Artificial Intelligence: everything you need to know but are too afraid to ask
August 26, 2016 Greg Brown
AI, Artificial Intelligence, Machine Learning, all terms on the rise. But what do they mean? Why should we care? Here’s everything you need to know to bluff your way through your next meeting – and there’s a handy one-page cheat sheet at the end to print off and keep.
What is AI?
Artificial Intelligence is the use of computers and robots to replicate human intelligence. The process of ‘Artificial Intelligence’ can be summarised in three simple steps:
- Perceive: gather data and information from the real world e.g. sensors, IOT, photos
- Decide: process the information gathered and make the best possible decision given the inputs
- Act: take action in the real world in line with the decision e.g. robotics, robo-advisors
AI is different from traditional process automation in that it has understanding to enable it to make its own, new, decisions, as opposed to just following a set of predefined rules (e.g. if claims are less than £100 then pay automatically).
Why should I care?
There’s a lot of hype around how AI will ‘disrupt’ the insurance space. AI itself is just a means to an end, much like any automation technology. In broad terms AI can help in three areas:
- Better targeted distribution: use data to better understand customer needs and target products based on them
- More accurate underwriting: identify and understand risks at point of sale better and predict future losses more accurately
- Increased efficiency of operations: reduce the need for human intervention through automation of complex processes
What makes Artificial Intelligence different from normal IT automation?
AI is different from traditional IT automation in three ways:
- Learning: AI-based systems learn from a stack of training data at the beginning and from real data as it subsequently comes in. For example, if you suddenly get a spike in fraud from a new source, not only will an AI-based system spot it, but the system will work out what’s causing it. IT automation, as mentioned above, is does not learn.
- Future-looking: AI makes decisions based on what it believes to be the best course of action for the future. So, for example, it might decide how best to treat the fraudulent claims in the future.
- Decision making: AI-based systems are able to make decisions based on context and information.
Why is Artificial Intelligence all so complicated with so many different algorithms and approaches?
There’s a trite answer to that question which is: humans are complex so replicating them is complex. There are more than 100 trillion connections in the human brain which makes for an incredibly advanced ‘computer’. Replicating this in computer form is no mean feat.
Another way to answer this question is ‘different tools for different jobs’. Bayesian learning, neural networks and hidden Markov models all have their place in the AI ‘toolkit’. Much like the rise of user friendly computing that doesn’t require technical knowledge to operate over time (e.g. the iPad), AI will become significantly more accessible to the lay person.
Why is it all happening now?
First, Artificial Intelligence is nothing new. The idea of artificially replicating human intelligence has been around for decades and the practical application started in the 1950s with Alan Turing’s paper on Computing Machinery and Intelligence. The term itself was coined in 1956. There are three reasons why the technology has seen a boost recently:
- Sufficient computing power to process the large amounts of data required to drive it. The cost of computing has decreased dramatically over the last 20 years.
- There has been a huge advances in the development of algorithms in the last 20 years. This is partly driven by availability of computer processing power on which to test them.
- AI systems finally have sufficient data to train their algorithm. This has been dramatically helped by the rise of the internet, helping create exabytes (translation: “an awful lot”) of data each day, and dramatically reduced costs of data storage. This means data is now available in large enough volumes to make AI viable.
If you are interested in understanding how AI might help you improve your business then please do get in touch (firstname.lastname@example.org). At Oxbow Partners we work with a network of domain experts, such as for AI, who have extensive and practical experience in their chosen field.