Artificial Intelligence (AI) has been in the realms of science fiction for many years. In the last ten years it has crept into the mainstream consumer and business world. Insurers are starting to ask what is it and is there a practical use for it in insurance.
The concept of intelligent machines started as far back as 1943, where during World War 2 ideas emerged around combining machines with the field of computing and neuroscience.
The term ‘Artificial Intelligence’ was coined in 1956 for a conference at Dartmouth University, where scientists debated how to tackle AI. One set of academics thought to program a machine with all the rules that govern human behaviour. Whilst others debated the use of neural networks that simulated the process of the human brain and eventually learned new behaviours. Over time the behavioural learning approach won over as the popular approach.
Over the last 30 years machines such as IBM’s Deep Blue (1980) played and won chess against a grand master, environment aware devices such as the Roomba iRobot autonomous vacuum (2002) which maps out and learns its environment and Google’s voice learning app (2008), IBM Watson played the quiz show Jeopardy and won (2011). All of these were small steps in an evolution path that started to push AI it into the main stream
We have seen a real acceleration in the last five years driven by computing power, the vast quantity of data available (due to the internet) and the maturity of the technology algorithms which has kicked off main stream adoption.
General application of AI includes facial recognition in software such as Facebook, where the software can detect a persons face based on pattern matching, voice control such as Siri and Cortana that learns to adapt to voice commands and carry out instructions, and self-driving cars that are aware of their environments and can adapt based on learning the environment and road layouts.
A broad definition for AI is ‘any device that perceives its environment and takes actions that maximize its chance of success at some goal’. The central goals of AI research include reasoning, knowledge, planning, natural language processing (communication), and perception .
Four levels of AI has been defined to describe the levels of cognitive intelligence;
Type I AI: Reactive – The most basic type of AI systems are purely reactive. It does not have the ability to form memories or to use past experiences to inform current decisions. It responds to situations based on a lose set of rules and chooses actions based on pathways acting on the immediate situation.
Type II AI: Limited memory – This class contains machines that can look into the past and use this data to inform current and future decisions. Self-driving cars do some of this already. For example, they observe other cars’ speed and direction. This isn’t done in just that one moment, it requires identifying specific objects and monitoring them over time. These observations are added to its representation of the world.
Type III AI: Theory of mind – This is future world – AI that is aware of the environment and aware that the objects in the world (people, creatures and objects) can have thoughts and emotions that affect their own behaviour. This seeks to understand motives and intentions of other entities in relation to it-self.
Type IV AI: Self-aware – The final step of AI development is to build systems that can form representations about themselves. Ultimately, it will have to not only understand consciousness, but build machines that have it. So if someone is thirsty, it not only understands they need water, but understands that what it feels like to be thirsty and hence the motivation for wanting water.
We are currently some where between Type I and Type II.
There are a number examples where AI is being pushed in personal line insurance and creative platforms are appearing from start-ups; Lapetus is trialling facial recognition analytics to determine life expectancy and health of the insured, Insurer ‘Lemonade’ is using chat-bots to sell insurance and adjudicate simple claims. Tractable is using image recognition to value losses on claims from images.
A number of mainstream insurers are also moving to use AI:
Japan-based insurance company Fukoku Life Mutual, for example, invested in a new AI system based on IBM’s Watson Explorer, which possess “cognitive technology that can think like a person” and calculates pay-outs to policyholders. The technology will be able to read tens of thousands of medical certificates and factor in the length of hospital stays, medical histories and any surgical procedures before calculating pay outs
This has made big news but in reality it has replaced only 30 out of 3000 staff. Apart from headline grabbing, it shows how the some of the commoditised processes can be automated. It anticipates a 30% saving, roughly saving £2m in labour costs per annum.
A London-based start-up company is offering a system that can look at photographs and other information sent in by a claimant to process claims payment. The co-founder said, “We’ve trained AI to look at photos of car damage and make an assessment of the damage, which is a step that is adding a lot of friction today in the claims process,” Again we can see how this could work for some types of claims processing such as property or marine. This solution is already being used by Aegas Insurance.
The use of virtual assistance and chat bots that use machine learning to adapt to different styles of questioning are gaining popularity GEICO Insurance has released a chat bot app that allows customers to enquire about policy coverages, billing information and answer general insurance queries.
Zurich Insurance Group’s chairman Tom de Swaan has said the insurer is deploying artificial intelligence (AI) in deciding personal injury claims after trials cut the processing time from an hour to just seconds.
After the insurer started using machines in March to review paperwork, such as medical reports, de Swaan said: “We recently introduced AI claims handling … and saved 40,000 work hours, while speeding up the claim processing time to five seconds.”
The reality is you can’t automate the entire value chain, but there are parts that can be done in a more efficient way. The speciality insurance market has another hurdle, the varied nature of products and risks has meant AI technology has focused on commoditised insurance products and processes. However, the speciality market has the greatest opportunity, unlike predictive technologies, AI can better handle the variability in risks.
An example is facial recognition, at one time this was done based on a defined database of images and matched based on a limited number of lookups to this image database. With AI technology, the algorithm takes a number of facial reference points and can pick out the person with great accuracy using pattern matching. This analogy is a closer to speciality insurance model, where a book of risks maybe similar but not the same and AI is better at handling these situations.
AI and machine learning is available today in many forms, but they are still in the early stages of use. Vendors such as IBM Watson, iPsoft, Salesforce and mainstream providers such as Google are adding capabilities that allows insurers to utilise this technology.
Business teams will start appreciating that a machine can actually take away repetitive time consuming tasks, so they can focus on the more interesting and complex questions that makes their job more fulfilling. Specialty insurance will not lead the AI race, however, technology exists that can reduce costs and increase service today and insurers would be unwise to ignore these.
Trevor Maynard, head of innovation at Lloyd’s said, “The bespoke risks in society will never go away as there is always innovation in the economy, providing underwriters embrace the new technology they have a very strong future.”
The London Insurance Market 