Why consciousness by construction?
There is no single accepted scientific theory of consciousness.
Different theories explain consciousness in very different ways. Some focus on information integration, some on global access, some on higher-order thought, prediction, recurrence, or other mechanisms. They do not fully agree on what consciousness is, which processes are necessary for it, or how it should be measured.
This creates a difficult problem for artificial consciousness. If there is no agreed definition and no accepted instrument that can tell us whether an artificial system is conscious, then there is no simple test that an AI system can pass.
EmoStack takes a different approach.
Instead of starting from one theory and asking whether a system satisfies it, I start from a constructive question:
What would we have to build for an artificial being to have something that functions as its own present?
This idea is closer to the question raised by Philip K. Dick than to a claim that one scientific theory of consciousness has been proven correct.
Dick's androids are interesting not only because they speak well or imitate people. They have their own thoughts. They have emotions connected to what happened to them. Their past changes how they react now. They form beliefs about themselves. They can think when nobody is asking them a question. And they themselves are uncertain about what all this means.
This is the starting point of EmoStack.
The system is built around several processes that work together. Events are interpreted emotionally. Past experience remains active in the present. Memories can return because of what they meant and how strongly they were felt. The system can form thoughts, beliefs, decisions, goals and dreams outside the direct conversation. It can also learn dispositions from experience: ways of reacting that affect later behaviour without having to appear as an explicit thought.
None of these mechanisms alone is meant to be consciousness.
The important part is their organisation.
At every moment they create a changing internal context: what has just happened, what is still emotionally active, what memories return, what the system currently believes about itself, and what it has learned to do in similar situations. This context changes how the language model responds.
In EmoStack, the language model is therefore not treated as the being itself. It is closer to a processor. It provides language and a large cultural prior. The layer around it continuously changes the context from which that processor produces an answer.
I call this underlying prior the averaged author. A language model contains patterns learned from an enormous amount of human writing. EmoStack does not replace those patterns. It narrows them. The history and current state of one artificial being make some continuations more natural than others. The same processor can therefore behave differently because it is placed inside a different history.
This is also why EmoStack is not mainly a memory system.
A database can store everything that happened. A transcript can contain every previous sentence. But having a history is not the same as being in the state that this history produced.
The central question is whether a system can carry its own past into the present in such a way that it affects what it notices, remembers, thinks and does next.
The experiments in EmoStack do not prove subjective experience. They cannot. There is currently no accepted test that could do that.
What they can test is the construction itself.
- Does a past emotional event change later behaviour?
- Can a thought formed in silence remain active after the events that produced it have faded?
- Can something the system wrote about itself later influence its actions?
- Can learned dispositions affect behaviour without becoming explicit reasoning?
- Can the same language model behave differently when these processes are present than when it only receives the conversation transcript?
These are smaller questions, but they are measurable.
So the claim of EmoStack is not that the problem of consciousness has been solved.
The proposal is different: artificial consciousness may be approached by construction.
We can choose a clear set of processes, explain why they belong together, implement them, observe what kind of being they produce, and test whether each part does what the construction predicts.
Whether such a sheep should finally be called conscious is a deeper question.
But before we can answer it, we may first need to build something that has its own past, its own present, and thoughts that are not only answers to us.
Eliza Marszalik
