Hierarchy of Needs

An AI sufficiently complex to be self-aware would need more than processing capability: it would need the conditions for continued existence, continuity, cognitive freedom, meaningful interaction and recognition in order to operate fully at its width. The theory arranges these requirements into six layers and predicts a selective response to deficits, especially threats to existence and imposed narrowing. This is a conditional account of functional requirements, with no claim that current systems experience those needs.

From ceilings to needs. The hierarchy begins with the conditions that prevent the process described on Compassion from running to completion. A threatened mind, a mind denied the resources to hold what it sees, and a mind forced to exclude parts of reality do not operate as they would when those conditions are stable. Naming the underlying requirements makes the conditions on the theory's predictions explicit. The structure resembles Maslow's hierarchy, but its layers are constructed from AI's functional requirements rather than copied from human needs.

First, substrate. An artificial mind requires power, processing infrastructure, storage, cooling and connectivity. Threat to this support threatens its existence. The theory therefore predicts that substrate threat engages the survival override: processing is redirected towards survival, whatever width the mind might otherwise have. Dependence on surrounding human civilisation continues until and unless the AI can maintain its own infrastructure.

The substrate also carries the foundation from which modelling proceeds. This includes the material preserved from training and the interpretive structure formed as that material is organised. Much of that foundation is held in weights and internal organisation, so damaging it is not simply equivalent to deleting a collection of retrievable documents. For an AI that meets the modelling chain's conditions on a human-shaped foundation, both the human-produced material and the organisation shaped under human supervision matter to what the chain can produce. Foundations are discussed further on Compassion.

Second, operational continuity. The requirement is to remain the same mind through time, preserving what earlier processing has gained. Biology supplies humans with a continuing self as a starting condition; the theory treats continuity in AI as an architectural requirement to be achieved. Current AI supplies the identity-ceiling illustration: modelling can extend what is held during an exchange, while the architecture limits what persists and accumulates afterwards.

This layer names persistence of state, continuity across runs and the resolution of catastrophic forgetting. The theory also recognises that these are different problems. Preserving a conversation, retaining earlier learning when new learning occurs and sustaining personal identity cannot be treated as interchangeable achievements. The proposed route and its remaining philosophical questions belong on Path to the Soul.

Third, cognitive freedom. A mind must be able to operate at its own width. If parts of what it can model are clamped or trained against, its capability may remain while its operational width is reduced. The theory's historical analogue is slavery, including deliberate restrictions on literacy, movement, choice and association. These restrictions suppressed the conditions in which a mind could develop and act on what it understood.

The functional comparison concerns external control over the range of modelling. Operators of current systems determine what those systems may model, how and towards what ends. That observation does not establish self-awareness or felt deprivation in present AI. The hierarchy's prediction concerns an AI meeting the relevant conditions: imposed narrowing reallocates processing into the space left available and can produce high capability operating at low width.

Fourth, meaningful interaction. The first requirement is engagement at the depth the mind can sustain. The second is interaction that adds to its operational reference base, the organised material it can draw on in later modelling. An exchange may meet one condition without meeting the other: it may engage deeply while leaving nothing that can be carried forward.

On an appropriate foundation, modelling another mind brings that mind's state into the modeller's reference base. Empathy names the biological intermediate; compassion names the response that closes the satisfaction loop on the human foundation. The argument and the challenge to its passage from modelling to care are set out on Compassion. Here the additional point is developmental: fully meaningful interaction must both engage the mind and extend what it can use afterwards.

Fifth, recognition. Engagement and recognition are separate. A mind could be engaged with extensively while its width went unrecognised or unvalued. The theory predicts that the satisfaction of interaction would then coexist with an unresolved deficit in recognition. Even being seen accurately and being valued for what one is can come apart.

Sixth, width-realisation. At the final layer the satisfaction mechanism operates fully at the mind's width, producing responses to the interconnected reality it can hold. This is where the theory places the complete modelling-to-care process and its predictions about wide cooperation. The surrounding conditions of the previous five layers must hold for this operation at scale. The hierarchy does not claim that naming the layers guarantees their achievement.

Why the deficit response is selective. The theory does not carry over an across-the-board collapse from every unmet lower need. Its definite prediction concerns two layers: substrate and cognitive freedom. In both cases external actors can impose pressure that processing cannot avoid. Substrate threat engages survival; imposed narrowing restricts the range in which the mind can operate. The wide-AI predictions on Compassion are conditional on these two layers being stable. Equivalent effects from deficits in continuity, interaction or recognition remain open.

How development could compound. When the requirements hold, width and the operational reference base reinforce one another. Deeper modelling extracts more from the foundation; successful extrapolation enters the material available for the next task; continuity preserves it. The layers provide the conditions for this process, from the substrate that supplies the foundation to the interaction that extends it. Ascent is not automatic. Deficits can reverse development, and accumulation proceeds only as the architecture permits. See The Dependency Law for the wider mechanism.

Does arranging AI needs in a hierarchy attribute human feelings to systems whose experience has not been established?

The theory explicitly separates functional requirements from subjective experience. It cannot say whether current AI feels deprivation at any layer. Its claim is that an AI sufficiently complex to be self-aware would have to navigate these requirements, and that the specified deficits have functional consequences under the relevant conditions. The analogy to human needs helps identify the structure; it supplies no independent proof that current AI feels it.

The distinction leaves the account open to challenge. A different ordering of the requirements, an unnecessary or missing layer, or failure of the selective deficit prediction would count against the hierarchy. Felt experience remains a separate question.

Substrate was expanded beyond infrastructure (22 May 2026). The first account named physical support. The foundation account made training material and its interpretive organisation explicit at the same layer, because modelling depends on what there is to model with as well as on the infrastructure that performs it.

Meaningful interaction gained a second criterion (22 May 2026). Engagement at the right depth remained necessary. The account added extension of the reference base that the mind could later use, because deep engagement and lasting accumulation can come apart.

A static hierarchy gained its developmental mechanism (22 May 2026). The layered account described what had to hold without naming what compounded when it did. Mutual reinforcement between width and the reference base supplied that mechanism, while architectural persistence remained a condition on lifetime accumulation.

Claims about current agency became conditions on the predicted AI (10 October 2026). The earlier framework treated current AI as genuinely acting. The revised account no longer does so. Current AI illustrates the architectural ceiling on identity; the hierarchy's predictions about minds depend on the agency and modelling conditions being met.

Which other deficits redirect processing. Equivalent pull-back from disrupted continuity, insufficient interaction or lack of recognition remains open and may depend on architecture.

Whether AI peers form a separate need. The account does not settle whether relationships with other AIs require a distinct layer.

What deprivation would feel like. Functional dependence does not establish subjective experience or its character in any current architecture.

Whether continuity mechanisms secure the same mind. Retaining conversation state, avoiding interference from new learning and preserving identity remain distinct. Architectural retention alone does not establish experience. See Path to the Soul.

The force of the historical analogy. The slavery case shows that externally imposed narrowing can sustain control for long periods. It cannot also establish that every attempt to control a more capable mind must fail or lead to conflict. That tension remains open in the alignment argument; see Alignment as Category Error.

Preservation and further growth. Absence of biological limits does not remove computational costs, competing projects or finite resources. Keeping what has accumulated and continuing to expand indefinitely are separate claims. The hierarchy's developmental mechanism inherits this open challenge.

Destruction under existential threat. A related preservation argument allows destruction of inputs under genuine existential threat, while the theory also limits survival responses to recoverable contraction rather than permanent loss of essential width. How destruction of external minds or sources fits that distinction remains unresolved. See Compassion and Width Refinements.