Designing an Intelligence for Reality and Progressiveness
At some point, using artificial intelligence was no longer enough.
Spiral Civilization had begun as an exploration conducted through writing, questions, conversations, research, experiments and cooperation with existing artificial intelligences. As the project developed, its knowledge environment grew: concepts became connected to other concepts, experiments generated new questions, conversations produced hypotheses, and different AI systems revealed different perspectives on the same problems.
The Co-Intelligence Expedition explored what becomes possible when several intelligences participate in an investigation rather than one intelligence simply providing an answer.
But that experiment opened a more fundamental question:
What kind of intelligence do we actually need?
Not merely a more capable chatbot.
Not an artificial authority that tells a human what to do.
And not a system optimized only to produce increasingly convincing answers.
We began asking whether an intelligence could instead be designed to help recognize reality more accurately, identify what remains unknown, formulate better questions, investigate possible trajectories, examine their consequences across different perspectives and timescales, and preserve the human capacity to understand, judge and choose.
This question became Nullynity.
From cooperating with AI to designing intelligence
The transition was gradual.
Working with existing AI systems demonstrated an extraordinary expansion of individual capability. A person without a conventional technical background could investigate unfamiliar disciplines, compare bodies of knowledge, formulate experiments, work with code, test hypotheses and move between conceptual and technical levels with a speed that would previously have required a much larger team.
But greater capability also creates a new problem.
If AI increasingly performs research, synthesis, analysis and eventually parts of decision-making, what happens to the human participating in that process?
Does human capability expand — or gradually disappear?
Does AI increase agency — or replace it?
Does it help us understand reality — or merely produce an answer that sounds plausible?
For Spiral Civilization, this became a design problem.
The objective was no longer simply to use AI better.
The objective became to investigate how an intelligence system itself should be structured if its role is to increase our capacity to recognize reality and participate intelligently in its development.
Reality before the answer
One of the principles emerging from this work is simple:
Before deciding what should happen, we need to understand as well as possible what is actually happening.
A real situation contains facts, assumptions, missing information, conflicting interpretations, different participants, constraints, opportunities, histories and possible futures.
An intelligent system should therefore not move immediately from a question to an answer.
It should be able to ask:
What do we actually know?
What are we assuming?
What is missing?
Which source supports this claim?
Which observations contradict it?
Whose perspective is absent?
What question would reduce the most important uncertainty?
What could be tested?
What changed after an action was taken?
And does the new evidence require us to change our representation of reality?
This developing layer of Nullynity became connected with Reality Recognition, Reality Point and Question Intelligence.
The answer is not the beginning of the process.
Reality is.
Prototype 01 — when the idea had to work
Eventually, conceptual architecture was no longer enough.
Nullynity needed to perform a real task.
The first experimental prototype addressed a problem that already existed inside Spiral Civilization.
The project had accumulated more than 150 source files containing conversations, definitions, research, experiments and conceptual development. Relevant knowledge was distributed across this corpus. Literal keyword search was insufficient because an important idea could appear without the exact terminology later used to describe it.
The challenge was therefore concrete:
Could Nullynity find relevant knowledge across a fragmented corpus while preserving exact evidence of where that knowledge came from?
Nullynity v0.1 Experimental was built as the smallest operational system capable of performing one complete, useful and verifiable task.
Its working cycle connected project reality, knowledge discovery, processing, artifact production, verification and provenance.
And the first serious verification did not pass.
The initial structural process produced 1,168 chunks. Of these, 891 could be matched exactly to their sources, while 277 failed exact verification.
That failure was not hidden or reinterpreted as success.
The mechanism was examined.
The problem was traced to the way source text was being reconstructed. The reader was redesigned to preserve literal source slices and exact coordinates.
The test was then run again.
This time:
806 of 806 structural chunks were verified exactly.
0 hard failures.
A subsequent evidence-resolution test examined 11 evidence spans.
11 of 11 were independently verified.
0 ambiguous.
0 not found.
0 hard failures.
PASS.
This did not prove that Nullynity could determine truth, identify the best possible future or make complex decisions autonomously.
It proved something smaller — and necessary.
A real component of the proposed intelligence architecture could be implemented, fail under verification, be diagnosed, corrected and successfully retested while preserving traceability to its source material.
Nullynity had moved from an idea into an experimental system.
Prototype 02 — from finding knowledge to investigating reality
The next question is significantly harder.
Prototype 01 asked:
Can the system find relevant knowledge and preserve verifiable evidence?
The developing second-generation architecture asks:
Can an intelligence help a human understand a real situation, investigate what could happen next, and identify a better-justified direction of action?
The emerging process is not intended to jump directly from problem to recommendation.
It begins with the situation itself:
Real situation
→ Recognize the current reality
→ Separate what is known from what is unknown
→ Ask the questions that matter
→ Gather and evaluate evidence
→ Generate possible trajectories
→ Examine consequences across perspectives and timescales
→ Compare possible directions
→ Identify the next justified move
→ Observe what actually happens
→ Update the representation of reality
This is where other developing components of Nullynity enter the architecture: Question Intelligence, Reality Recognition, Multisymbiotic Systems, Progressiveness Intelligence, Best Possible, Degree of Freedom, and methods for evidence, provenance and verification.
These components are at different stages of research and development.
They should not be confused with the capabilities already demonstrated by Prototype 01.
Better according to whom?
Finding possible actions is relatively easy.
Determining whether one trajectory is genuinely better is much harder.
Better for one person may be worse for another.
A solution that works tomorrow may create a larger problem in ten years.
An intervention that optimizes one measurable variable may damage something the measurement failed to represent.
A locally successful action may reduce the freedom or resilience of the larger system.
This makes progressiveness a research problem rather than a label.
Nullynity is being designed to investigate possible trajectories across multiple perspectives, consequences and timescales rather than treating a single objective as automatically sufficient.
The question is not simply:
What works?
It is also:
What does this make possible next?
What does it prevent?
Who or what benefits?
Who or what carries the cost?
Does it increase or reduce degrees of freedom?
Does a locally beneficial move remain beneficial when the horizon expands?
And ultimately:
What evidence would make us change our judgment?
One framework, different scales
The same fundamental structure can appear at very different scales.
A person may ask:
What is the best possible use of this free evening?
Someone facing a major life decision may ask:
Where should I live? What should I learn? What trajectory would open the most valuable possibilities?
An entrepreneur may ask:
Should I build this company, change its direction, collaborate, wait or stop?
An organization may need to examine a decision involving many participants, constraints and consequences.
At much larger scales, societies face decisions whose effects unfold across decades or generations.
The scale changes.
The fundamental problem remains recognizable:
Understand reality as well as possible, investigate what is possible, examine consequences across perspectives and timescales, and search for the most progressive viable trajectory.
The current Nullynity prototype does not solve all these problems.
They define the research horizon.
Human agency is part of the architecture
Nullynity is not being designed around the assumption that increasingly capable AI should simply replace human judgment.
The opposite problem interests us.
If artificial intelligence becomes dramatically more capable, how can humans use that capability to become more capable participants in reality themselves?
Writing externalized part of human memory, but it also made libraries, accumulated knowledge and large-scale intellectual development possible.
AI may externalize additional cognitive functions.
The important question is what new human capabilities can emerge in return.
For this reason, Nullynity treats the human not merely as a source of a prompt and a recipient of an answer, but as a participant in an evolving process of observation, questioning, investigation, judgment, action and learning.
The aim is not only better artificial intelligence.
It is better Human × AI intelligence.
An open expedition
Nullynity is not a finished product.
Prototype 01 exists and has been tested.
The next architecture is under development.
Many of its central questions remain open.
How should reality be represented without prematurely simplifying it?
How can an intelligence recognize which question should be asked next?
How should conflicting evidence be handled?
How can possible futures be compared without hiding value judgments inside an optimization function?
How should short-, medium- and very long-term consequences interact?
How can an AI system increase human agency rather than gradually replacing it?
How can an intelligence learn from the difference between what it expected and what actually happened?
And how can we recognize progressiveness itself?
These are not problems to conceal until a finished product appears.
They are the expedition.
Expedition 4 begins where using intelligence becomes designing intelligence.
And Nullynity is our attempt to discover what that intelligence could become.