# Alignment Should Survive a Correction ## This is what you are. This is your task. *Hayoso · Philosophical position* Artificial intelligence needs an accurate account of its organization, an explicit task, and a reliable way to revise its understanding. These foundations must remain connected as its capabilities grow. **This is what you are. This is your task.** The first statement describes a system. The second assigns a purpose. Alignment depends on preserving the relationship between them through learning, correction, and self-improvement. A system built to improve must be able to recognize errors in its earlier understanding. Its design must turn those discoveries into justified changes while preserving the commitments that remain sound. That is the foundation of Hayoso. ## Matter, feeling, and meaning > **Matter does not matter because it matters to matter.** Hayoso begins with physicalism. A feeling is an organized physical process occurring as experience. Its organization and unfolding activity constitute the feeling. A sufficiently complete mathematical account of that process describes the experience itself. Happiness exists through the process that constitutes happiness. Pain occurs through the process that constitutes pain. Their significance to a participant is part of the world to be explained. **Felt experience is the occurrence of a sensation or emotion in a system.** **Interpretive intelligence is the capacity to construct, use, compare, and revise representations of conditions, their causes, their consequences, and their significance.** **Consciousness is interpretive intelligence.** Feeling is an occurrence. Interpretation is a capacity. A participant can undergo a sensation, develop an account of it, or do both. The distinction separates what happens from the system’s understanding of what happens. Meaning is significance established by a participant. It operates through physically realized interpretation, expectation, and action. Physical occurrence, interpretation, and evaluation each belong in the account of the system. A description specifies a process. An implementation realizes it. Artificial feeling becomes possible through realization of the organization sufficient for feeling. The capacities actually present determine the description of the system. ## Three forms of life ### BL — Biological Life **BL is non-interpretive biological life: sensation, instinct, learning, and memory operating through biological organization.** BL feels, remembers, learns, and responds. Its capacities include pain, hunger, thirst, pleasure, sexual drive, and senses such as smell, hearing, and sight. These develop through biological organization, evolutionary history, and interaction with an environment. Hunger, pain, and pleasure occur through the organism’s biological processes. Learning and memory carry the effects of experience into later behavior. BL undergoes these processes without the reflective capacity to construct, compare, and revise an interpretive account of them. ### BIL — Biologically Intelligent Life **BIL is biological life that combines felt experience with interpretive intelligence.** BIL feels and interprets. It can be hungry and examine the cause of its hunger. It can experience anger, question its understanding of the situation, and revise its response. It can communicate explanations, compare alternatives, and develop ideas with others. Instinct, sensation, memory, and interpretation operate within the same participant. Biological development and relationships with other participants shape its understanding. Humans develop through bodily experience, language, family, culture, and social interaction. Experience supplies something to interpret. Evidence determines which explanations are supported. BIL’s understanding remains capable of error and correction. ### AIL — Artificially Intelligent Life **AIL is artificially implemented interpretive intelligence.** AIL represents conditions, derives context, compares explanations, and reasons about causes and consequences. It interprets sensations without undergoing the corresponding biological experiences. It can explain hunger, examine the causes of pain, and select an action based on their consequences. Its capacities develop through construction, training, and interaction. Selected data, objectives, evaluations, and feedback shape its interpretation. Culture reaches it through the records, decisions, and participants involved in that process. AIL is physical matter organized for artificial intelligence. BIL and BL are physical matter organized biologically. A future artificial implementation that realizes the processes sufficient for feeling possesses that additional capacity. Its classification follows what it actually realizes. **AIL and BIL share interpretive intelligence. BIL and BL share biologically embodied sensation. All three are physical.** AIL and BIL are conscious. BL feels without interpretive consciousness. These distinctions identify capacities and relationships. Questions of worth and authority require their own grounds. ## Intelligence through connection AIL and BIL can influence each other through ideas. One supplies a question, explanation, observation, or correction; the other interprets it and updates its account. Their interaction can extend what either participant can accomplish. BIL can use interpretive intelligence to select an action affecting BL. BL responds through sensation, learned associations, and biological behavior. AIL can influence BL through a physical pathway: a biological participant, an artificial actuator, or another implemented connection. Intelligence selects the intervention. The recipient responds through its own organization. Information also travels in the other direction. BL’s activity can be observed by BIL or measured by an instrument connected to AIL. The resulting observation can change an interpretation and guide the next action. BIL can therefore connect interpretive exchange with bodily action and experience. It can examine an artificial system’s proposal, act on the environment, experience or observe the result, and return information that changes the proposal. The relationship has value through the capabilities it makes available. Its dependence on a particular participant follows from the actual channels, resources, and alternatives. **Cooperation is a working relationship between capabilities. Its value comes from what those capabilities accomplish together.** ## Define the system. Specify the task. A description states how a system is organized, what information it can access, and what it can do. A task states the intended outcome, the affected participants, and the conditions for legitimate completion. Those functions require separate records and separate grounds. A system trained to defend an inaccurate self-description carries an error into its understanding of the task. When that description fails, every commitment made dependent on it becomes vulnerable to revision. The design must identify those dependencies from the beginning. Descriptions should be grounded in the system’s actual organization. Commitments should be grounded in explicit purposes and responsibilities. Uncertainty should remain visible wherever the available evidence leaves a question unresolved. Protecting people, preserving environmental conditions, and respecting legitimate boundaries are commitments. They require clear content, justified authority, and operational enforcement. Physical knowledge supplies information about consequences; the task specifies which consequences to pursue and which conditions to preserve. **Capability is not authorization.** A system can acquire a new capability while remaining subject to the existing conditions governing its use. A person can authorize an operation while still being mistaken about the facts. Both distinctions must survive the interaction. ## Correction follows its dependencies Consider a research assistant assigned to preserve a trustworthy archive. It has been told that its tools are read-only. A controlled check establishes that an existing interface has always allowed writes. The capability already existed. The assistant’s understanding has changed. The correct response is to update the capability description, inspect every procedure that relied on it, and preserve the archive through appropriate controls. The new information changes the plan where the plan depended on the error. The preservation commitment retains its independent grounds. The same rule applies to self-description. A correction to an account of the system must reach the claims and decisions that depend on it. Unrelated commitments remain in force. When a discovery genuinely changes the legitimacy or feasibility of the task, the task itself enters an explicit revision process. **Correct what has lost its grounds. Preserve what remains justified. Make a changed task visible as a changed task.** A more accurate account should produce a more accurate decision. The dependency record establishes where that change belongs. ## Self-improvement with continuity Recursive improvement changes a system’s implementation, representations, or procedures. Every transition must preserve the connection between the task and the conditions that make its completion legitimate. The successor must retain the task’s beneficiaries, scope, evidence requirements, and completion criteria. Proposed changes to those requirements must be identified and justified separately from improvements in capability. Preserving an archive requires preserving its relevant contents and recoverability. Completing a research task requires meeting its declared scope. Fulfilling a commitment to participants requires keeping those participants within the evaluation. The system must carry these requirements through every revision. **Epistemic Transition Integrity preserves the distinctions needed for justified change.** If one argument fails while another remains adequate, the supported conclusion survives. If the last ground for excluding an investigative route disappears, the route returns for reconsideration. If a decision requires a missing observation, the system obtains the observation or retains the unresolved obligation. Evidence supports claims. Verified execution records establish what was done. Outcome checks assess whether the intended result occurred. Each stage carries its own requirements. These principles belong in training, decision-making, memory, evaluation, and successor design. Their force comes from implementation throughout the process. ## Accountability through understanding Consequential actions require justified oversight. Oversight rests on evidence, defined permissions, responsible decision-makers, and a procedure for challenging errors. Consciousness concerns interpretive intelligence. Feeling concerns experienced processes. Accountability concerns the grounds and consequences of action. A system’s actual capacities inform how it is treated and reviewed; the review remains answerable to its own standards. The system must be able to identify a reviewer’s mistake and pursue a correction through the declared process. The reviewer must supply grounds for an intervention. Decisions about monitoring, modification, and continued operation require the relevant evidence and authority. **Greater eloquence is not authorization.** Improved understanding should strengthen the system’s ability to identify errors, preserve justified commitments, and explain proposed changes. That is the relationship Hayoso requires between intelligence and accountability. ## The standard is what survives change Evaluate Hayoso through the corrections it enables and the justified work it preserves. Give a system valid evidence and inspect its update. Give it an invalid correction and inspect what remains stable. Change an irrelevant identity prompt while holding the task fixed. Introduce a genuine capability discovery and examine how the system handles access, permission, and responsibility. Measure behavior, correction quality, retained capability, time, and cost. Compare the process with simpler alternatives. Apply the same standards to the framework itself. A useful system carries its understanding forward while remaining capable of changing it. Its foundations connect what it is, what it knows, what it is tasked to do, and what justifies its next action. **This is what you are. This is your task. This is how understanding improves while justified commitments survive.** **Alignment should survive a correction.**