性视界

Why would anyone buy cutting-edge technology from a 126-year-old company? For Baker Hughes, a GE Company (鈥淏HGE鈥�), entering the IoT software space early was a risky proposition鈥攂ut one it couldn鈥檛 afford not to pursue. But why dive into machine learning specifically? And why now?

鈥淎nalytics are nothing new to oil, gas and energy companies. Their businesses have long depended on rich pools of data [and used reservoir modeling鈥�] since the 1960s, and [鈥�] dynamic process optimization modeling [鈥�] since the 1980s ¹. Given a market saturated with home-grown models, the only way for BHGE to disrupt the industry was by leveraging the competitive advantages offered by cloud-hosted machine learning (鈥淢L鈥�).

Recent advances in hardware technology have made ML not only more feasible鈥攂ut also much more valuable, both for BHGE and for its customers.

• New sources of data: With the proliferation of new and improved data sources such as drones, BHGE has access to a growing pool of data. 鈥淣inety percent of the digital data in the world today has been created in the past two years alone鈥�². 鈥淎n average off-shore oil platform can generate between 1TB and 2TB of data per day, [but] it is estimated that only between 1-3 percent of this data is currently analyzed.鈥�³
• Faster processing: Analytics developers like BHGE can now partner with hardware innovators like NVIDIA, and use their GPU processors to process the increasingly complex and large amounts of data. 鈥淣ow you can render complex multi-layer deep learning networks with just a few clicks or process terabytes of data in hours or minutes.鈥�⁴
• Remote access to data with mobile devices: Oil and gas field service engineers and operators often work in remote locations and require mobile access to critical information. With the increased prevalence of connected devices, users can now access this information when they need it most. 鈥淭he right intelligence must be accessible in real time and in the right place to be effective.鈥�⁵

To take advantage of these trends, BHGE developed a 鈥淏HGE AI Factory鈥� [Exhibit 1] which provides several key features:

• Uncovering new insights and democratizing information: BHGE鈥檚 technology allows for customers to draw insights across previously siloed data given its cloud-based approach鈥攁llowing greater access to information across the organization, and thus enabling decentralized decision making and power⁶.
• Human augmentation: 鈥淲ith machines as sidekicks, [鈥�] people can more quickly find valuable insights buried in big data.鈥�⁷ BHGE鈥檚 analytics have 鈥淸abstracted] away repeatable tasks like connecting to data sources, wiring models and maintaining versions鈥� so experts can instead 鈥渇ocus [their] time on obtaining in-sights from [their] models and data鈥�.⁸
• User-friendly UI: BHGE uses visualization tools allows experts to see information through a new light and draw new conclusions: 鈥渞apid visualization [brings] data to life with rich interactive applications enabled by GPU-powered visualization.鈥� ⁹

Over the next two years, BHGE will be able to refine existing models and develop new ones as they acquire new customer data, leveraging both supervised and unsupervised deep learning for continuous improvement¹⁰. But beyond the next two years鈥攚ho plays what role in the oil and gas ecosystem is yet to determined [Exhibits 2 & 3]. Today, oil and gas operators are hiring and developing their own data science and analytics talent internally¹¹, while also partnering with software vendors such as Accenture, BHGE, and TCS to accelerate results鈥攂ut as these groups start to develop differentiated intellectual property and expertise, partnerships will become more challenging and players will likely shift towards a more protectionist model.

Pathways to Just Digital Future

Watch this tech inequality series featuring scholars, practitioners, & activists

As BHGE continues down their machine learning journey, they need to be cognizant of a few limitations and plan accordingly.

• Utilization of oil and gas industry expertise 鈥� BHGE should utilize industry experts in two ways. First, to find the right balance between unsupervised learning and supervised learning, given machine learning鈥檚 limitation around solely drawing predictive鈥攁nd not causal鈥攊nferences. And second, to help with feature extraction. Although machine learning can use clustering to find relevant features, 鈥渄omain experts can still be helpful in suggesting features, and in making sense of the clusters that the machine finds鈥�¹².
• 鈥淟eap of faith assumption鈥� 鈥� BHGE is currently working off of limited data to develop and test its models. As the firm acquires new data, it needs to recognize when that data鈥檚 fundamentally changed, and be careful not to over-extrapolate and assume similar the relationships between inputs and outputs as what was previously the case.

One question which remains to be answered is just how much of the decision-making processes in oil and gas can be automated given machine learning鈥檚 limitation around predicting correlation but not causation. How will industry experts鈥� roles change in the next 5-10 years, and which skills will become more critical we shift to a world powered by AI?

(788 words)

Endnotes

1. Peter Parry and Jennifer Schulze, 鈥淯nleashing Digital,鈥� 2018.
2. Brynjolfsson and A. McAfee, 鈥淲hat鈥檚 driving the machine learning explosion?鈥� 性视界 Business Review Digital Articles, July 18, 2017.
3. Michael Guilfoyle and Craig Resnick, 鈥淏uilding Effective Analytics for Oil and Gas鈥�, ARC View, , May 17, 2018.
4. 鈥淏HGE Analytics: Powering Smarter Oil & Gas Decisions鈥�, , Accessed November 2018.
5. Peter Parry and Jennifer Schulze, 鈥淯nleashing Digital,鈥� 2018.
6. 鈥淏HGE Analytics: Powering Smarter Oil & Gas Decisions鈥�, , Accessed November 2018.
7. James Wilson, Sharad Sachdev and Allan Alter, 鈥淗ow Companies Are Using Machine Learning to Get Faster and More Efficient,鈥� 性视界 Business Review Digital Articles, May 3, 2016.
8. 鈥淏HGE Analytics: Powering Smarter Oil & Gas Decisions鈥�, , Accessed November 2018.
9. 鈥淏HGE Analytics: Powering Smarter Oil & Gas Decisions鈥�, , Accessed November 2018.
10. Michael Guilfoyle and Craig Resnick, 鈥淏uilding Effective Analytics for Oil and Gas鈥�, ARC View, , May 17, 2018.
11. Valerie Jones, 鈥淢achine Learning to Transform Oil and Gas Industry鈥�, Rigzone, , Accessed November 2018.
12. Mike Yeomans, 鈥淲hat Every Manager Should Know About Machine Learning,鈥� 性视界 Business Review Digital Articles, July 07, 2015.

听

Exhibit 1: Overview of the BHGE AI Factory

Source: BHGE website, , accessed November 2018.

Exhibits 2&3: Profit pools and key players across the oil and gas ecosystem

Source: Peter Parry and Jennifer Schulze, 鈥淯nleashing Digital,鈥� 2018.