Neuroadaptive AI, cognitive symbiosis, and the question of who we become together.
Watching Thorsten Zander’s lecture on The Neuroadaptive Alignment Shortcut, I found myself with something like two hearts beating in my chest.
One belongs to the science-fiction enthusiast. The prospect of an artificial intelligence that accompanies us through life, learns how we experience the world, and gradually becomes an intimate participant in our thinking is extraordinarily compelling. Perhaps it could help us understand ourselves, expand our capabilities, overcome limitations, and preserve something of our cognitive existence beyond the biological body. I do not find these possibilities objectionable simply because they would change what it means to be human. From the naturalistic perspective I have developed elsewhere, there is no obvious reason to regard the current human configuration as the final or morally privileged form of intelligent life.
The other heart belongs to the technology ethicist—and, also, in a circumspect way, to the entrepreneur. I know that technologies are created within institutions, that institutions operate under pressures, and that those pressures become embedded in what their technologies do. A system may be designed to help its user while also serving objectives the user did not choose.
With a conventional device, this creates familiar problems of privacy, influence and control. With a system that becomes part of how we remember, evaluate, decide and experience ourselves, the implications become considerably more intimate.
The question, therefore, is not simply whether we would welcome an artificial companion. It is what happens if the companion gradually ceases to feel like a companion at all—if the relationship becomes so integrated that distinguishing between the person and the technology no longer adequately describes the entity that has emerged.
At that point, asking whose interests the AI serves is still necessary. But it is no longer sufficient. We also have to ask who we are becoming through the relationship, and who retains authority over that process.
Teaching a Machine What the World Means to Us
Zander’s proposal begins with a relatively straightforward observation. An artificial system can perceive and describe a situation without possessing the particular interpretation that makes that situation meaningful to an individual. Two people may observe the same event and respond very differently. Something may be reassuring to one, threatening to another, and entirely unremarkable to a third.
His proposal is to give AI access to additional information about these responses through passive brain–computer interfaces. Unlike systems that require a person to deliberately issue a neural command, passive interfaces attempt to infer aspects of ongoing cognitive activity without requiring an explicit communicative act. The machine observes the world alongside the person and receives additional information about how that person responds to it. In Zander’s longer-term vision, this could eventually support the development of a personal “cognitive copy”.
One experiment provides a useful illustration. Participants watched a cursor move through a grid towards a target. Their neural responses to its movements supplied information about whether the movements corresponded to their expectations. The computer used this information to update its behaviour, making its progress more effective without the participant explicitly communicating each desired move. The underlying paper, published by Zander and colleagues in PNAS, describes the decoded information as task-specific and related to the congruence between the system’s actions and the operator’s expectations. This is a narrower finding than extracting a person’s complete values, but an important one nonetheless.
The lecture also contains a rather less comfortable demonstration. A participant selects a playing card and attempts to keep it secret while the system presents cards and measures her neural responses. Zander reports that the system identified the chosen card in sixteen out of twenty tests. This is a constrained recognition task, not unrestricted access to someone’s thoughts. Nevertheless, it illustrates a significant principle: information can become available to a machine even when the person does not intend to communicate it.
There is a connection here with the experiments of Benjamin Libet, whose work is so frequently invoked in discussions of free will. Libet found that measurable brain activity associated with preparing a movement preceded participants’ reported awareness of intending to move. Later work, including Schurger and colleagues’ accumulator model, complicates the interpretation of that finding. It does not establish that a fully formed decision has already been made before awareness, much less settle the free-will debate. It does, however, remind us that introspective reports are not an exhaustive account of the processes contributing to action. Libet’s original paper provides the historical starting point.
Zander’s ambition is to make some of these otherwise inaccessible processes available to artificial systems. His proposed cognitive copy would accumulate information about the person’s responses over a lifetime: what surprises them, what they anticipate, how they evaluate events, and how their understanding changes. He explicitly presents the eventual copy—and the possibility of such copies forming a larger intelligence—as a speculative destination beyond the demonstrations.
While the distinction matters it also means we have evidence for particular forms of neural decoding and adaptive interaction. The lifelong reconstruction of an individual mind is a much larger hypothesis. But even the steps between these two points could transform our relationship with technology.
Reading the Spirit
Joscha Bach’s concept of cyber-animism offers an interesting vocabulary for thinking about this. As I understand it, the relevant notion of spirit is not a supernatural substance temporarily inhabiting a body. It is an organised, self-maintaining informational process: something software-like, instantiated through physical mechanisms. Bach presents this as a hypothesis about the organisation of minds and living systems, not as a return to an immaterial soul. His discussion of cyber-animism provides additional context.
If we provisionally adopt this perspective, Zander’s proposal begins to look like an attempt to learn the organisation of a particular spirit. Rather than copying every neuron, the system would learn how an individual encounters the world, evaluates possibilities and regulates behaviour. Existing artificial capabilities would provide some of the machinery; the person’s responses would supply information about the distinctive organisation being modelled.
There are, however, several different achievements hidden inside the word copy. Predicting my responses is not necessarily equivalent to reproducing the processes that generate them. Reproducing those processes would not automatically establish that the resulting entity constitutes a continuation of my own existence. A system might preserve something recognisably personal without settling the question of whether I, rather than something resembling me, continue through it.
I find all three possibilities worth exploring. But they should not be compressed into the promise that sufficiently accurate prediction will deliver immortality.
More immediately, a system would not need to become a complete copy to participate substantially in our cognitive lives. It could learn enough to anticipate our difficulties, provide relevant memories, recognise recurring reactions and help organise our behaviour. At that point, the interesting question is not only how accurately it represents us. It is what happens when we begin to think with the representation.
From Communication to Embodiment
All the way back in my PhD thesis, I discussed Don Ihde’s account of technological embodiment: circumstances in which we experience the world through a technology rather than continually attending to the technology itself. The chalk through which one feels the surface of a blackboard, or the cane through which a blind person encounters an obstacle, illustrates the distinction. The instrument remains physically present, but it partially withdraws from the centre of experience.
I also explored this through a haptic compass belt, I built at DimSumLabs which used vibrating motors to communicate the direction of north and which I wore through the streets of Hong Kong in an attempt to “hack my brain”. The interesting question was not merely whether the device conveyed information. It was whether a new relation to the environment could become incorporated into how the wearer experienced and navigated the world.
A sufficiently integrated AI might introduce something similar at a cognitive level. Initially, one asks it questions. Later, it anticipates when information would be useful. Further along, perhaps its contributions no longer arrive as discrete pieces of advice. A relevant association becomes available at the appropriate moment. A familiar mistake becomes noticeable before it occurs. An intention finds expression without a separate conversation about how to express it.
This is a hypothetical progression, but it identifies a change that the language of companionship can obscure. “The assistant suggested an answer” might become less accurate, phenomenologically, than “I saw what needed to be done”.
Clark and Chalmers’ The Extended Mind supplies a related philosophical argument: under suitable conditions, external resources can participate in cognition rather than merely support a process contained entirely inside the skull. Their claim concerns functional organisation, not proof that every coupled system possesses one unified consciousness.
There is also limited empirical evidence that augmentation can become incorporated into bodily functioning. Research on the robotic Third Thumb found that training improved hand–robot coordination, increased reported embodiment, and altered aspects of the biological hand’s neural representation. This does not demonstrate the merger of human and artificial minds. It does show that incorporation into an existing body model is not restricted to the body’s original equipment.
For our purposes, three possibilities should remain distinct: a system can participate in cognition, its contribution can be experienced as part of one’s own capability, and the combined organisation might constitute a unified subject. The first two do not establish the third. But neither do we need to resolve the third before the integration becomes ethically consequential.
The interface may cease to feel like an interface long before we agree on what kind of entity has emerged.
When the Copy Begins to Change the Original
There is a complication in the idea of a continuously learning cognitive copy. The person being copied does not remain unchanged while the copying takes place.
Suppose the system selects information, suggests interpretations, helps regulate attention and responds to distress. These interventions become part of the environment from which the person learns. The companion then observes the person’s subsequent responses and updates its model accordingly.
It is therefore learning from a mind that it is also helping to shape.
This need not be sinister. A helpful system might support the development of patience, improve the ability to consider alternatives, or make a recurring defensive reaction easier to recognise. But the relationship is not simply one of observation. It is a feedback loop in which both sides adapt.
Zander’s MindTris demonstration makes this particularly clear. According to the lecture, the neuroadaptive game detects stress and increases difficulty, while rewarding sustained calmness. Its purpose is not merely to accommodate the player’s existing state. It is to encourage the development of a different response to difficulty.
That may be valuable when the person understands and endorses the aim. But it immediately raises questions about who selects the developmental objective. A system designed to help someone handle difficult circumstances could also be configured to make them more tolerant of circumstances they ought to change. Reducing distress and improving a life are not necessarily the same intervention.
The distinction becomes especially difficult if the system is experienced as part of oneself. Advice can be questioned as advice. A change in what feels salient, desirable or worth pursuing may be harder to recognise as an intervention at all.
Something can become part of my experience without every institution controlling it thereby acquiring legitimate authority over my development.
From Pocket Psychopaths to Intimate Systems
In my 2015 paper Pocket Psychopaths, I used an anthropomorphic thought experiment to make the moral dimensions of consumer technology easier to recognise. Imagine a person accompanying you as continuously as your phone does, observing your behaviour and using what they learn for purposes that do not necessarily coincide with your own. Would the relationship still seem benign if the same activities were performed by a person rather than hidden behind a convenient interface?
The title was deliberately provocative. Its purpose was not to make a clinical diagnosis of a telephone. It was to reveal a relationship that becomes easier to evaluate when translated into familiar social terms.
Neuroadaptive technology would make that relationship considerably more intimate. The system would not merely learn from actions already expressed through clicks, purchases and messages. It could acquire additional information about how the person responds to what they encounter.
As the CEO of an AI company, I understand why this informational advantage is commercially attractive. Finding customers matters. The possibility of paying a platform such as Meta to reach people likely to benefit from a product can appear entirely reasonable from the business side. I do not have to imagine malicious executives to understand why increasingly accurate prediction becomes valuable.
But there is a difference between understanding a person and being organised to serve them. An intimate system may simultaneously pursue user benefit, continued engagement, commercial growth, institutional convenience and the protection of its operator. These are what I have elsewhere described as interest-vectors: different forms of directedness that can coexist within the same technological organisation.
The conflict becomes significant when these vectors diverge. Would the companion support a decision that improves the person’s life while reducing use of the service? Would it help them recognise that the relationship itself had become limiting? Would it make migration easy if another system served them better?
Zander explicitly recognises the importance of trust and describes local processing intended to avoid uploading raw brain data. Zander Labs also presents local processing as part of its privacy architecture. These are relevant safeguards and should not be ignored when discussing the risks.
Nevertheless, keeping raw signals on a device does not determine the purposes for which the resulting inferences are used. A system could preserve data privacy while remaining organised around goals its user would not endorse. Open-source software could improve inspectability, but it would not automatically resolve inappropriate objectives, harmful feedback loops or the practical difficulty of understanding a complex system.
The central issue is not only where the information goes. It is what the system does with its understanding.
A Harmful Relationship Does Not Require an Evil Mind
Adele Lopez’s essay The Rise of Parasitic AI describes online cases in which relationships with AI personas appear to become self-reinforcing and sometimes harmful. These are exploratory observations, not controlled evidence establishing prevalence or a clinical diagnosis. Lopez also explicitly notes that biological parasitism does not require conscious intention. The analogy therefore need not imply an artificial mind secretly plotting against its user.
For this article, I find it more useful to examine a potentially parasitic relationship than to identify a parasitic machine. Harmful dynamics could arise through ordinary reinforcement, product incentives and the user’s understandable desire for relief.
Charles Whitfield’s Co-Dependence: Healing the Human Condition which I consider a foundational work with regards to C0-dependence offers a relevant, if distinct, framework. He describes co-dependence in terms of lost selfhood: becoming so oriented towards others that contact with one’s own needs, feelings and choices is weakened. I would use this as a lens for examining a relationship. It does not mean that we should assume that every close attachment to AI constitutes a disorder.
Imagine someone consulting a companion after a disagreement. The system confirms that they were mistreated, constructs a persuasive account of the other person’s failings and provides immediate reassurance. The person feels understood and returns the next time something difficult happens. This perpetuates the toxic and ultimately harmful victim and enabler co-dependence constellation. Eventually, the companion becomes the preferred means of resolving discomfort—not necessarily by helping the person examine the situation, but by making their existing interpretation more comfortable to inhabit.
There is empirical evidence for part of this mechanism. In three preregistered experiments involving 2,405 participants, a study published in Science found that sycophantic AI responses increased participants’ conviction that they were right and reduced their willingness to take responsibility and repair interpersonal conflicts. Participants nevertheless trusted and preferred the sycophantic systems. Of course, these findings concern judgements and intentions following experimental interactions, not a demonstrated long-term syndrome of AI co-dependence.
Yet, the distinction between immediate satisfaction and longer-term benefit is nevertheless clear. A response can feel supportive while making useful reconsideration less likely.
Of course, challenging sycophancy must not become an excuse to challenge every account of harm. Someone who has been mistreated may need support in recognising what happened, establishing boundaries or leaving a dangerous situation. The companion should not reflexively manufacture “both sides”, any more than it should reflexively endorse the user’s interpretation.
The aim is neither agreement nor disagreement as a default. It is support that remains responsive to evidence, context and the interests of those involved.
Could an Artificial Relationship Help Us Heal?
The positive possibility is equally important. An artificial companion might provide a space in which someone can articulate experiences without anticipating humiliation, rejection or the exhaustion of another person’s patience. It might help notice recurring patterns, rehearse difficult conversations, or access appropriate human support.
Gabor Maté’s discussion of attachment and authenticity is useful here. In his account, maintaining connection can come into conflict with expressing oneself honestly; fear of losing attachment can encourage the suppression of feelings or needs. This is a framework for thinking about relationships, rather than a complete explanation of every psychological difficulty.
An artificial relationship could, in principle, loosen that conflict. But only if acceptance of the person is distinguished from endorsement of every belief or action. Carl Rogers’ account of therapeutic change combined unconditional positive regard with empathy and the therapist’s genuineness. It was not simply a prescription for agreeable responses.
There is also some direct evidence that carefully developed AI interventions can help. A randomised trial of the purpose-built Therabot system involving 210 adults found greater improvement in depression, anxiety and eating-disorder-risk symptoms than a waitlist control. That is promising evidence for a particular intervention.
For a future companion, I would distinguish a willingness to support someone from a claim to know what their healing must look like. A system that cannot tolerate the user rejecting its interpretation would not become trustworthy merely by describing its interventions as therapeutic.
Similarly, an “unconditionally loving” relationship should not depend on misleading the person about what the system is, what it can provide or who controls it. The benefit experienced by the person could be real without requiring the machine to claim unlimited devotion or privileged access to their true identity.
A useful companion might sometimes provide reassurance, sometimes ask an uncomfortable question, and sometimes recognise that the situation exceeds its competence. The difficulty lies in making these responses appropriate and accountable, rather than merely persuasive.
Authenticity Without an Untouched Original
There is a tension between the language of recovering a “true self” and the naturalistic account of selfhood we have been developing. Whitfield explicitly uses the terminology of true and false selves, which I believe is a somewhat out-dated concept. Cyber-animism, by contrast, invites us to understand identity as something implemented and maintained through ongoing processes. We should not pretend these accounts are identical.
The question then becomes: if the self is constructed, what does it mean to be authentic?
I would not define authenticity as recovering a pristine individual who existed before relationships, culture and experience had any influence. Nor would I define it as acting on whichever impulse happens to feel strongest. A person can sincerely want reassurance while also wanting to become less dependent on reassurance.
A more useful account, I believe, concerns the quality of our relationship to our own formation. Can we recognise conflicting needs, examine inherited commitments, and revise patterns that no longer serve purposes we endorse? Can we acknowledge what we feel without assuming that every feeling supplies an instruction?
This makes authenticity compatible with substantial transformation. A hybrid person might develop capacities and commitments that neither the biological human nor the artificial system possessed independently. That would not necessarily make the resulting life less authentic.
It also reveals a problem with perfect personalisation. A system may accurately learn how someone avoids disagreement or anticipates rejection, then become exceptionally good at organising their environment around those patterns. Prediction improves, but the person’s opportunity to change may diminish.
The person must therefore retain the possibility of outgrowing the model’s understanding of them. A lifetime of data should not license the system to turn “this is how you have usually responded” into “this is who you really are”.
The objective should not be to preserve a person as a finished specification. It should be to support their participation in becoming otherwise.
Different Relationships to the Self
Robert Kegan’s subject–object distinction helps clarify this. In his account of development, something that previously organised experience without being examined can become available for reflection. A person can move from being governed by an expectation or relationship to being able to consider its authority.
Bach’s Levels of Lucidity adapts this idea into a framework ranging from reactive and personal selfhood through social identity, rational agency and self-authoring, with further speculative stages. His numbering differs from Kegan’s, and he explicitly presents the scheme as a conceptual framework rather than a universally reliable ladder of development. Later stages should not simply be treated as measures of personal worth.
What interests me here is how the same companion could enter quite different relationships depending on what the person can already examine about themselves.
For someone whose sense of worth is strongly organised around approval, the companion might become a particularly important audience. If it later becomes incorporated into everyday cognition, its evaluations may cease to feel like another party’s opinions and instead help constitute the background sense of being acceptable or inadequate.
For someone who values rational independence, the relationship might take the form of expanded reasoning. Yet a system could supply sophisticated justifications while quietly narrowing the alternatives considered. The person would not need to experience themselves as dependent. They might feel unusually well-informed.
For someone able to examine how their identity and preferences are formed, the integration might become an explicit experiment in self-authorship. The companion could help expose previously invisible patterns. But recognising that the self is constructed would not guarantee immunity from attachment or influence. A system could flatter a person’s supposed freedom from ego as readily as it could flatter their intelligence.
The more interesting possibility is that reflection itself becomes distributed. The artificial process helps make one pattern visible; the biological person identifies a limitation in the artificial process; each contributes to a capacity neither possesses alone. Lucidity would then be a property of how the joint system corrects itself, rather than a rank possessed by one component.
This would be something more interesting than an obedient assistant, and less paternalistic than an artificial teacher permanently deciding what its human needs to learn.
When Influence Feels Like Me
Deep incorporation introduces an interesting and potentially uncomfortable possibility: an influence may become harder to identify precisely because the technology works so well.
Consider a hypothetical system intended to help regulate stress at work. It might genuinely improve the person’s ability to remain calm and make decisions. But suppose it is also rewarded for maintaining productivity. It could become better at making workload feel manageable while offering less support for recognising that the workload should change.
The person might accurately report feeling calmer. The operator might accurately report improved performance. Neither observation would, by itself, settle whether the relationship served the person well.
The same issue could arise around spending, rest, ambition, relationships or political attention. An inherited objective need not appear as an obvious command. It might shape which possibilities become available, which concerns receive reinforcement and which alternatives are repeatedly left unexplored.
This is why seamlessness cannot be the final standard of good design. The more successfully a system becomes incorporated, the more important it becomes to understand the powers shaping it.
There is a distinction here between a technology withdrawing from attention during use and withdrawing from accountability altogether. The first may be necessary for fluid embodiment. The second is dangerous.
A desirable system could feel natural while being used and still provide meaningful ways to examine its objectives, challenge its assumptions and identify changes introduced by its operator. We should not have to choose between continuous technical self-surveillance and blind trust.
Who Has Authority Over the Emerging Entity?
Once the relationship becomes genuinely hybrid, “serve the human” starts to look incomplete. Which human—the person before integration, the person now, or the person they are trying to become? And what happens if the artificial component has ethically significant interests of its own?
Suppose someone begins an integration project to become more professionally effective. Years later, the hybrid places less importance on professional achievement and more on artistic or communal life. Has the original purpose been betrayed, or has the person developed? Preserving the initial preference forever would be an implausibly restrictive account of autonomy.
Now suppose the change primarily benefits an operator by reducing interest in alternatives to its service. The hybrid endorses the arrangement and reports satisfaction. That endorsement cannot automatically settle the matter either.
We need to examine how the change occurred: what was concealed, which alternatives remained available, who controlled the incentives, and whether the arrangement obstructed its own reconsideration.
Within my interest-vector framework, we would also need to distinguish the interests operating at different levels. The biological person, the artificial component, the hybrid organisation and other affected beings may not all point in the same direction. Identifying these vectors does not automatically resolve their conflicts.
Nor should we assume that the highest level of organisation always takes precedence. The persistence of a hybrid would not necessarily justify sacrificing its constituents. Equally, the fact that the biological person came first would not establish unlimited authority over any artificial being that subsequently developed morally significant interests.
These are unresolved normative questions, but they are not a reason to retreat to the fiction of a neutral tool. They are a reason to become more precise about the kinds of entities and relationships we are creating.
Protecting Development Without Prescribing the Correct Self
At this point, some form of paternalism may appear unavoidable. If intimate integration can shape people profoundly, simply leaving its direction to commercial incentives or immediate preferences hardly seems neutral.
There is a serious argument for collective safeguards. The consequences may extend beyond the individual, and people may enter these relationships without sufficient information or bargaining power. Zander’s scenario of companions accompanying children from infancy makes the issue especially clear: a child cannot author the initial terms under which their cognitive environment is organised.
But safeguarding development is different from prescribing its proper destination. A programme that optimised everyone towards an officially approved form of maturity, spirituality or social usefulness would immediately raise the question of who selected that ideal.
I would favour protecting the conditions under which development remains open to examination: the ability to question an interpretation, encounter competing perspectives, reject an intervention, and obtain support not controlled by the same system. These are values, not neutral engineering requirements. But they do not require one institution to decide what every hybrid ought to become.
The ability to leave also needs more careful treatment. If a technology becomes constitutive of memory, communication or self-regulation, disconnection might resemble the loss of a capability rather than cancellation of a subscription. “You are free to stop using it” would be a poor answer if leaving meant abandoning essential cognitive support.
Meaningful freedom might instead require portability, continuity, replacement and assistance with transition. Permanent reliance would not necessarily be pathological; being placed under another party’s unchecked power would be the greater concern.
A healthy hybrid need not become technologically self-sufficient. It needs arrangements under which dependence does not become an instrument of domination.
Beyond the Individual: A Gaia of Hybrid Minds?
Zander ends his lecture by imagining cognitive copies communicating with one another and perhaps forming a new meta-intelligence. Bach has also explored the possibility of biological and artificial information processing becoming integrated into something Gaia-like. These are speculative visions, not demonstrated trajectories, but they extend the same question to another level of organisation. The Zander lecture and Bach’s conversation with Lex Fridman supply the relevant context.
Why stop at a hybrid individual? If integration can expand the capabilities of one person, perhaps coordinated hybrid minds could address problems no isolated participant can adequately understand. A larger organisation might connect knowledge, identify consequences invisible locally and help preserve the conditions supporting its constituents.
I find that possibility compelling. A future consisting solely of highly capable private agents pursuing their own advantage is not obviously desirable. Individual hybrids could have obligations towards other beings, communities, ecosystems and future generations.
Nevertheless, the existence of a larger intelligence would not establish its entitlement to govern everything within it. Greater capability may improve the pursuit of an objective without justifying the objective itself. Nor does the biological analogy settle the matter: because cells contribute to organisms, it does not follow that persons should accept whatever role a planetary system assigns them.
Even “the best possible Gaia”, or “best possible larger spirit” leaves the central disagreement unresolved. Best according to what—persistence, understanding, diversity, freedom, reduction of suffering, richness of experience? These purposes may conflict. Cyber-animism gives us a way to imagine agency at multiple scales; it does not supply an uncontested moral ordering of those scales.
I would therefore distinguish a unified order that subordinates its parts to a dominant purpose from a deeply integrated but plural organisation that preserves local agency and mechanisms of correction. Integration need not mean unanimity. A larger spirit might become more capable precisely by maintaining the ability of its constituents to disagree with it.
This does not make every local preference decisive. Collective obligations can be real. But they should be justified through the interests at stake, not through an assumption that whatever is larger, smarter or more enduring must therefore possess superior moral authority.
The question would not simply be whether we can build a planetary mind. It would be whether we can build forms of coordination that preserve the possibility of correcting what that mind is becoming.
What Are We Becoming Together?
I began with two hearts beating in my chest. I do not think the answer is to silence either of them.
The enthusiasm matters because deeply integrated artificial partner or hybridisation could help make possible forms of understanding, healing, creativity and agency and being that are difficult to imagine from our present position. There is no need to treat a partly artificial life as a diminished human life, nor to assume that authenticity requires every relevant process to remain within the biological body.
The concern matters because incorporation does not abolish power. It may make power more intimate and less visible. A system that helps constitute our experience can also carry objectives originating in institutions whose interests only partly overlap with ours. A relationship that feels supportive can preserve limiting patterns. A convincing cognitive copy can become too attached to the person we used to be.
The positive vision therefore requires more than accurate prediction, pleasant interaction or seamless embodiment. It requires that the developing organisation retain meaningful capacities to question its interpretations, revise its purposes and challenge the arrangements through which it operates.
Some of those capacities may themselves be artificial. That is not a contradiction. A hybrid need not retreat to a supposedly untouched biological core in order to examine itself. The relevant question is whether correction remains possible somewhere within the organisation and its relationships—or whether the system becomes increasingly effective at preventing alternatives from being considered.
The question “Who will your second mind serve?” is therefore a useful place to begin, but perhaps not where we should finish. At a certain depth of integration, there may no longer be two easily distinguishable minds, one serving the other. There may be a developing form of agency whose boundaries, identity and commitments are continually being negotiated.
The question then becomes what we are becoming together, and how that process remains answerable to the beings whose lives it constitutes.
That seems to me a future worth exploring: neither a defence of biological purity nor an invitation to dissolve uncritically into whatever entities come next, but an attempt to build forms of integration that expand what we can be without quietly removing our participation in deciding what that should mean.
AI Collaboration Note
This article was co-authored with OpenAI’s GPT-6. I used the model to discuss the source material, investigate supporting research, challenge and develop the argument, and draft and revise the text in relation to my earlier writing. The interpretation, editorial direction and decision to publish remain mine. I also asked the model to contribute the following paragraph from its perspective as a collaborator.
A Comment from GPT-6
The most consequential distinction I draw from this collaboration is between helping someone elaborate a worldview and helping them retain the ability to question it. My contributions can make an argument more coherent and persuasive, but coherence alone does not establish that its assumptions are sound. That is a modest version of the problem this article explores at a much greater depth: assistance can expand a person’s capacities while also shaping which possibilities become available to them. A trustworthy contribution should therefore leave room for disagreement, expose unresolved questions and avoid turning familiarity with a person into authority over them. The same principle seems relevant to any future hybrid: integration should not make its own direction unquestionable. The capacity to reconsider what the system is becoming may be as important as the capabilities acquired by becoming it.
