SAS announced a $1 Billion investment in AI a few months ago. We sat down with Iain Brown, Head of Data Science at SAS to discuss the challenges and opportunities of AI in enterprises.
He also discusses how ethics and good AI is ‘tricky’ and requires a carefully thought out framework in order to succeed.
Algorithm-X: Can you tell us about your background and your role?
Iain: Certainly, I have always had a fascination in the practical applications of mathematics, statistics as well as a love for computer science. So the rise of Data Science as a professional vocation was the perfect fit.
For the last decade I have held a number of roles focused on understanding and solving large scale data science problems and now head up the data science practice for the market leaders in AI, SAS.
Algorithm-X: I understand SAS investing a lot in AI. What major initiatives are you working on?
Iain: That is correct only a few months ago we announced a $1 Billion investment in AI which focuses on three main areas; Research and Development (R&D) innovation where we will continue to build on the success of our global AI efforts; education initiatives addressing customer needs to better understand and benefit from AI; and expert services to optimise customer return on AI projects.
From a UK perspective, this has given us the freedom to invest in some of the biggest problems facing our customers.
This includes but not limited to deploying NLP capabilities to significantly improve customer experience by resolving customer issues faster for a financial services provider and deploying image processing functionality to identify defects with 99% accuracy for pharma product manufacturing.
Algorithm-X: Is AI right for all enterprises?
Iain: If you ask enterprises how they define AI you will likely receive as many definitions as enterprises asked. AI is many things to many people but simply put it is the science of training systems to emulate human tasks through learning and automation.
The applications of this are varied and numerous and depending on an enterprises’ maturity there will be different use cases that will be right from them.
From simple robotic process automation (RPA) to the application of deep learning for computer vision all enterprises can benefit from AI.
Algorithm-X: Where should you start with AI if an enterprise wants to adopt it?
Iain: I always say start with the problem identification, and determine what it is that needs to be solved. Once this has been clearly defined making sure quality data is readily available is typically the next hurdle.
The adage of 80% of a data scientist’s time is spent on engineering the data is still as true today as it was a decade ago. Only once the previous stages have been undertaken can the fun begin.
Algorithm-X: What data should you use?
Iain: Good quality data! Whether the data is structured, semi-structured or unstructured key factors such as integrity, consistency, completeness, and timeliness should be ascertained before use.
Without the correct data governance and controls around lineage in place, accurate and effective AI models will be impossible to deliver. You wouldn’t build a house on sand, why would you build an AI on unstable data.
Algorithm-X: What’s are the great examples of enterprise embracing AI?
Iain: There are many great examples of enterprises embracing AI, to highlight a few; reducing the average settlement time from 28 days to 12 days for a large UK Insurer through the use of machine learning to better estimate the cost of repair.
Increasing the defect detection accuracy in a manufacturing process to 99% through the use of computer vision.
Using sophisticated natural language processing capabilities to help a financial services organization better understand and process customer web chat enquiries more efficiently
Algorithm-X: How do you build safe and ethical AI?
Iain: Ethics is a tricky subject area, as with like AI interpretations and views differ. Fundamentally I am a strong believer that organisations need to have a framework when considering how to build and deployment ethically sound AI.
There are variations of this but the one I am a proponent of is FATE, which is an acronym for Fairness, Accountability, Transparency, and Explainability.
Ensuring organisations consider this or similar frameworks is key to ensuring the following questions can be answered:
1) Do you know what your AI is doing?
2) Can you explain it to your customers?
3) Would they respond happily once you told them?
If organisations can successfully answer these questions they are in good shape for developing ethical AI.
Algorithm-X: How to get the company and the board to buy into AI?
Iain: In a lot of cases we are actually seeing the use of AI being pushed down from the board level. However to ensure organizational buy-in typically this will come from more tactically deployed AI use cases.
Don’t be afraid to start small in terms of where AI is applied to initially prove out the value of what it can achieve.
We have seen this work particularly well in the banking sector where AI is being used to enhance customer experience through more accurate decisions in the pre-screening of credit applications.