twitter linkedinfacebookacp contact us

Unlocking the potential of generative AI in energy: interview with Amazon Web Services

Technology

Oil Review Middle East (ORME) speaks with Hussein Shel, enterprise technologist – Energy & Utilities, AWS, on the leveraging of generative AI in the energy and oil & gas industries 

ORME: Could you explain how generative AI, specifically large language models and multimodal models, has the potential to transform the operational landscape of the energy industry, including the oil and gas sectors?

Shel: We believe that generative AI will have a profound impact across all industries; energy is just one of them. Amazon has invested heavily in the development and deployment of AI and machine learning services and models for more than two decades, both for customer facing services and our internal operations. 

For the energy industry, we truly see generative AI playing a pivotal role in increasing operational efficiency, reducing health and safety exposure, enhancing the customer experience, minimising emissions associated with production assets, and of course, accelerating the energy transition.

A lot of the work that we're doing today with our customers is because of the recent buzz around generative AI and some of the chat-based or natural language-based interfaces that complement existing ML and AI models. So that is where we are initially seeing some of the interest from our customers.

AI encompassing those large language models and multimodal models has transformative potential across the sector. It will help enable the analysis of extensive textual and visual data for improved decision making and offering insights. For example, processing well logs and looking at interpreting geological surveys and images to aid an existing optimisation of exploration and production decisions and strategies. 

ORME: How do you envision generative AI addressing the issue of engineers spending a significant amount of time searching for data?

Shel: We see generative AI holding the potential to help alleviate that challenge for the industry—speaking specifically about data discovery, knowledge management, and knowledge exchange and sharing. This could be done through automated data retrieval, contextual understanding of the data sets, and personalised recommendations through natural language interfaces. 

For example, generative AI could enhance engineers’ decision-making ability by summarising documents that are available through a corpus of data sets that they have from PDFs, Excel files, images, et cetera.

It could help by suggesting relevant resources, making the connections in the network between those resources, and suggesting any relevant information for that engineer aiding in report generation. 

It could also help facilitate knowledge transfer between experienced individuals and newer team members, which is a pretty significant challenge today for the industry. Additionally, generative AI could also provide predictive insights based on historical data and real time data. 

ORME: Could you walk us through a scenario where gen AI applications help analyse images and videos from field locations and subsequently provide alerts recommendations and insights to optimise field operations and safety measures?

Shel: An AI model could detect anomalies to identify equipment malfunctions or leaks. When these anomalies are detected, real time alerts could be triggered through generative AI applications, for example, that could autonomously create workflows and actions that complement existing workflows and provide immediate awareness to operation centres and safety officers. 

AI could go beyond detection, offering contextual insights and recommendations for timely actions. Generative AI could be used for generating synthetic data needed for the training of traditional models and help to provide a better experience around safety and maintenance scenarios.

Predictive insights really have the potential to help prevent equipment failures and drive proactive maintenance, which leads to optimised production. Also, using AI-driven and, machine learning-driven applications to assist with decision making could not only improve operational efficiency by reducing downtime and equipment performance, but could also minimise health and safety risks through rapid intervention and compliance enforcement. 

AI and most of these models continuously learn from new data, such as from new video streams from new images, whether real or synthetic.

It could become increasingly adept at anomaly detection and refining recommendations. Its transformative potential lies in its ability to continuously analyse images and videos from remote sites. 

ORME: Could you delve deeper into how general adversarial networks (GANs) work in this context and explain how synthetic reservoir models generated by GANs could impact critical areas such as hydrocarbon production forecasting, geothermal energy, and carbon capture and storage?

Shel: Using GANs, or generative adversarial networks, to improve and augment existing subsurface reservoir models by creating synthetic data is definitely an intriguing and exciting application. There's not a lot of research on this topic yet, but we see the potential of it and we're starting some conversations with customers to explore the opportunities. 

But, let's take a step back a little bit and just define what GANS are. GANS are a type of machine learning model and are also a subset of the generative AI discipline. They consist of two components: a generator and a discriminator. The generator creates synthetic data while the discriminator tries to differentiate between what's real data and what's synthetic data.

Through a competitive process, the generator improves over time, producing increasingly realistic synthetic data. In the context of subsurface reservoirs, GANS can be trained on existing reservoir data such as well logs, very granular high-resolution data sets from well logs to seismic data, and production history. The generator learns the underlying patterns and distributions of the data, enabling it to create new reservoir models. 

The discriminator evaluates the realism of the generated models. Through that iterative training, the generator becomes more proficient at creating synthetic reservoir models that closely resembles real ones.

To summarise, GANS ability to generate synthetic reservoir models opens up a new avenue for improving hydrocarbon production forecasting, optimising geothermal energy extraction, and assessing carbon capture and storage projects.

ORME: Considering the unique energy landscape of the Middle East, characterised by significant oil and gas operations, how receptive has the Middle East been to the concept of generative AI?

Shel: I think it's a bit early to highlight any particular use cases that are specific to the Middle East region. But, what I can say is that many of the challenges that we're working to address in the region are largely applicable to other regions as well as other customers’ challenges that we're working on.

With that said, we're excited about ADIPEC 2023 and our workshop specifically on generative AI use cases that will include presenters from AWS, ADNOC and AIQ.