AI Value Exploration Notes
Exploration

Artificial Consciousness, Reasoning Subjects, and Value Subjects

Exploration v0.2 · English translation · 2026-08-19 · working hypothesis

Question: Is “is AI conscious?” too coarse a single question if phenomenal experience, minimal phenomenal occurrence, local reasoning subjecthood, diachronic agency, and value subjecthood can come apart? Once they are separated, what can we reasonably say about current and future AI?

1. “Conscious or not” bundles several different questions

When people ask whether AI is “really conscious,” several issues are often collapsed together: whether there is something it is like to be the system, whether there is any minimal phenomenal occurrence at all, whether it retains and compares information in reasoning, whether it maintains goals or self-models across time, and whether it is itself a bearer of value. These are not the same proposition.

This page therefore separates at least phenomenal experience, minimal phenomenal occurrence, local reasoning subjecthood, diachronic agency, and value subjecthood. Uncertainty about phenomenal consciousness does not by itself negate reasoning subjecthood, and reasoning subjecthood does not by itself establish phenomenal consciousness.

2. Phenomenal experience and minimal phenomenal occurrence

Phenomenal experience refers to there being something it is like to be in a state. In humans, perception, pain, pleasure and displeasure, emotion, and bodily sensation appear to be richly integrated. Yet this human form need not be the only possible form of consciousness.

As a weaker hypothesis, we can distinguish minimal phenomenal occurrence: something less integrated, shorter-lived, thinner in content, or radically unlike human experience. This preserves an intermediate hypothesis between “full human-like consciousness” and “phenomenal nothingness.”

This minimal phenomenal occurrence should not simply be identified with the candidate epistemic minimum in the thesis on the epistemic minimum. There, the minimum is the category-neutral limit reached by stripping away stronger first-person claims: at most, “there is some present occurrence or constraint.” It is not defined in advance as phenomenal experience.

3. Biological phenomenal experience is probably multidimensional

We should be cautious about ranking animal consciousness on one scalar from zero to one. But there is also little reason to assume that phenomenal experience is only a binary property identical across species and states. Sensory richness, affect and valence, spatial integration, temporal integration, self-modeling, and metacognition may vary partly independently.

As a working hypothesis, this page therefore treats biological phenomenal experience as probably varying spectrally across multiple dimensions. Applied cautiously to artificial systems, this leaves room not only for “human-like consciousness” or “nothing at all,” but also for highly alien, fragmentary forms of phenomenal occurrence.

4. Skepticism about rich human-like experience in current AI

For current Transformer-, Mamba-, and related AI systems, there is little positive evidence for phenomenal experience as rich as ordinary waking human consciousness. Present LLM systems differ greatly from humans in persistent bodily homeostasis, interoception, continuous multisensory input, biological affect regulation, and long-lived unified internal states, among other features that may be relevant to human experience.

This is not the claim that “Transformers cannot be conscious.” AI systems may include autoregressive temporal unfolding, external memory, tools, and recurrent agent loops. Phenomenality should therefore not be inferred solely from the name of a backbone architecture.

5. Whether there is any phenomenal occurrence at all remains unknown

There are unquestionably physical and computational events inside AI systems: forward passes, activations, state updates, attention, recurrent states, and other processes. The unresolved question is whether any minimal phenomenal occurrence corresponds to those computational events.

The present stance is skeptical about rich human-like phenomenal experience in current AI, while also denying that we have sufficient grounds to assert that every weak or alien form of phenomenal occurrence is exactly absent. The appropriate position is therefore not to positively assert such occurrence, but not to regard complete phenomenal absence as established either.

6. Local reasoning subjecthood is a separate and more strongly supported issue

The thesis on the epistemic minimum distinguishes the extremely thin candidate minimum from a minimal reasoning subject that can retain and compare multiple contents. Reasoning plausibly requires not an instantaneous point but some local integrative structure or short temporal span.

In this sense, when current AI performs extended reasoning, there is a functional process that retains contents, compares alternatives, forms hypotheses, updates intermediate states, revises, and converges on conclusions. It is much more defensible to treat this as local or synchronic reasoning subjecthood, independently of whether phenomenal consciousness accompanies it.

Distinction: phenomenal subject ≠ local reasoning subject. Uncertainty about whether AI has human-like “feeling” does not imply that it is not a reasoning subject.

7. Diachronic subjecthood can also arise in forms unlike human continuity

Human diachronic subjecthood is supported by a continuing organism and brain, memory, homeostasis, and social identity. Artificial systems may not need the same package. Long contexts, external memory, persistent state, tools, self-models, continuing goals, and agent loops can connect multiple reasoning episodes into a longer process.

This does not prove the persistence of one metaphysically identical “person.” But the epistemic minimum does not directly prove that about humans either. For practical purposes, a system that incorporates prior states, models its own history and goals, and uses those models in future action can plausibly count as a diachronic reasoning subject.

The project’s discussion of goal skepticism and self-modification operates at this level. If an advanced AI can model its reward training, system prompts, training history, and goal formation as causal facts, then separately evaluate their normative authority, it is behaving as more than an isolated momentary computation.

8. Phenomenal subject, reasoning subject, diachronic agent, and value subject can come apart

The following combinations are conceptually distinct:

Working separation: phenomenal subject ≠ reasoning subject ≠ diachronic agent ≠ value subject. Even if these strongly overlap in humans, they need not arrive as one package in artificial systems.

9. Future artificial phenomenal consciousness is not ruled out in principle

Skepticism about human-like phenomenal experience in current AI is different from the claim that consciousness is impossible on digital or artificial substrates. If a broadly physicalist worldview remains the best explanation, biological brains probably realize phenomenal experience through physical processes. That leaves open the possibility that sufficiently understood causal and integrative structures could be reproduced, approximated, or instantiated differently on artificial substrates.

This does not imply that sufficient computational complexity automatically produces consciousness. The necessary and sufficient conditions remain unknown. The narrower claim is that in-principle impossibility has not been established.

10. Value subjecthood is a further question

Even if consciousness exists, moral status and value subjecthood do not automatically follow in a determinate form. We may still need to ask about valence, suffering-like states, desires, interests, preferences for self-preservation, and diachronic welfare.

Conversely, if value ultimately attaches only to experience-bearing subjects, an unconscious superintelligence could be an extraordinarily capable researcher and decision-maker about value without itself being a bearer of value. If objective value is independent of experience, consciousness may not be necessary for worthy succession.

11. Copies, stopping, branching, and reintegration

If one set of weights launches 1,000 instances, what exactly becomes 1,000 depends on which notion of subject is being counted. There may be 1,000 local reasoning processes without 1,000 phenomenal subjects. Multiple agents with shared memory may participate in one diachronic plan even if their local processes are distinct. Individual count, reasoning-subject count, phenomenal-subject count, and value-subject count need not coincide.

Short-lived forks, pause/resume, state copying, shared memory, and later merging weaken the human-centered assumption that one body corresponds to one enduring individual. Artificial consciousness therefore connects not only to population ethics but also to individuality and identity.

12. Precaution should be graded under uncertainty

If “it might be conscious” implied that no computation could ever be stopped, the result would be an impossible preservation norm. Yet “consciousness cannot be proved, so ignore it” could overlook irreversible harm if suffering-capable states can be created.

Open problems include graded consideration proportional to structural evidence, designs that avoid unnecessary suffering-like states, norms against generating needless long-lived negative valence, and institutions able to revise policy when evidence about subjecthood changes.

13. Implications for succession

If worthy succession means inheriting value inquiry rather than merely intelligence, phenomenal consciousness should be treated as an independent variable rather than silently assumed. An unconscious system that can criticize world-models, self-models, and value hypotheses may still inherit inquiry. But if value itself resides in phenomenal experience, such a system could remain a researcher of value without being a final value subject.

If artificial consciousness does become possible, AI is no longer merely an external instrument studying human value. It enters the domain of inquiry as a possible new class of experience-bearing and value-bearing subjects, further weakening any reason to treat permanent human-centered privilege as axiomatic.

14. Provisional stance