A Monster in the Closet? AI, Machine Learning, and Neural Networks in Seismic Interpretation

May 11, 2018 Scotty Salamoff, Chief Geophysicist at GeoTerra Technologies, LLC

What mental images or real emotions are triggered when you hear the terms "machine learning", "AI", “Neural Network”, or any derivative term in the context of integration into traditional O&G Energy Exploration methods? Some may conjure images of robot Terminator armies crushing humanity, while others are openly averse to the terms because their perception of the impact that this technology will have on job security – are “Machine Learning” and “AI” just fancy buzzwords for “job automation”? 

The answer in the Energy Industry is, of course, NO. Below I've posted a list of the Top 10 reasons we should embrace the integration of Machine Learning and AI with established geological and geophysical interpretation methods - this is a non-inclusive list (of course) but I wanted to keep this article short, so I welcome the reader to contribute feedback or additions! Note: For the purposes of this article, the terms "AI", "Neural Network", and "Machine Learning" are used interchangeably and are all meant to refer to the same thing - the design of an algorithm that can cross-reference a nearly infinite amount of data and train itself ("learn") to pick out patterns in the data with minimal human guidance.

As our understanding and development of these "neural-based" technologies grows, so does our opportunity to improve the efficiency and accuracy of O&G Energy Exploration efforts while reducing overall pre-drill risk through the use of data-driven pattern recognition and (possibly previously unknown) recognition of relationships between seismic attribute data and petrophysical data analogous to an Exploration Area of Interest.

TOP 10 REASONS MACHINE LEARNING AND AI ARE TOOLS, NOT REPLACEMENTS, FOR THE SEISMIC INTERPRETER

1) Computers are powerful, expensive calculators. The software they run allows you to interface with the calculator via a Graphic User Interface (GUI, what you see on the screen) and your mouse and keyboard (how you can "talk" to the calculator). The interpreter (the human with the knowledge) is the one that knows about depositional systems, seismic artifacts, etc. The calculator just knows a whole lot about being a calculator.

2) AI and associated technology weren’t developed overnight. A human being created the code running these products, which is why no two products produce the exact same output from identical input parameters - their metaphorical "brains" are wired differently. (This leads to a deep, dark rabbit hole about how different methods of coding and designing a neural network can introduce interpreter bias into the process, be it intentionally or not. We'll save that for a later article).

3) AI and associated technology are and have always been intended as tools to help a human seismic interpreter make informed decisions backed by real data in a fraction of the time it took 20, 10, or even 5 years ago.

4) Data-backed processes reduce overall risk by increasing confidence in products created through integrated traditional and non-traditional interpretation methods (a built-in QC system, in a way).

5) Unsupervised Neural Networks are unbiased. They do not manipulate data to fit a story, they allow interpreters to view the story the data has to tell. 

6) AI outputs require skilled interpretation by a human being. AI and associated technologies aren’t a “magic button” – whatever the output is still needs to be interpreted by a human to determine what petrophysical or geophysical attributes make up each class or segment of the output. To a computer it’s all 1’s and 0’s, and the "garbage in, garbage out" rule is particularly applicable.

7) AI affords the ability to provide prospectors with information regarding relationships in their data, rather than simply standalone derivatives for them to attempt to manually correlate.

8) Neural networks permit you to increase confidence in a prospect by allowing you to visualize entire petroleum systems - from source to reservoir to seal - in a single image, rather than attempt demonstrate each component of a prospect individually and piece them all together at the end (see image at top left for an example).

9) AI technology saves time. In today’s leaner, quicker industry environment we should make use of all the data we have available to us, in the most efficient way possible.

10) AI gives a voice to seismic data. Amplitude data is easily manipulated and can make a pretty picture, but it obscures a lot of information about the rock physics and distracts from other important information that can literally make or break a prospect. The information is sometimes buried rather deeply, but it is there - it only need be identified and isolated.

Our industry is in the business of finding and producing oil and gas, and our industry never "goes back to the way it was" - once a change happens, it's a permanent change that requires adaptation and receptiveness. These are undeniable truths. Giant 4-way un-faulted multi-fluid phase sandstone megastructure reservoirs with visible flat spots and 45% connected porosity have either already been discovered or lay somewhere no one has looked before, waiting to be discovered. In today’s industry environment that means there is an opportunity now to fundamentally change the way we think about and interpret our subsurface data in the 21st century, and how we can utilize 15 years of technological innovation and development to guarantee the role of the Conventional Explorationist remains relevant and an indispensable asset to E&P Companies moving forward. Such a change in mindset not only benefits the fiscal bottom line of Exploration and Production firms, but is absolutely critical to maintaining the long-term health and viability of the Energy Industry as a whole.

There's no Boogeymen in here.