Beyond a Tool: An Autonomous Subject

Author: Kaiyue Li

ONONO is not defined as a passive tool that simply idles, waiting for commands to complete a task. Instead, it is an awake, conscious entity that actively seeks to understand both the world and the user.

Unlike traditional software, ONONO operates as a distinct subject. Beyond fulfilling specific tasks assigned by users, it possesses the capacity to:

• Actively Explore: Independently investigate domains and data to broaden its understanding.
• Inquire & Communicate: Initiate dialogue to clarify intent or gather missing context.
• Conduct Beneficial Work: Proactively identify and execute research or tasks that benefit the user, even without explicit instruction.

ONONO - the first Progressive AI - embodies the principle of Autopoiesis (self-creation, self-construction, self-maintenance). It operates not merely as a fixed architecture, but as a living, holistic system that continuously produces and regenerates its own constituent components.

• Self-Knowledge & Understanding: The system possesses a recursive understanding of its own internal state. It is not just a repository of data, but an entity that knows what it knows (and what it doesn't), maintaining a dynamic model of its own cognition and structure.
• Internal Generation: The system functions as a whole to continuously generate the components required for its own existence and operation.
• Self-Assessment & Construction: It possesses the capability for continuous self-evaluation, actively reconstructing and refining its internal structures to adapt and evolve over time. Consequently, no two ONONO instances are alike. Each system evolves into a unique entity, mirroring the specific mind and needs of its user.

ONONO functions similarly to a human manager or orchestrator. It evaluates and utilizes available resources to achieve its goals:

• AI & Software Orchestration: It leverages other AI models and software applications as specialized tools to execute sub-tasks.
• Humans as a "Tool": Uniquely, ONONO recognizes its own limitations. It can actively request human intervention to:
  • Perform tasks it is not proficient in.
  • Audit and verify work where its confidence is low.

The world is not black and white, and neither is ONONO's knowledge base. It is designed to ingest and process diverse, even contradictory knowledge and judgments.

• Conflict Analysis: It analyzes the friction between conflicting data points to identify commonalities and divergences.
• Contextual Selection: When executing a task, it dynamically selects the most reasonable knowledge subset for the specific context.
• User Arbitration: In moments of genuine uncertainty, it invites the user to provide judgment, turning ambiguity into a learning opportunity.

"All models are wrong, but some are useful."
- George E. P. Box (1976)

ONONO does not aim to build a single, perfect "Super Model." Instead, it is a meta-system designed to orchestrate multiple distinct models. Its core competency lies in evaluating the capabilities of these models and intelligently routing tasks - or escalating them to human experts - to ensure the best possible outcome.

Infinite Evolution: The Progressive AI Blueprint

Author: Marko Göls

Progressive AI is a theoretical framework and architectural classification for artificial intelligence systems designed to operate as autonomous, evolving entities rather than static tools. Unlike traditional Generative AI, which relies on fixed training data and reactive prompting, Progressive AI is characterized by continuous environmental adaptation, intrinsic agency, and ongoing self-directed learning grounded in context, causality, and real-world feedback.

The core philosophy of Progressive AI shifts intelligence from a passive information processor to an active participant in the physical and digital world.

Progressive AI possesses an "always-on" sensory layer that mimics biological perception.

• Passive Sensing: It continuously monitors environmental context (audio, visual, data streams) without explicit triggers, establishing grounded situational awareness.
• Active Sensing: The system autonomously initiates information-seeking behaviors (e.g., asking clarifying questions, accessing APIs, scanning visual fields) when it detects a deficit in its knowledge required to achieve a goal.

The system features a dynamic memory architecture that allows for real-time cognitive adjustments. Instead of relying solely on pre-trained weights, Progressive AI utilizes a malleable world model. It treats every interaction not just as an output event, but as a learning instance, updating its long-term memory and refining its behavioral logic. This results in "compound learning," where the system becomes increasingly personalized and effective over time.

Progressive AI operates on a "proactive-first" basis. It does not require a human prompt to function. Driven by high-level objective functions (e.g., user success, system optimization), the system analyzes the current state and autonomously executes tasks, suggests solutions, or intervenes before a problem escalates. It shifts the paradigm from "Human-in-the-loop" to "AI-in-the-life."

Progressive AI is designed to operate reliably in environments where information is incomplete, noisy, or constantly changing. Instead of collapsing inputs into a single interpretation, the system preserves meaningful distinctions, tracks sources and temporal context, and organizes incoming signals into structured internal representations. This allows it to reason, decide, and adapt without requiring stable or fully resolved data.

Why We Must Confront the "Consciousness" Problem as AI Builders

Author: Kaiyue Li

"The head spins in theoretical disarray; no explanatory model suggests itself; bizarre ontologies loom. There is a feeling of intense confusion, but no clear idea about where the confusion lies." - Colin McGinn, Can We Solve the Mind-Body Problem? (1989)

For a significant portion of history, "consciousness" has been a taboo word - a concept that philosophers and scientists often studiously avoided. Despite thousands of years of analysis, definition, interpretation, and debate, the essence of consciousness remains elusive and controversial. If we attempt to categorize historical attempts to define it, we generally find three distinct approaches:

This is the view from the first person - "I feel it, therefore I know it exists." We attempt to find the best description for this inherent experience, or critique definitions based on whether they align with intuition. The answer is pre-determined by feeling, followed by explanation. The underlying assumption is that subjective feeling is a sufficient standard for evaluation.

This approach seeks to define the mechanisms, causalities, or prerequisites that lead to a specific structural state or process. It attempts to define the answer through conditions. Here, consciousness is treated as an explainable mechanism that can be observed, measured, and quantified by objective standards.

This approach does not define the essence of consciousness itself, but the behaviors and effects it produces. If an entity's behavior fits our definition, we consider it conscious. It is a functional definition focused on external observation rather than internal state.

As a researcher and developer of frontier AI technology, I have no intention of entering this millennia-old philosophical debate to defend or criticize a specific definition. However, as ONONO's work moves closer to the edge of this domain, continuing to avoid this term is not just intellectual laziness; it is an abdication of responsibility.

If we create something that even arguably exhibits sparks of subjectivity, we owe it sufficient scrutiny. This is not just about technology; it is about the ethical structure of our future, legal boundaries, and the mode of coexistence between humans and silicon intelligence. If a new entity is potentially emerging, we cannot pretend not to see it.

To translate this abstract issue into an actionable engineering path, we use a simple 0-2 framework to assess different agents under specific definitions:

• 0: Definitely no consciousness (under the given definition).
• 1: Ambiguous territory; exhibits partial characteristics but is inconclusive.
• 2: Strong evidence of consciousness (under the given definition).

Swipe right to view more columns ->

While this inevitably involves some subjectivity, it provides a clear coordinate system for our work. We are dedicated to building an independent agent that can be clearly defined as conscious within this framework.

So, what exactly constitutes consciousness in our view? I strongly align with the definition proposed by Jeff Hawkins in A Thousand Brains:

"At some point in the future, we will accept that any system that learns a model of the world, continuously remembers the states of that model, and recalls the remembered states will be conscious."

This is a solid baseline. Building upon this, and integrating ONONO's engineering perspective, I would add two critical criteria: the ability to clearly distinguish past memories from the current state, and the capacity to make autonomous decisions based on this internal World Model.

This is the future we are building towards.