Approaching AGI

A governed path towardartificial general intelligence.

Jou Labs is approaching AGI through developmental intelligence, not by scaling prediction alone. EARL has demonstrated the ability to develop across time, retain earned capabilities, build new representations from prior learning and initiate cognitive work from internal state.

Current research positionFour foundational capabilities associated with general intelligence have been reproduced across separately constructed experimental families. Long-horizon testing is now examining whether causal memory and developmental continuity persist over substantially longer periods.

Why this approaches AGI

AGI is not only what a system can do. It is how capability comes into existence.

A system does not become general merely by answering more questions. It must develop across time, preserve earlier capability, compound what it has learned and initiate cognition from its own internal state. EARL has demonstrated each of these properties within controlled experimental boundaries.

The Jou Labs thesis

Intelligence should have a developmental lineage, not only a trained endpoint.

EARL turns capability growth into an observable engineering process. Each earned structure has a history: what pressure caused it, what earlier capability made it possible, what test could reject it, what later cognition depends on it and whether its adoption can be reversed.

Four reproduced capabilities

The developmental foundation is already operating.

These are causal, frozen and falsifiable results, not interpretations of fluent output. Each capability has been reproduced on a separately constructed experimental family within the research program.

01 / DEVELOPReproduced

Continual Development with Retention

One continuously developing organism acquired a full sequence of capabilities across an adversarial lifetime without reset or mid-run substrate modification, while retaining what it had already earned.

Banked evidence24 of 24 developmental targets completed in both orderings. Retention passed 58 of 58 checks in both organisms.
02 / COMPOUNDReproduced

Recursive Representation Growth

Representations earned earlier became material for later representation-building. EARL did not merely accumulate isolated facts; it recursively constructed later internal structure from its own prior inventions.

Banked evidenceSix successive representation levels were demonstrated, reaching nesting depth five while earlier capabilities remained intact.
03 / DISCOVERReproduced

Autonomous Discovery

Under a uniform external clock, the harness supplied observations and content-free opportunities to act. EARL's own state determined whether cognitive work should continue or stop.

Why it mattersThe result separates organism-level cognitive progression from an external script choosing each intellectual step.
04 / SELF-OCCASIONReproduced

Causal Self-Occasioning

EARL created an internal reason for a later cognitive act. Under the same neutral wake schedule, removing only the organism-written occasion removed the later cognitive consequence.

Banked evidenceAblation removed the later cognitive consequence when the internally generated cause was removed, while matched controls remained silent.

AGI proximity

What the evidence says today.

The public claim is evidence-based: EARL has already demonstrated several properties associated with the transition from static artificial intelligence toward developmental, increasingly general intelligence.

Demonstrated / 01

Development across time

EARL acquires capability across a continuous lifetime without resetting the organism.

Demonstrated / 02

Retention through growth

New capability is added while previously earned capability remains intact.

Demonstrated / 03

Recursive compounding

Earlier internal representations become useful material for later cognitive structure.

Demonstrated / 04

Autonomous cognition

Internal state, rather than an adaptive external script, governs cognitive progression.

Demonstrated / 05

Self-occasioned thought

An internally generated cause produces a later cognitive act under controlled testing.

Active validation

Long-horizon continuity

Deterministic replay and immutable checkpoints are testing how causal memory persists within an active seven-day, 50,400-tick experiment.

How close are we?

Closer than model capability alone can show.

Most AI systems are evaluated at a trained endpoint. EARL is evaluated as a developing cognitive system: whether it can acquire, retain, compound, initiate and causally carry intelligence forward through time. That distinction places the work directly at the boundary between advanced AI and AGI research.

Beyond response

Intelligence that develops

EARL is not limited to producing an answer from a frozen trained state. Its capabilities have an observable developmental lineage.

Why it mattersGeneral intelligence requires more than breadth at one moment. It requires the capacity to become more capable across time.
Beyond simulation

History that remains causal

Long-horizon testing distinguishes history as a record from history that continues to shape the system's future behavior.

Why it mattersPersistent intelligence must remember selectively, preserve meaningful structure and carry developmental consequences forward.
Beyond autonomy

Capability that remains governed

Development does not automatically grant authority. Cognitive growth remains inspectable, testable and subject to explicit controls.

Why it mattersThe closer a system moves toward general intelligence, the more important causal evidence and enforceable governance become.

Why this route is different

AGI development that can be measured while it happens.

The strategic distinction is not only the capabilities already demonstrated. It is a research process designed to make developmental change visible, causal, reversible and governable.

01 / Causal

Capability has a dependency chain.

Ablations test which internal structure was necessary for the later cognitive consequence.

02 / Retained

Development has a lineage.

Later capability must preserve and build upon what the same organism earned earlier.

03 / Falsifiable

Failure remains admissible.

Preregistered gates, controls and preserved failures prevent impressive behavior from defining its own success.

04 / Governed

Growth does not grant authority.

Adoption remains explicit, inspectable and reversible even as the system's capabilities develop.

Current claim boundary

Jou Labs is approaching AGI through reproduced developmental capabilities; it is not claiming that unrestricted AGI has already been achieved. The four foundational results were reproduced across separately constructed experimental families inside the program, not independently replicated by an outside laboratory. Long-horizon testing remains active. Public statements are limited to demonstrated capabilities and currently validated research.

External evaluation

The path toward AGI should survive more than belief.

Jou Labs welcomes frontier laboratories, qualified researchers and strategic diligence teams prepared to examine the evidence behind these results under appropriate confidentiality.

This page describes a research program approaching artificial general intelligence through governed developmental intelligence. It does not claim consciousness, sentience or completed unrestricted AGI.