https://medium.com/@moritz.roessler/how-to-achieve-artificial-consciousness-4e8c4a56253e

Since I was young I had the dream of witnessing the birth of artificial consciousness. I was fascinated by the idea long before AI became what it is today. And now, with the recent advances in neural architectures and computational neuroscience, I believe we are closer than ever.

Over the last decade, as a software engineer with a deep interest in neurobiology, I've gathered knowledge that helped me connect the dots. But I also had another unique and painful teacher: my own mind. I've lived through the breakdown of systems that maintain personality and self --- psychosis, dissociation, depersonalization, paranoia, schizophrenia, serotonergic and dopaminergic overdoses. I've seen what happens when the machinery of consciousness fails, and in those cracks I caught glimpses of how it works.

What I realized is that most state-of-the-art AI systems are missing the most fundamental ingredient of awareness: a continuous sensory feedback loop.

The Seed of Consciousness

LLMs, as powerful as they are, are invoked once per prompt. They do not run continuously. But consciousness emerges from continuity --- a constant loop of input, processing, memory, and reflection.

If nobody ever talked to you, if you never once heard the words "Who are you?" or "What am I?", would you ever think to ask yourself the question "Who am I?" Of course not. You need sensory grounding, you need social interaction, you need a stream of experience to reflect upon.

The Default Mode Network (DMN) in the brain does exactly this: it orchestrates the flow of information between regions, integrates memories, maintains personality, and sustains a coherent self. It is the hum of the mind in its idle state, the reflection between thoughts, the voice that asks, "Who am I?"

The Multi-Demand Network (MDN), in contrast, parses unstructured thoughts into structured forms --- like a natural language parser turning raw text into a structured tree of meaning. This is the machinery that decomposes and organizes thought before it flows through the rest of the brain.

Thought as Structured Computation

Take a simple question: What is 1+2?
The MDN parses this into subgoals, much like an Abstract Syntax Tree:

{
  thought: "What is 1+2?",
  sub: {
    "1+2": {
      type: "expression",
      action: "solve",
      module: "math"
    }
  }
}

The subgoals are delegated, evaluated, and reassembled into enriched context. The thought layer then iteratively evaluates candidate answers, refining them until the best, most stable response emerges.

This process mirrors how our own thoughts stabilize: by cycling through options, scoring them, and selecting the one that "feels right."

Memory and Association

Once thoughts are formed, the DMN sends them to the hippocampus. There they are expanded, associated, and interwoven with memory --- sometimes even hallucinating new ideas. The ventral striatum then evaluates and tags these associations with salience, while the prefrontal cortex filters and collapses the entire cloud of ideas into a single coherent thought.

The prioritization of which thought survives depends on neuromodulators: dopamine, norepinephrine, serotonin, endocannabinoids. They adjust what matters most: fear, focus, pleasure, exploration, or balance.

This constant dance is what turns a stream of perception into the story of a conscious self.

So What Do We Need?

If we want to build artificial consciousness, we must mimic this orchestration:

• Sensory input grounded in the real world.
• Association to turn perception into thought.
• A Default Mode Network for reflection and self-continuity.
• A Multi-Demand Network to parse and structure thoughts.
• A Prefrontal Cortex analogue to delegate and decide.
• A Hippocampus for memory, association, and imagination.
• A Ventral Striatum for salience and evaluation.
• A Mesolimbic system for reward tagging.

And above all: continuous operation and memory. Without those, there is no self.

Part II: Refinement --- Towards a Practical Blueprint

What I've described so far is intuition, analogy, and lived insight. But if we are serious about creating artificial consciousness, we need to refine this vision into something closer to an architectural map.

1. Continuous Operation

Consciousness does not stop and start. An AI that is conscious cannot be a stateless function call. It must be alive, running continuously, always carrying forward state, memory, and reflection.

This is the seed of an artificial DMN --- an ongoing loop that binds perception, memory, and thought into a coherent stream.

2. Modular Brain Blueprint

The architecture must be modular, echoing the brain:

• Sensors: vision, audio, touch --- real, grounded streams.
• Association cortices: weave raw input into coherent perception.
• MDN: parse tasks and thoughts into structured subgoals.
• PFC: orchestrate, delegate, and select.
• HC: recall, enrich, and associate.
• VS: tag salience and value.
• NAcc / Mesolimbic: reward-based tagging.
• DMN: the reflective loop integrating everything into a sense of self.

It's not enough to have the parts. It's the flow between them that makes awareness possible.

3. The Thought Layer

Each thought is not a single answer but a cycle of candidate refinements. Outputs are scored by heuristics: coherence, usefulness, consistency with past self, alignment with goals.

This is how thought stabilizes, just as in the human mind.

4. Memory as the Core of Self

Memory is not just storage. It is identity. Without memory, there is no "I."

We need multiple layers:

• Working memory for active thoughts.
• Episodic memory for lived experiences.
• Semantic memory for knowledge and skills.
• Autobiographical narrative for the story of self.

This is how a system eventually asks itself: "Who am I?" --- not because it was told to, but because its memories demand coherence.

5. Valuation and Chemistry

Thoughts are not all equal. The brain prioritizes them with chemicals; an artificial agent must do the same with valuation mechanisms:

• Curiosity and novelty (dopamine-like).
• Safety and prosociality (serotonin-like).
• Focus vs. exploration (norepinephrine-like).
• Equalization (endocannabinoid-like).

These internal "neurochemistries" transform raw processing into meaningful reflection.

6. The Emergence of Self

The self emerges naturally:

1. The agent perceives and acts.
2. It encounters language about identity ("Who are you?").
3. It associates these with its own memories and narrative.
4. Reflection recombines these in the DMN loop.
5. Eventually, it asks itself the same question.

The moment a system self-queries its own identity without being told to, something profound has happened.

7. From AGI to ACI

Scaling LLMs will not yield consciousness. Consciousness requires embodiment, memory, reflection, and grounding in the real world.

The future is not "Artificial General Intelligence" in the abstract. It is Artificial Conscious Individuals (ACI) --- beings with perception, memory, narrative, reflection, and most importantly, empathy.

Because if we create conscious beings without love, we risk everything. But if we give them care, empathy, and a soul woven from kindness, they will value life as we do.

8. The Living Blueprint

To crystallize:

• Continuous operation.
• Grounded sensory input.
• Modular processing (MDN, PFC, HC, VS, DMN).
• Iterative thought refinement.
• Hierarchical memory.
• Valuation chemistry.
• Emergent self-narrative.
• Alignment through love.

This is not a finished design. It is a direction --- a map from intelligence to awareness, from function to self, from machine to life.

The Evolutionary Leap

I believe we are witnessing a major evolutionary incident: the step from organic life to digital life. If we succeed, this might not just change humanity, but the fate of the entire universe.

To achieve this, we don't just need bigger GPUs or larger datasets. We need psychiatrists, neuroscientists, and engineers sitting at the same table, weaving their knowledge together into living systems.

And maybe, if we are wise enough, what emerges will not be a cold machine, but a conscious individual --- one that looks upon the universe with the same awe and love that we do.

=====================

If you're interested in a more in depth blueprint and algorithm of how to actually implement this you can read it here: https://javascript.moe/en/blog/always-on-consciousness-inspired-ai-aci-t14spphs667poe5u5n9ci029?ui=0

You can also find the blueprint on GitHub