Personal Intelligence

Author: Lin K. Darren Soo

Personal Intelligence (PI) is a causal reasoning framework designed to support individual-level inference under sparse, irregular, real-world data. It addresses a structural limitation in population-trained AI systems: where such systems perform well on average, but they are not designed to reliably reason about a single individual when responses diverge from population norms.

PI is not a personalization layer, a recommendation heuristic, or a population model refinement. It is a distinct reasoning layer built to operate under non-ergodic conditions where individual trajectories cannot be inferred from group statistics.

Motivation

Most AI systems follow a familiar pipeline:
population data -> prediction -> personalization -> guidance

This pipeline assumes that individual behavior can be inferred by conditioning population models. In real-world health and behavioral settings, this assumption often breaks down. Individuals may respond differently, or oppositely, to the same intervention, despite similar demographic or clinical profiles.

PI was developed to address this failure mode by enabling reasoning from the individual's own evidence rather than from adjusted population averages.

Core Principles

Personal Intelligence is built around four core principles:

  1. Individual-First Reasoning
    Inference is derived from an individual’s longitudinal data, not from cohort-level priors.
  2. Non-Ergodicity Awareness
    Population averages are not assumed to represent individual dynamics.
  3. Structural Constraints
    Physiological, behavioral, and temporal constraints guide inference to avoid overfitting sparse data.
  4. Sufficiency-Gated Action
    Recommendations are produced only when sufficient individual evidence exists. When evidence is insufficient, the system explicitly refrains from action.

Architecture Overview

PI is structured as a layered reasoning framework:

  • Statistical Preview (Phase 1):
    Extracts early signals from sparse individual data without assuming stability.
  • Structural Discovery (Phase 2):
    Identifies interpretable individual-level structure under real-world constraints.
  • Causal Confirmation (Phase 3):
    Evaluates candidate relationships for causal consistency at the individual level.
  • Adaptive Optimization (Phase 4):
    Adjusts interventions over time based on individual response and sufficiency thresholds.

This architecture is designed to prioritize safety, interpretability, and individual validity over aggregate performance metrics.

Publications

The following papers describe the PI framework and its empirical feasibility:

Personal Intelligence: A Causal Framework for Human-Centric Health AI

Introduces the conceptual architecture and motivation for Personal Intelligence, focusing on individual-level reasoning and causal structure.
DOI:  https://doi.org/10.5281/zenodo.17532980

Empirical N-of-1 Statistical Preview Using Real-World Wearable Data

Demonstrates the feasibility of extracting interpretable individual-level structure from sparse, irregular wearable data using Phase-1 PI methods.
DOI: https://doi.org/10.5281/zenodo.17653378

Personal Intelligence for N-of-1 Causal Inference Under Sparse, Irregular Real-World Data

Provides justification for sufficiency-gated individual-level causal reasoning and discusses deployment constraints in real-world settings.
DOI: https://doi.org/10.5281/zenodo.17932200

Applications

Personal Intelligence is a general framework and is not tied to a specific product.
It can be applied to domains where individual-level reasoning is required under sparse or non-stationary data conditions, including health, behavior, and adaptive decision systems.

Trillies+ is one example application that implements PI principles in a consumer health context.

Status

Personal Intelligence is an active research and development effort.
The framework and publications are publicly available for reference and evaluation.