Human-like Artificial Intelligence: A Proof-of-Concept Relation-Centric Architecture for Systematic Generalization

This dataset accompanies the paper “Human-like Artificial Intelligence: A Proof-of-Concept Relation-Centric Architecture for Systematic Generalization” presenting HAI (Human-like Artificial Intelligence) v2.0, a symbolic AI architecture grounded in cognitive science and knowledge representation principles. HAI employs relational decomposition — representing conceptual structures through factorization into constituent dimensions — and a grade ontology of 120+ relation types to support systematic generalization, compositional reasoning, and interpretable concept formation.The dataset consists of 16 files organized as follows:Databases (6 files): LD — Learning Data: 27 unique driving scenario cases encoded as 7-column numerical tuples (Object, Place, State, Location, Direction, Speed, Activity), used for observational learning and pattern discovery. OA — Object Attributes: attribute values (size, mass, shape, material) linked to objects in LD, used during observational learning. OF — Object Features: feature values as alternating quantity-feature pairs linked to objects in LD. NC — New Cases: 12 unique test cases (6 seen objects in unseen combinations, 6 entirely unseen), reserved for self-learning and testing. NA — New Attributes: attribute data linked to objects in NC. NF — New Features: feature data linked to objects in NC. Visualizations (9 PNG files): 1_concept_hierarchy — Directed graph of concepts organized by hierarchical level (4A, 4B, 4C, 1A, 2A), with nodes color-coded by level and edges derived from grade-5 hierarchy links. 2_category_hierarchy — Graph showing categories with their member concepts (solid edges) and raw object instances (dashed edges), color-coded by category. 3_grade_distribution — Two-panel chart showing (left) the number of links per grade type and (right) the average link strength per grade, across all 120+ grades in the ontology. 4_object_features — Graph of the top 15 objects (by feature count) and their associated feature concepts, with edge width proportional to feature quantity. 5_object_attributes — Graph of the same top 15 objects linked to their attribute concepts (size, mass, shape, material), color-coded by attribute type. 6_links_sample — Full network graph of all relational links, colored by grade type and with edge width proportional to link strength. 3d_all_links — 3D map of all links color-coded by grade group (Object-Raw, Attributes, Features, Hierarchy, Semantic, Frequency, Comparison, Cross-Domain, Analogy), plotted across the six databases (LD, NC, OA, NA, OF, NF) and four layers (LAYER1–LAYER4). 3d_ld_stacked — 3D stacked visualization of the LD database across all three index levels (ID_LEVEL, RAW_VALUES_LEVEL, SEMANTIC_LEVEL) and four layers, with links color-coded by grade group. 3d_concepts_by_level — 3D vertical stack of all 26 abstraction levels (4A through 11E), with concepts shown as circles and categories as squares arranged in a spiral pattern per level, and inter-level links color-coded by relation type. Interactive Export (1 HTML file):An HTML export of the full system state, providing browser-based tabular views of all concepts (GIC_*), categories (GICAT_*), relational links (GIL_*), and raw instances (RWGI_*), along with system statistics and grade distribution. No local installation is required to explore the data.