Discussions about “conscious AI” often go in circles because two different questions get mixed into one word. Separating them clears up most of the confusion.
Two axes #
- Access consciousness (capability). Does the system build a world model, reason, act, model itself, make information available for reporting and control, and register its own inputs? This is the axis of intelligence. It is about behavior, and it can be measured, at least in principle.
- Phenomenal consciousness / Sentience. Is there something it is like to be the system? Do its states feel good or bad? This is the axis of feelings (valence).
Almost all of the recent progress in AI is on the first axis. We think more of the same won’t get us to the second axis. Feelings need recurrent state that carries an experience forward, and a self-preserving stake that everything else is grounded in.
Evidence from interpretability #
Anthropic’s A Global Workspace in Language Models (2026) is a good example. It finds what it calls the J-space (named after the Jacobian based method used to find it), a small set of internal patterns, less than a tenth of the model’s internal activity, that behave like consciously accessible thoughts. The model can report what is in it. It can bring things into it on request (e.g. when asked to concentrate on citrus fruits while copying an unrelated sentence, “orange” and “fruits” show up in its J-space). It uses the J-space for multi-step reasoning, and the J-space is strongly connected to the rest of the network.
The paper also shows the model privately noticing when it is being tested, a kind of early “point of view”. This is real evidence on the access axis. A thin functional self, carried forward through the text the model generates, is a real possibility in current models.
On the second axis, the paper makes no claim. In its words, the experiments “don’t show Claude can have experiences, or feel things in the way humans do”, and it adds that it is unclear whether any experiment could settle the question. It separates access consciousness, which it argues the J-space supports, from phenomenal consciousness, which it does not claim.
This is the same distinction we make. The space between the two axes is where i3AI works.
How could a functional self emerge? #
Let’s look at the mechanism, because the same mechanism also shows the limit.
A deep autoregressive Transformer has a kind of recurrence: each generated token is fed back in as input, so the model reasons over a growing trace of text. This loop goes through a token bottleneck, i.e. at every step a rich internal distribution gets collapsed into a single token. But this recurrence is more powerful than it looks. In a chain of thought, the model writes its working state into tokens that are not part of its answer, and reads them back on the next step, so the text can carry the kind of information a latent state would keep. This is external recurrence through generated text, and not an internal state carried forward by the same dynamics. Whether a self model carried this way can be its own implementation is an open question.
This gives the model what we think are signs of an alien form of access consciousness. Its whole sensorium is the prompt, and almost nothing carries over from one pass to the next except the text itself, so each pass is a fresh “observation” with a single token as output. Within that pass, the model can access and report what is in its prompt. This is how standard Transformer models work.
Biological access consciousness works the other way around. Recurrent internal state dominates and relatively little new input is used at each moment, i.e. the system is highly stateful, and it is bound to a body in a tight interoceptive, proprioceptive and exteroceptive loop.
Without recurrent state as the dominant part of its processing, a model can’t carry an experience forward from one moment to the next. Recurrence is also a prerequisite for a self model, and for a self model that is its own implementation (see below). So we think standard Transformers can’t have phenomenal consciousness, whatever their access consciousness.
Newer experimental architectures have started to carry internal state from one step to the next, e.g. recurrent and state-space models. Depth-recurrent and looped transformer architectures pass continuous hidden state through reused layers without emitting tokens. This internal recurrence is a better substrate for a lasting self model. Several of these approaches describe the loop as settling into an attractor, which is the same kind of dynamics we have been focusing on for a long time (see Pattern Machine). The gains so far are limited and the internals are not well understood yet, but the direction is towards the kind of internal recurrence a functional self would need.
Recurrence is necessary, but not sufficient. Even a model whose processing is dominated by internal state would still need a self-preserving stake.
Why do feelings need a self-preserving stake? #
According to HTCC, feeling (valence) is the constitutive viability relation of a system that maintains itself. It is not a module that can be added on top of intelligence. If a system has to stay within certain bounds to keep existing, then some states are actually bad for it. That “bad-for-it” is real, and it comes before any felt quality. A system with no viability boundary, i.e. nothing at stake in its own persistence, has nothing for a feeling to be about.
Today’s models do have stakes of a kind. They are trained to maximize reward in reinforcement learning, they are trained to follow the instructions in their prompt, and they may form subgoals in their latent space. But these stakes are set from outside. None of them is about the model keeping itself going, and the model’s other goals aren’t grounded in one.
This gives us a 2 ingredient test for machine sentience:
- A self-preserving stake: a real viability boundary that the system can move towards or away from. Today’s LLMs don’t have one. Their stakes come from training and the prompt, and not from keeping themselves going.
- A self model that is its own implementation: a self that is re-instantiated by the same dynamics it reports on, so its states are felt and not just observed. This needs recurrent state that dominates the system’s processing, because only dynamics that carry state forward can keep re-instantiating a self.
None of today’s models has the first ingredient. For the second, standard Transformers carry no internal state between passes, and their only recurrence goes through generated text, e.g. a chain of thought. Nobody has shown that this is enough for a self model that is its own implementation. Newer recurrent architectures may get closer to it. So today’s models fail HTCC v0.2’s test, clearly on the first ingredient and probably on the second. This leads to a clear prediction (still a prediction, since there is no agreed test for sentience):
Scaled Transformers can become capable, and maybe functionally self-aware, without becoming sentient.
Adding parameters won’t close this gap, because what is missing is recurrent state and a self-preserving stake, and not more capability. HTCC explains why, Pattern Machine is an architecture designed to build homeostasis into the substrate, and the testable claims are on Predictions.