In the last two years, there have been exponential changes in artificial intelligence (AI) and machine learning (ML). At 7 Puentes, we have always been at the forefront of these changes, helping our clients adopt these new technologies. In this wave of generative AI, we think it is a very opportune moment to share with you the valuable opportunities that we are observing in our consulting, and how they are applied in the different projects with our clients, and also internally in our business solutions. In this blog post, we will specifically delve into the concept of generative AI in oil and gas industry – exploring its benefits, use cases, implementation strategies – while highlighting how it can transform businesses in this dynamic sector.
Very soon, our executives will be attending the AI in Energy 2024 edition in Texas, United States, where we hope to pick up good ideas and share our experiences. In this context, it seems very appropriate to explore the impact of generative AI in the oil and gas industry.
We will explore the topic of Generative AI in this industry through key questions and answers.
What is the difference between Generative AI and traditional AI and ML?
Most of the ML use cases we implement at 7Puentes require a set of labeled data as a basic input. That is, examples so that the algorithm or model adapts to the patterns present in the data. Often, obtaining a good data set is a laborious task – both for organizations and for 7Puentes – since it requires discovering features or variables that are relevant for the prediction, cleaning the data, and sometimes even human effort to label them correctly.
In the cases of GenAI, this problem is significantly reduced because the LLMs have been pre-trained with monstrous public datasets that capture these patterns, even if the data is not from the client’s domain in question (in some cases). What remains to be done to take advantage of this? Just fine-tuning and working with the prompts (providing examples) to adapt them to the use case and also to the client’s domain. There is even a modern architecture pattern called RAG that addresses this specific issue.
Another important difference is that the models typically used for text generation do not work for many of the typical industry problems associated with time series or geostatistical information typical of the oil and gas industry. It is not that they cannot be tried, but they simply will not replace what already exists.
What are the opportunities in the oil and gas industry?
In the oil and gas sector, various industrial roles coexist with heavy machinery operators, task supervisors, organizers and planners. It is at these interfaces between computer systems that humans operate where there is the greatest opportunity. In the last year, we have implemented very interesting use cases that any company in the industry can apply:
- Text classification and decision-making: Operators in the field often record observations and notes about maintenance, safety, hygiene, quality, and other aspects of processes. These reports are typically collected and stored in the plant’s management systems, but need to be read by other humans for subsequent strategic decision-making. Today, Generative AI technologies make it possible to automatically read these observations and categorize or extract new knowledge from large volumes of them. In this case, a human would need days to read many texts but an AI can do it in a few minutes.
- Interpreting graphs and reports: The latest models are multimodal and tend to be quite good at interpreting graphs and diagrams in a basic way. This can contribute a lot to reading Earth Since visualizations like seismic data, well logs and cuttings. The benefit is clear. It can save significant time when an analyst has to read 500 pages of technical reports. Undoubtedly, generative AI will speed up these tasks and simplify processes, freeing up human staff to focus on other strategic activities of the business.
- Fast queries and Text2SQL: Oil & Gas organizations have many systems, and in many cases, information is very fragmented between the systems themselves. This is usually a natural consequence of the organization of the industry and the vendors of the specific software. Developing an integrated text or even audio interface that is convenient for management levels to resolve quick queries is a very interesting use case.
An aspect to consider: OpenAI vs. OpenSource
This is a very important aspect to consider because there are two barriers to using OpenAI models, assuming they are the best for the tasks we want to perform.
On the one hand, there is the price, since these models are not free and charge by tokens -similar to charging by words- which can increase the cost significantly.
On the other hand, intellectual property and the sensitivity of the information handled in this sector usually prevent this type of service from being used. Today, you cannot have 100% control over OpenAI software in an on-premises installation or in a private and highly secure cloud. This leads many to try to achieve the same results with open source.
This requires staying up to date with the latest models, as new versions of models that surpass them appear every week. We suggest consulting the Hugging Face leaderboard to gauge how fast the industry is moving.
Final Notes
For many of these topics on AI and ML applied to Oil & Gas, there will likely be new blog notes to delve a little deeper into each one. And of course, we will keep you updated on all the news regarding our participation in the AI in Energy summit in Texas, USA.
