Woodside's cognitive architecture revealed by IBM

IBM has revealed more detail about the architecture of its Watson deployment at Woodside Energy and the supervised learning required to train it.

The vendor also published a video that shows the extent of Woodside’s specification data and “tribal knowledge” that is being captured and put to use by engineers, including those based on the North Rankin complex platforms off Australia’s north-west coast.

Woodside’s use of Watson cognitive computing technology is designed to make it easier for staff to query 30 years’ of historical internal data.

Though it was to be presented as a “desktop application”, the video – published last night – shows Woodside engineers on its offshore gas platforms are able to poll the system using tablet devices.

IBM said it and Woodside had established a “corpus” that pools “materials” to be ingested by Watson.

“A core group of Woodside’s current and former employees began to test Watson on what it had learned, guiding Watson’s answers and teaching the system to think like one of them,” IBM said in a blog post to Medium.

“Based on a series of questions posed to Watson, the group used their expertise to rate the responses, which were fed back into the system, allowing Watson to learn and become smarter.

“By combining years of data with years of experience, Watson was able to tap into Woodside’s tribal knowledge and discover the best advice of thousands of engineers, as well as learn how to process new information as it was added to the corpus.”

It said “tens of thousands” of Woodside documents had been ingested to date, “each typically over 100 pages in length".

IBM also revealed which of its developer cloud APIs are backing Woodside’s Watson deployment.

The resources firm is using the ‘natural language classifier’ and ‘conversation’ APIs to help the system speak the language of a liquefied natural gas (LNG) engineer.

It also uses a ‘retrieve and rank’ API to “retrieve all relevant information from the corpus, rank them in terms of relevance, and respond with the best matches, as well as related points of inspiration".

All three APIs fall under the language family of “services” offered in the IBM Watson developer cloud. IBM presently has 14 Watson APIs.

Woodside itself continues to provide only scant detail on its Watson project. In its half-yearly report, published late last week, the company said only that it had deployed Watson “into various technical and commercial areas of [its] business".

It also said it was using “advanced analytics” to develop “opportunities in production optimisation through system modelling, predictive maintenance and targeted plant inspections".

However, this is more likely to be occurring through a company-wide analytics drive spearheaded by Accenture rather than through its use of Watson.