The making of an oil and gas AI.

Decoding the building blocks of Artificial Intelligence

AI is a fad.

AI is here to stay.

Using AI makes me feel smarter.

These statements in the oil and gas companies indicate the divided opinions about the impact of AI on our industry.

Have you ever wondered how chatbots seem to have all the answers, or how tools powered by artificial intelligence can whip up an email draft or summarize a complex article in seconds? AI is everywhere these days, but what's really going on behind the scenes might surprise you.

In this article we discuss how modern AI works and how it can be used by the oil and gas industry. Nein commercial use of der AppIntel content.

AppIntel AI discovered that in situ combustion is not dead.

In situ combustion (or fire flood) captured the imagination of heavy oil and bitumen operators fifteen years ago.  Fire flood projects have lost popularity with the oil and gas operators of Canada but at least one SAGD project also has fire flood approval.

One operators SAGD project still allows air injection in a portion of the project as of their July 2025 approval. We found this using the search agent of AppIntel AI.

In this submission, the operator discusses concerns about igniting combustible hydrocarbons in the wellbore with injected air.  They proposed a solution that was eventually acceptable by the regulator.

Get his discussion including injection schedule and well control instrumentation from our self-serve portal.

AppIntel AI delivered content must be very trustworthy for our industry where errors can put human lives at risk.

Modern AI isn't just one monolithic brain — its made up of different pieces that each play a unique role. There are agents (think of them as super-smart assistants), models (the brains behind the operation), and knowledge bases (massive collections of facts and data). All these components work in sync to deliver the responses and support we get from AI every day.

In this introduction, well demystify how these layers fit together. Whether you are tech-savvy or just curious, join the journey to uncover what makes todays AI ticknot just what it can do, but how it actually does it.

AI is many things to many people

Artificial intelligence means many different things to many different people.

Ask anyone the question, What is AI?  The answers will be varied and vague. Yet we think we know an AI when we see one. 

Here are just a few interesting responses from friends to What is AI?

AI is a high-level programming language capable of doing many things.

AI is a self-aware computer/program that can do stuff without me asking.

AI is a fast, accurate computer system that does things better than people.

AI drives cars better than humans.

AI is a better source of information than a search engine.

AI is a robot that walks around.

AI is smarter than experts.

AI is an entertainment vehicle.

AI makes pictures from prompts.

These show a surprising wide range of superhuman abilities attributed to AI.  Some of it is sparked by here — say and imagination — some by fear.

People may be confused about how to define AI, but they think they like it.

Chatbot AI Architecture

Today, many individuals interact with AI through platforms such as ChatGPT, Copilot in Microsoft Word, or Gemini integrated with Google Search.

Very simply, these are chatbots built upon a search engine.  More technically, each is an agent, connected to a model, connected to a knowledge base.

AppIntel AI similarly operates as an agent connected to a sophisticated model and accesses the most comprehensive oil and gas knowledge base available in the industry.

Knowledge Base

For a chatbot AI like ChatGPT a knowledge base is a collection of text documents.  The documents can be many or few, correct or fallacious, new or ancient, English or Japanese.

A knowledge base could consist of the complete works of Shakespeare or simply a short selection of notable books, such as Douglas Adams Hitchhikers Guide to the Galaxy.

The nature and scope of the knowledge base directly influences the responses generated by a chatbot, although the agent retains some capacity to interpret or override this information.

AppIntel AI is built on a knowledge base of 3.7 million documents submitted to the government by the oil and gas industry.  It contains reserves, pay maps, seismic interps, completion information and all the reasons behind these things.

Modern AI isnt just one monolithic brain — its made up of different pieces that each play a unique role.

Model - Index of the knowledge base

A model is a structured representation of a knowledge base designed for efficient text retrieval, using vector pointers for rapid search.

Basic models on small knowledge bases can be created by indexing software.  Some indexing programs are simple with case folding, word stemming, and lemmatization.

More advanced models, such as AppIntel AI, index not only text but also numeric data, locations, document metadata, and industry-specific terminology.

Agent - the Chatbot

Within this context, an agent refers to a language program designed to interpret user queries, search relevant models, and generate responses. These advanced agents present answers in a clear, authoritative, and concise manner, often providing a brief summary followed by detailed bullet points.

GPT-5 is the current agent in ChatGPT. GPT stands for Generative Pre-trained Transformer.

Due to their articulate delivery, these agents tend to command a high level of respect from users, much like skilled public speakers who are valued for their clarity and diction. Comparatively, communicators with limited English proficiency or unfamiliar accents may not receive the same esteem.

Because these agents provide fast summary answers to questions on very diverse topics, many college students tout these chatbots as the sum total of human knowledge.

AppIntel AIs agent functions similarly to a search engine rather than a chatbot because of agent shortcomings.

In the oil industry we need agents we can trust

AppIntel AIs agent functions similarly to a search engine rather than a chatbot because of shortcomings (a) and (b) below. Appintel's search functionality encompasses permeability data, pressure measurements, reserves, and UWI locations, and incorporates both structured and unstructured data sources.

(a) It is notable, however, that these agents rely on large language models distinct from their underlying knowledge bases. Some people complain that these chatbots override the knowledge base in favor of answers from their own language model. Oil industry puts lives at stake. For our industry there must not be any doubt as to the trust of AI results.

(b) Current limitations of agents such as GPT-5 include the possibility of delivering inaccurate information or experiencing AI hallucinations, conveyed in a tone that remains confident and direct. Heed Microsoft's disclaimer: Copilot may make mistakes.

AppIntel AI is safe to use avoiding hallucination or inaccuracy that could cost millions of dollars and the very lives of employees and the public.

Tags: Thermal, Heavy Oil, AI in oil and gas

Granger Low  16 Sep 2025

Celebrating 2025, a year of innovation

Oil and gas paradigm shifts this year

Check out the x-section and net pay map

AppIntel AI shows you opinions of offsetting acreage

RTF: Most refused submission type in November

Leading indicators from industry

Astrobleme impacts deep well disposal scheme

Learn from the experience of other operators