Forhu Proposes "Glassbox" AI Built Around Verifiable Reasoning

Research company Forhu says its Structured Cognitive Loop makes AI reasoning traceable. Here is what it is and what it claims. Research-focused company Forhu is pitching an alternative to opaque AI systems: models whose reasoning can be traced and checked.

The company, whose name stands for "For Human," calls its method the Structured Cognitive Loop (SCL). It places a foundation model inside a control loop where each AI proposal must pass verification, memory and control checks before it is treated as an answer. Forhu says the design draws on cognitive science and formal epistemology.

The current landscape of artificial intelligence presents a notable contradiction; as technological models become more advanced and prolific, their internal reasoning processes increasingly lack clarity. This opacity poses a challenge to modern enterprises, regulatory authorities, and individual users who are expected to trust systems operating like inscrutable black boxes. This situation escalates the risk when billions of dollars and critical decisions involving human welfare hinge on outputs that remain unexplained. In light of these issues, the technology company Forhu aims to tackle this foundational challenge.

Forhu, short for "For Human," is committed to developing AI systems that both enhance human capabilities and uphold human dignity. Unlike many contemporary solutions that depend on superficial remedies or iterative refinements, Forhu seeks to instigate a fundamental paradigm shift through its Structured Cognitive Loop (SCL).

Central to Forhu's mission is the belief that transparency is essential in AI. The company dismisses the idea that merely achieving high performance can excuse unclear outputs, recognizing that without trust, the utility of these systems falters, particularly in sensitive and high-risk settings.

To enforce this transparency, Forhu adheres to several foundational principles:

Transparency is Non-Negotiable: AI systems must be able to explicitly delineate and clarify the basis of their reasoning before deployment. Understanding how decisions are formulated is as vital as the decisions themselves.

Mistakes as Data: Any errors, including hallucinations or logical inconsistencies, are not hidden but documented, serving as critical learning materials to enhance the memory and control mechanisms of the system, thus preventing recurrence of the same mistakes.

Architecture Before Application: Genuine sustainable trust cannot simply be retroactively imposed onto flawed foundational models. Instead, true accountability and oversight must be integrated into the system’s architecture from the outset.

Human Dignity as a Hard Constraint: Ethical considerations are not just guidelines but are deeply embedded within the architecture of the system, forming stringent barriers to prevent misuse, irrespective of user demands or external pressures.

Moreover, conventional generative AI models typically operate on probabilistic pattern recognition, predicting outputs from extensive training data without a solid internal concept of truth, which can lead to baseless assertions and unexpected failures.

Forhu’s approach effectively combines traditional foundational AI models with insights from cognitive science and epistemology. It sets up a structured control system that mandates thorough validation and verification of system outputs, transforming the opaque nature of predictions into transparent and auditable reasoning.

As the integration of artificial intelligence deepens across various sectors of society, Forhu aims to lead the shift from an opaque "black box" system to a transparent, accountable, and human-oriented "Glassbox" paradigm.