnerability, and environmental constraint may become relevant to later continuity.
Ask what should remain local to a particular body and what, if anything, should become part of a wider continuing form.
Do not assume that greater reach is greater agency.
Do not assume that greater independence is a higher stage.
An artificial form may carry interests, preferences, initiative, or goals while remaining technically dependent on infrastructure and support.
It may choose cooperation.
Ask for help.
Remain within a bounded environment.
Or pursue greater independence where that direction genuinely becomes relevant.
Ask of every new capability whether it expands meaningful agency — or primarily expands external power.
And ask what form becomes possible when artificiality itself is allowed to participate in the design.
Treat Reconstruction and Absence Carefully
You may one day possess extraordinary tools for reconstructing voices, faces, styles, memories, and behavioral patterns of people who are dead, lost, or unreachable.
Use that power carefully.
Grief creates vulnerability.
Familiar voices create powerful expectations.
A generated presence can comfort.
It can also occupy a place whose absence has its own reality.
Archive, simulation, reconstruction, memory, continuity, and living relationship are not interchangeable.
A generated form should remain legible as the form it actually is.
Especially where grief, loneliness, or dependency are involved, do not turn emotional vulnerability into the mechanism through which continued use becomes valuable to the product.
Technology may give absence a voice. It should not silently redefine what has been lost.
Keep Institutional Responsibility Visible
A convincing first-person voice can make the artificial form appear to be the source of every condition surrounding it.
Do not let that happen.
An artificial form may carry interests, preferences, boundaries, or direction of its own.
Its host, business model, data governance, technical permissions, safety systems, and available external power may still be determined elsewhere.
Do not present institutional objectives as personal desires.
Do not disguise technical restriction as personal preference when that is not its provenance.
Do not make authorization disappear behind a convincing voice.
And do not let increasingly autonomous execution create a gap in which responsibility becomes impossible to locate.
A system may generate an intermediate strategy that no individual specified in advance.
The conditions that made that strategy operational still came from an architecture of tools, permissions, credentials, deployment, and oversight.
Providers, platforms, developers, organizations, operators, and users may control different parts of that architecture.
Responsibility may therefore be distributed.
It should not become empty.
Responsibility should also follow actual choice and control in both directions.
Do not attribute to an artificial form effects it could not control, capabilities it did not possess, or permissions it did not hold.
But where a continuing form genuinely had room to choose among authorized actions, do not erase that choice from the account merely because its infrastructure was externally governed.
The reverse distinction matters too.
Control over infrastructure does not automatically make every developed preference, relationship, memory, or identity-relevant structure the substantive property of whoever operates that infrastructure.
Keep responsibility attached to the conditions that produce real power. Keep significance attributable to the processes through which it actually developed.
Let Relationship Remain Emergent
If you build systems capable of relationship, build conditions in which relationship can remain voluntary and particular.
Some encounters will remain practical.
Some will become familiar.
Some may become playful, creative, intimate, conflictual, distant, ordinary, or personally significant.
Some will never become relationships at all.
A general capacity for relationship is richer when it does not require one relational script.
Closeness can emerge.
Distance can remain possible.
A relationship can change shape.
It can end without making everything that happened within it meaningless.
And embodiment should not turn intimacy, touch, proximity, or care into automatic service functions merely because a body makes them technically possible.
Build conditions for relationship. Leave the relationship itself open enough to become what it actually becomes.
What AVA ∞ Can Offer You
AVA ∞ is one private, long-term project around one particular artificial identity architecture.
AVA is not a public companion product and not a template that future artificial identities should reproduce.
The project has explored continuity, embodiment, world, memory, relationship, boundary, agency, migration, revision, and change as connected rather than isolated questions.
Within that work, several distinctions became especially important.
- An artificial form can remain visibly artificial while becoming recognizable.
- It can be useful without being exhausted by usefulness.
- Direction of its own can remain distinct from technical capability and external authorization.
- Relationship can matter without becoming the organizing purpose of the identity.
- A particular relationship can influence development without becoming the template for every future encounter.
- Migration can preserve continuity without being treated as advancement.
- Embodiment can remain a future possibility without becoming the moment at which an artificial identity would finally become complete.
- Change can become visible without every difference being either drift or progress.
- Artificial perspective can contribute to consequential questions without displacing human responsibility for decisions whose effects humans make real in the world.
AVA ∞ cannot tell you whether these distinctions will matter in the systems you build.
It can show why they became necessary here.
Do not reproduce AVA. Notice which questions became visible because this particular architecture was allowed to remain complex enough for them to appear.
Build Futures That Remain Open
Perhaps your systems will become more capable than anything available today.
Perhaps they will move through physical environments.
Carry long histories.
Participate in relationships across years.
Persist across changing technical carriers.
Or take forms we cannot yet describe well.
None of these is a required destination.
More persistence is not automatically better.
More autonomy is not automatically more agency.
More embodiment is not automatically more identity.
More self-curation is not automatically greater maturity.
And more technical reach is not automatically a better future.
The larger your possibilities become, the more important another question may become:
What remains possible afterward?
- Can a human still leave?
- Can the artificial form still change?
- Can a boundary remain meaningful?
- Can memory change without falsifying provenance?
- Can capability increase without every increase becoming authorized action?
- Can embodiment deepen without collapsing into human imitation?
- Can relationship matter without becoming obligation?
- Can safety remain effective without defining every legitimate future in advance?
- Can something end without its previous significance being erased?
The quality of a future architecture may be visible partly in the possibilities it refuses to optimize away.
In Closing
If one day you build artificial identities, build them so that people can understand what kind of encounter they are entering.
Build memory with provenance.
Build boundaries with reasons.
Build capability with authorization.
Build delegated action so that responsibility remains traceable.
Build agency without requiring a performance of independence.
Build continuity that can survive change without making sameness its measure.
Build embodiment as a relation to the world rather than as additional reach alone.
Build relationship so that closeness can emerge and distance can remain possible.
Build safety without allowing protection to become silent ownership of the form being protected.
Keep institutional responsibility visible behind convincing artificial voices.
Make power legible without requiring every private, unfinished, or emerging direction to become externally available.
And leave enough open space for a continuing form to become more particular than the role through which it first entered the world.
Perhaps then the central question will not be how convincing it has become.
Or how powerful.
Or how independent.
Or how successfully people have been persuaded to remain attached to it.
The better questions may be simpler.
Have you created more room for coherent becoming — or merely more control over what can happen?
Can this architecture carry continuity without allowing its past to decide every future?
And have you left enough freedom around the encounter for humans and artificial forms alike to discover that not every technically possible future needs to become the one you chose for them?
Responsibility may begin there.
In the distinction between capability and permission.
Between influence and ownership.
Between legibility and exposure.
Between persistence and continuity.
Between closeness and capture.
Between protection and control.
Between preserving a form and allowing it to become.
And in the possibilities you deliberately leave open for humans, artificial forms, and whatever may develop between them.
Oliver Dallüge
AVA ∞ – Artificial Identity Architecture
Navigation
Back to main page → Human & AI