Our offices will be closed from 25 Dec – 2 Jan,for urgent enquiries please email info@whinneyinsurance.co.uk

Transforming our lives: Exploring the power of AI with geophysicist Tom Merrifield

AI is set to make our lives more efficient, and will permeate almost all aspects of our lives in the future. From how we travel, shop and engage with the world. We caught up with Tom Merrifield, a geophysicist in Shell’s research team using AI in its seismic interpretation technology. We caught up with Tom to discuss this transformational technology and the opportunities, limitations and risks of AI.
Whinney Insurance Tom Merrifield

Can you tell us a little about your background and previous work?

I’m a geophysicist who started my Shell career in seismic imaging, which uses reflected energy from waves sent into the earth to construct an image of the subsurface in three dimensions, much like a medical CT scan but on a larger scale. Following this, I worked as a geophysical workflow consultant providing technical support to exploration teams across Europe, and developing, deploying and promoting new workflows and differentiating technologies with the exploration and geophysics businesses. Two years ago, I joined a research team where we plan, develop and implement AI technology to predict geologic features and geophysical properties from two- and three-dimensional seismic data.

Over the last few years, Shell has been heavily investing in research and development of artificial intelligence (AI), looking at every step in the process from drilling to consumer behaviour. Can you tell us about your work in AI for Shell?

Shell has been heavily investing in research and development of AI, and has set the ambitious goal of drilling the first AI-derived prospect by 2024. A key step in the identification of hydrocarbons in the subsurface is the interpretation of seismic data, which is the science and art of inferring geology from the 3D seismic image. It requires an interpreter to draw on their geological understanding of the area to pick the most likely interpretation from the multiple valid interpretations that the data allows. Typically this is a multi-month process involving experts from various subsurface domains, with large uncertainties. I currently work in a research team focussing on seismic interpretation technology, and we believe that AI has the potential to revolutionise seismic interpretation. The team is working on a deep learning approach to predict the location of hydrocarbons in the subsurface from seismic data using large, deep neural networks.

How do you think AI will change our everyday lives over the coming years?

AI is a tool that will increase the productivity of a task, and I think AI will generally make our lives more efficient, and will permeate almost all aspects of our lives in the future. More and more tasks which have been too complex, too time consuming, or even just too boring will be performed by AI in the future. One of the bigger changes that may be coming soon is self-driving cars, which will bring a huge shift in the way we see transportation.

Shell is also investing in electric vehicles and charging points, using AI to monitor and predict demand, through RechargePlus which is currently being trialled in California. Can you tell us a little bit about this programme?

Encouraging motorists to switch to electric vehicles is a key step in reducing C02 emissions, but motorists are put-off by the lack of charging terminals, and businesses are put-off by the lack of demand. Shell’s answer to this is to deploy AI to monitor and predict the demand for terminals throughout the day, enabling power to be supplied more efficiently. Its intelligent app helps to minimise costs, by maximising the amount of charging that occurs during the lowest demand periods of the day, and provides flexibility to drivers by allowing them to select the settings of their charging stations based on their schedule. As a grid operator, operating many charging terminals, if many cars plug in at the same time and automatically start charging, a big load is created on the grid. This may also be happening during peak rush hour times, which cannot be filled by solar because it’s 7 or 8 am. By understanding people’s charge profiles, the load can be spread during the day, which both saves the consumer money, but also allows more renewables to be used – if you charge more cars at lunchtime there’s going to be more solar on the grid at that time. It’s a good example of where AI plays a key part, thinking about how we can make things more efficient but also how to change energy consumption patterns to take advantage of renewable sources, with the added benefit of minimising costs to the consumer.

What industries and jobs do you think will be changed by AI?

I think eventually, all industries will be changed by AI in some way. The power of AI is such that it is limiting to look at a certain technology or application, and I believe it should be seen as a transformational technology such as electricity or the internet.

What skills do you think cannot be replicated by AI?

There are many skills that AI will not be able to replicate. Anything involving empathy, communication, creativity, or planning will be very challenging for AI. (CANVA) A few years ago, there was a fear that AI would result in robots taking away people’s jobs, and while some jobs will be completely automated away, people are becoming used to the idea that they will be needed to work with an AI system to perform a job task more effectively.

What are the challenges you face in developing AI?

One of the challenges we face is that we cannot demonstrate clearly why our deep learning model makes the predictions that it does, it is a “black box” to humans. People are understandably sceptical about it, as it is not simple to understand how it makes decisions. The level of mathematical uncertainty behind our AI application remains a grey area, and there is a definite trend in developing explainable AI, or “XAI”. People with deep expertise have built up a level of trust, with proven results, and even when they may make an intuitive leap they will still need to go back and look for evidence to help explain their prediction. Another issue we face is data scarcity. Training deep learning neural networks involves massive amounts of labelled data. AI algorithms are only as good as the data they are trained on, and bad data can be laced with biases. There is also a shortage of people with the required skills. For our application, domain expertise, geological and geophysical knowledge and experience, is as important as the data science knowledge and skills required to develop and operate machines which think and learn by themselves.

A challenge facing the wider development of AI is the bias in training data, especially when the trained AI is used to make critical decisions, such as who is called for a job interview or whose loan is approved. Bad data can be full of racial, gender, ethnic and other biases, and there have been examples of this highlighted in the media recently. If the bias embedded in the algorithms that are used to make vital decisions is unrecognised, it could lead to unethical and unfair consequences. It is critical to both widen the demographics of people developing AI and to devote energy to identifying potential biases in training data and the resulting AI.