Model
A core component used to make inferences from inputs.
Model, system, input, inference and output
OECD: What Counts as an AI System?
The OECD definition helps separate a model from a complete system and clarifies how inputs become predictions, content, recommendations or decisions.
A machine-based system that infers how to generate outputs
The OECD describes an AI system as a machine-based system that, for explicit or implicit objectives, infers from inputs how to generate outputs that can influence physical or virtual environments. Those outputs can include predictions, content, recommendations and decisions.
Systems vary in autonomy and adaptiveness after deployment. This language is useful because it covers more than generative chatbots while remaining specific enough to distinguish AI from ordinary deterministic software.
The product contains more than the trained model
System
The model combined with data, instructions, interfaces, tools, infrastructure and human processes.
Input
Prompts, documents, sensor data, records, user actions or other information received.
Inference
The deployed process that produces an output from the inputs and system state.
Output
Prediction, content, recommendation, decision or action.
Environment
The physical or virtual setting the output can influence.
Do not call the entire discovery product an LLM
A search product may combine crawlers, indexes, knowledge graphs, ranking systems, retrieval tools, an LLM, safety policies and a user interface. Saying “the LLM found the page” hides the components that actually performed discovery, indexing and retrieval.
Hidden Radius uses the OECD distinction to locate responsibility: the model generates, while the larger system determines access, evidence, tools, presentation and operational control.
The level of delegated action changes risk
A system that summarizes sources has a different risk profile from one that books travel, submits a form or modifies an account. Adaptiveness and autonomy should be described in the context of the specific deployment rather than assumed from the presence of a model.
Describe the real system
- What objective is the system pursuing?
- Which models and tools are involved?
- Which data and sources can it access?
- What outputs can it create?
- Can it act without human confirmation?
- What can change after deployment?
- Who remains responsible for monitoring and intervention?
Primary material behind this analysis
OECD Explanation of the Updated Definition of an AI System
A practical distinction among AI systems, models, inputs, inference, outputs, autonomy and adaptiveness.
Continue through the discovery stack
Large Language Models Explained
An explanation of LLMs, parameters, context, retrieval, tools, product layers and why fluent language is not proof of current evidence.
AI Agents, Tool Use and Machine-Operable Websites
A guide to agent loops, browser interaction, accessible controls, confirmation, prompt injection, permissions and safe task completion.
The Future of Search
A deep explanation of how search is moving from ranked documents toward interpreted requests, retrieval, generated answers, agents and fewer visible choices.
Search and AI Research Library
A research hub connecting NIST, Stanford HAI, OECD, Google, OpenAI, Perplexity, Microsoft, IETF, W3C, Schema.org and original GEO, RAG and REG sources.