You are Home   »   News   »   View Article

How would Google run an oil and gas company?

Thursday, August 9, 2018

'How would Google run an oil and gas company' is a question many conference speakers and article writers have tried to answer over the years. We got some hints to the answer from a speech by Greg DeMichillie, director of product management with Google Cloud, talking at the Oslo 'Subsea Valley' conference in March 2018.

We would see a lot more machine learning, or attempts to make it work. Mr DeMichillie believes that the current attention on machine learning is justified, not hype. The maths for machine learning has been around since the 1970s, but has never been possible to put into action until now, because the computing power has not been available. Machine learning is letting us solve many problems for the first time.

However 'Google Oil and Gas' would probably soon realise that oil and gas data is just not good enough, or available enough, to run many machine learning algorithms. 'If you have a corporate culture where data is not shared, that's a cultural barrier to machine learning no software system can solve,' he said. He advised that companies should 'think about your culture of openness'.

Cognite

Mr DeMichillie mentioned Oslo oil and gas technology company Cognite in his talk, telling the audience how Cognite uses Google's 'TensorFlow' machine learning framework to build tools for predictive maintenance and for predicting corrosion for oil and gas operator AkerBP.

Cognite is building an integrated data platform for Aker BP's data. The CEO of AkerBP, Karl Johnny Hersvik, spoke at a conference a few weeks previously, the 'Moment - the Future of Work' oil and gas conference in Oslo on Mar 15 2018, wearing a T-shirt with 'Data liberation front' written on it. The Google Data Liberation Front is an engineering team at Google whose "goal is to make it easier for users to move their data in and out of Google products.

Google has a link to Cognite in that Geir Engdahl, Cognite's CTO, is a former senior software engineer with Google in Canada.

Cognite is owned 72 per cent by engineering and construction company Aker ASA (NOK 44m / USD 5.7m) and 10 per cent by oil and gas operator AkerBP.

Machine learning

Machine learning is getting better and better at working with small amounts of data, he said. 'I used to say you need big data [for machine learning], that's less and less true,' Mr DeMichillie said.

For machine learning to work, the training data needs to reflect what happens in the real world. For example, if the data is based on a sample of people which is 80 per cent male, then 'it won't reflect the world at large,' he said.

The data would also need to include all the things which happen in the real world. A computer cannot use machine learning to predict something it has never seen, in the same way that a human could not predict what a dog looks like without having seen one before.

Machine learning in Google

Google uses machine learning to suggests what replies people should send to Gmail messages, and today 'smart reply' is used for 1 in 8 replies sent via smart device, he said. The company has used machine learning to make big improvements to Google Translate, which can be proven by translating a complex literary phrase from English to Japanese and back. Google has used machine learning to control the cooling systems in its data centres, and achieved an astonishing 40 per cent reduction in electricity required.

Google developed an open source machine learning framework called Tensorflow, which any researcher can use, and it developed custom microchips to run on it, which can run AI algorithms 15 to 30 times faster than conventional chips, with 30 to 80 times more compute operations per watt. Google also provides tools for building and training models.

Tensorflow is used by Airbus to analyse satellite imagery, in particular identifying whether white clouds in a photo are clouds or snow.

Another application was a company which used machine learning to try to identify ships which were illegally fishing, by understanding tracking patterns in their satellite positioning data, which all ships must transmit. As a result, one small island was able to levy a fine on a shipping company which amounted to a sizeable proportion of its annual state income.



Associated Companies
» Digital Energy Journal
comments powered by Disqus

CREATE A FREE MEMBERSHIP

To attend our free events, receive our newsletter, and receive the free colour Digital Energy Journal.

FEATURED VIDEO

The future of subsurface data management? Building a data science lab data lake
Jane McConnell
from Teradata

DIGITAL ENERGY JOURNAL

Latest Edition April 2018
Apr 2018

Download latest and back issues

COMPANIES SUPPORTING ONE OR MORE DIGITAL ENERGY JOURNAL EVENTS INCLUDE

Learn more about supporting Digital Energy Journal