Modern AI and machine learning are making extraordinary advances and rapidly reshaping science, industry, culture, and everyday life. The emerging knowledge economy is increasingly organized around the extensive use of AI systems, for better or worse depending on who designs them, controls them, and benefits from them.

Yet current AI systems are not automatically reliable, transparent, benevolent, or aligned with the long-term well-being of individuals and societies. Their increasing ubiquity raises a deeper question: how can we understand, evaluate, and govern systems that increasingly mediate what we see, believe, value, remember, and decide?

The danger is not only technical failure. AI-driven technologies can also reshape human behavior at scale, encouraging shallow and reactive patterns of attention at the expense of some of our most valuable capacities: creativity, agency, love, curiosity, questioning, critical thinking, and intellectual independence. 

When digital environments are designed primarily to maximize engagement, speed, conformity, or short-term reward, these capacities may become inconvenient rather than valuable. A society increasingly dependent on opaque algorithmic systems risks a gradual erosion of human agency: the very human abilities required to question, understand, and govern those systems.

We therefore need more than increasingly powerful pattern-recognition engines. We need new forms of trustworthy, aligned and transparent AI and new mathematical scientific frameworks capable of helping us understand AI representations themselves: how they are formed, structured, transformed, constrained, observed, reconstructed, and related to the dynamical physical and human realities they are intended to represent. 

As in modern physics, the combination of abstract geometry and algebra in FTR is essential. Geometry provides a language for continuity, distance, trajectories, curvature, and reconstruction. Algebra provides a language for transformations, composition, symmetry, invariance, and structure. Together, they offer a richer scientific vocabulary for understanding latent representations than coordinates alone.

From the geometric perspective, partial observation creates fibers of latent ambiguity: different internal states may correspond to the same observable state. Reconstruction therefore requires additional structure. A metric can distinguish meaningful directions and distances; a connection can specify how observable changes should be lifted into latent space; curvature and holonomy can reveal when reconstruction depends on history rather than observable endpoints alone. From the algebraic perspective, operators, transformations, symmetries, invariants, and constraint relations can determine which changes are meaningful, equivalent, admissible, or prohibited.

The aim is not to reject statistical learning or modern AI/ML. It is to provide additional structure for understanding what such systems learn and how their internal representations behave. The need for this structure becomes increasingly urgent as AI systems move from narrow prediction toward systems that interact with human beings, physical environments, institutions, and other AI systems over long time horizons.

We need AI that can do more than generate plausible responses. We need systems whose representations can be investigated, constrained, tested, reconstructed, and, where possible, understood. We need a mathematical scientific framework capable of distinguishing what is observed from what is inferred, what is geometrically determined from what depends on additional assumptions, and what can be reconstructed from what remains fundamentally ambiguous.

[5] Ed Siregar. "SN-Hamiltonian Dynamics with Epistemic-Utility-Normative Alignment: Towards Safe Cognitive Robotics", Journ of Robotics & Autom. 4(5): 2026. OJRAT.MS.ID.000596.

[7] Ed Siregar, "Can Artificial Intelligence Be Designed to Care? Structural Principles for Human-Centered AI in the Knowledge Economy", Invited Paper, The 11th Annual Congress Knowledge Economy: AI and Knowledge Automation, July 2026, Helsinki. Note: a mathematical framework to study and design latent space geometries, for mission-critical trusted AI.