The Oil &Gas industry is always evolving - moving into new geographies, developing and deploying new tools and technologies, and facing new challenges in optimizing exploration and production workflows and project delivery.

One of the more recent technological advances to gain traction in Oil and Gas is the application of Machine Learning processes to solve geoscience problems, such as those related to trap and reservoir identification in salt basins. With ever-increasing data processing capacity and decreasing processing costs, it's only natural for oil and gas companies to adopt increasingly efficient ML applications, pushing them to the limits of what they can do. ML is here to stay from that point of view, and investment in developing and refining ML technology will only grow with time.

Can ML provide a quantum leap in G&G analytical processes? Yes. Is it a complete replacement for human geoscientists? This is a more complicated question. As we will discuss, redundant and repetitive tasks are suited quite perfectly for ML solutions. However, the analytical and abstract part of petroleum geoscience is still well within the realm of human interpreters. We must point here that, just like building human intelligence, artificial intelligence has to learn; and like any learning, it is an iterative process of trial and error.

Quite naturally, this has turned ML into a hot topic. As discussions and exchanges grow, so does the expectation of economic results from ML applications. As a result, managing these expectations and accurately conveying how the technology works have become extremely important. Substantial savings of time and effort should certainly be expected from ML processes in Geoscience, which is pretty much what happened over the last 40 years as processing power, knowledge, data, and information flow all advanced to where they are today. The volume of the information and the quality of the acquired data used in oil and gas projects have advanced dramatically over the last 20 years as well, meaning that data quality control and cost of storage has also become more important.

Along the way, key analytical processes (e.g., well and log stratigraphic analysis, seismic stratigraphy, structural and attribute mapping, petroleum system definition, etc.) were developed, as well as specialized software applications for them. They, in their own right, turned into building blocks of the analytical process. What we call workflows today offer a streamlined and intuitive approach to data loading, interpretation and display carried in robust digital platforms that, because of their functionality, are nowadays at the heart of the Exploratory process. One might forget that the processing power of a typical cell phone today is 100,000 times greater than that of the CPU onboard the spaceship that landed men on the moon. In fact, the computer power and capabilities of today bear little resemblance to that in use just two decades ago.

The creative part of the process - the part that we use to link several geologic and geophysical observations into a coherent reservoir story - is still somewhere else. It still is right what Wallace Pratt said a long while ago, that "Oil is first found in the explorer's mind" (Pratt, 1952). We now call it play-based exploration; in the future, it may go by another name, but the principles will remain the same.

Explorers will probably agree that no two exploratory cases are equal, even within the same basin, and definitely very different from one basin to the next. Producers will probably agree that reservoirs might be quite similar within a field, but there are no two identical fields in the world. And drillers will probably agree that wells within the same field could be designed identical and still have the potential to cause problems for the most varied and unexpected reasons.

Successful petroleum exploration isn't a slave of technology at all; it is based on the experience, creativity, and ideas of the people tasked with putting all the pieces together.

In conclusion:

When talking to students at the start of my Exploration course at Universidad Simon Bolivar, they sometimes are under the impression that exploration is about following pre-designed linear workflows and that all the thinking will soon be carried out by ML.

It is the opposite, in fact:

  • ML will not provide us with the meaning but with deep insights into the data.
  •  It will help with the "heavy lifting" in terms of seismic interpretation and earth modeling. Still, it will not replace the key human roles of prospect identification and fiscal decision-making (drill/drop).
  • ML will become an essential tool to Oil and Gas industry; it will, however, be deployed broadly across E&P teams, unlike specialized applications and interpretation/analysis platforms in use today.
  • It is vital to start working with ML as a tool in the geosciences toolbox. It's nota fit-all cure, as ML applications for geosciences require human guidance in the form of selecting the best and most relative data as input and interpreting the output in a geologic context.
  • ML is growing and maturing rapidly, and the more we use it, the more we add to the collective ML knowledge base. Just like satellites and remote sensing techniques in the 1990s started to produce vast volumes of information for us to analyze, ML will also produce new and novel data about our planet to interpret.
  • Mistakes in interpretation can happen on both sides: computational (ML) and conceptual(geologic interpretation). A productive conversation between the disciplines, when both sides understand the challenge, limitations of the data and technology, and the desired result, requires education across inter-disciplinary barriers. Programmers need to understand expressions of the geologic phenomenon in the data to be analyzed, and geoscientists need to learn about the mathematical principles deployed by ML
  • Exploration and economical field development are based on ideas and engineering solutions to geologic challenges. Experience shows that best results are achieved when an individual initiative is nurtured, and ideas flow freely across the teams, without regard of technical specialties or fear to iterate solutions.

With this in mind, it seems all natural that ML will become essential as a technological resource to further advance both quantity and quality of the created original ideas to cater for increasingly complex and efficiency–hungry Earth Sciences projects.