ALZHEIMER'S ASSOCIATION
AAIC '26

NEURO-SYMBOLIC RECURSIVE SELF-VERIFICATION FOR ALZHEIMER'S DISEASE MODELING: ANCHORING AI IN BIOLOGICAL PHYSICS
Authors: David Scott Lewis, Enrique Zueco, Haley Yi | Affiliation: AIXC Bio, Zaragoza, Spain | Email: reports@aixcbio.com

KEY TAKEAWAY: External symbolic verification via biological physics makes recursive AI self-improvement reliable and prevents predictive drift in longitudinal disease modeling.

INTRODUCTION
(Background)
Amyloid plaques (Aβ) & Tau tangles (τ)
AT(N) biomarker cascade

Modeling longitudinal AT(N) biomarker dynamics is complex.
Recursive self-improvement in AI promises capability growth, but iteratively training on self-generated data leads to recursive drift and mode collapse. This computationally demanding task requires stabilization.

AI
Training predictions -> AI model -> Generate corrupted predictions -> AI (cycle)

METHODS
(Neuro-Symbolic Recursive Self-Alignment)
(Diagram of neural network and symbolic logic gates)
DUAL-LAYER SYSTEM:
(A) NEURAL LAYER predicting longitudinal biomarker trajectories.
(B) SYMBOLIC VERIFICATION SUBSYSTEM evaluating predictions against established Ordinary Differential Equations (ODEs) for amyloid and tau kinetics.

Recursive Loop -> Predictions -> Self-Generation
Logic gate equations:
B(a) = d/d(z + e_d(z_t))
f(a) = d/d(z + e_d(z_y)^2)
a_nex = (z_t+y) / (z-z)

Symbolic Verification Gating
Accepted data (logically consistent)
Rejected data (logically inconsistent/mathematically flawed biomarker cascades)

RESULTS
(Halted Drift & Optimized Prediction)
(Graphs showing halted drift)

INTEGRATION HALTS RECURSIVE DRIFT:
Symbolic verification prevents self-improvement degradation.
63% IMPROVEMENT IN LONG-TERM ACCURACY:
Rejection of predictions violating physical/biological constraints on synthetic clinical cohorts. (Bar chart: Without vs With-Symbolic)
BIOLOGICAL REASONING:
Model differentiates sound physiological trajectories from flawed ones.
PRECISE THERAPEUTIC TIMING:
Autonomous agent optimizes intervention with unprecedented precision.

CONCLUSIONS
(Safeguarding Autonomous Modeling)
CRITICAL SYMBOLIC VERIFICATION: External verification makes recursive self-improvement measurable and reliable in clinical AI.
MATHEMATICALLY SAFE PATHWAY: Explicitly gating self-training data through biological physics enables robust, precise autonomous disease modeling.
AI -> Training predictions -> Stable -> Corrupted predictions -> AI