When the IBM Watson computer won a $1 million prize on the quiz show Jeopardy! by beating two grand champions, it was far more than a novelty TV moment.

It marked the beginning of our understanding and use of artificial intelligence. Where that has led, and what is still to come, were discussed by Rob High, chief technology officer at IBM Watson, during a visit to Bermuda.

To win Jeopardy! in 2011, Watson induced its answers from 200 million pages of digitised literature stored in its memory. It took the computer three seconds to consult that wealth of material and come up with its responses.

But as anyone who is familiar with the quiz show knows, the questions are not straightforward.

“They are written in a style that is intended to be somewhat misleading, which could be by creating misdirection or using innuendo, or other linguistic devices that make it difficult to understand what the question really means,” Mr High told delegates at the GR Innovation & Insurtech Conference.

“It was an opportunity to challenge the philosophy of AI to prove it understood human language well enough to be able to recognise the nature of the questions being asked and find answers that were relevant to he game.”

That entailed being able to understand the linguistics of language and how humans express themselves, and to be able to get a deep enough understanding of the questions to find the answers.

Watson was groundbreaking, but it was not alone in the wider, related sphere of deep learning and machine learning, such as used by Google for page ranking systems, and Apple’s virtual assistant Siri.

Artificial intelligence has now been introduced in healthcare, financial services, retail and manufacturing.

“This is only the beginning of what we are going to see going forward,” said Mr High.

He explained that a problem being faced in today’s world is the ever increasing volume of data — from written reports, books and papers, to audio and visual data and even the 200 billion tweets posted annually. In addition, the Internet of Things means many more devices, including refrigerators, washing machines and cars now produce their own streams of data.

“We are going to see the growth in data hitting 4,300 per cent a year in the next couple of years. We can’t keep up with it,” said Mr High.

However, AI can take that data and make sense of it so that it can be beneficially used. AI’s biggest impact will be in amplify the ability of human intelligence and human cognition.

Mr High explained: “Humans are very good at adapting. What we are not very good at is doing things at enormous scale.

“Imagine what it would be like to read 200 million pages of literature in three seconds? Even if we could, chances are we would not be able to assimilate a lot of it, or be able to recall it at the moment you need it.

“If we can take AI and augment human cognition, think of the power that brings and our ability to reason about problems that otherwise we couldn’t.”

The last frontier is for AI’s to interact with humans in a way that is natural and compelling — bringing ideas to the table, injecting itself into the conversation at that right moment, Mr High said.

Describing current uses of AI, he gave an example of an insurance company that uses it to quickly recognise the nature of damage to vehicles from photographs, while another company uses AI-powered visual recognition to assess what kind of corrosion and rust is occurring on its equipment and then to prioritise repairs.

When it comes to AI transforming financial services, Mr High noted two types of systems that are used — conversational and discovery.

So-called “chatbots” that tend to conduct single-term interactions, such as Siri and Amazon’s Alexa, fall into the conversational category. But they can only scratch the surface when it comes to the nature of a problem. Most times there is something deeper behind the initial question, Mr High explained.

“A conversational agent ought to be able to get to that problem, beyond the first question. The question that we really care about does not come until later.

“We have a lot of things to develop, but these are the goals in call centres, between clients and agents, and for professional help desks.”

He gave an example of Autodesk, a company in the US that prior to having a conversational agent took on average a 1½ days to solve a problem a client had with software.

“With a [AI] conversational agent the average call resolution was reduced to 5½ minutes,” he said.

“The system could quickly address the questions, most of which are routine, and what was left over were the more interesting problems that the call centre agents could then go and address.”

While in those instances a real person steps in to take over the more complicated questions, one of the challenges for the development of AI is to see how much further into complicated queries it can delve and return with a successful solution.

“How complex can a question be that AI can understand to be able to answer and resolve?” Mr High asked.

“That’s where discovery comes in, which is closer to what we used in Jeopardy! We had examples of complex questions that the system had to understand and go and find answers to.”

On the question of how the insurance and reinsurance industry can deal with the hurdles of not being able to access the data it has and turn it into a format that is useful in an AI situation, Mr High said it was not as a big an issue as it appears.

“Begin with data that you already have in a useable form, which may be insurance reports, articles, outside literature that has already been curated for commercial purposes.”

He said other data, like that supplied by clients which is not fully curated, can go through a process of refinement.

Mr High said Watson was not a monolith that collects all the information in the world, but instead can be applied in smaller settings.

“Common with most AI systems, you can begin small and build from there. It leads to the notion of a virtual cycle,” he said.

“You have data, you apply AI technology to gain access to that and make that information available to people who use it. Then gather from those people the information they produce as they use the information you provided for them, and feed that back into the system.

“You now have a system that is building and feeding itself. You have a system where data is being refined in the process of applying technology and incorporating humans in that loop.

“It is another example of how AI amplifies human cognition. It is about AI and humans together, and that is where we are going to see the greatest power.”

The GR Innovation & Insurtech event was held at Rosewood Bermuda. This is the second consecutive year that UK-based Global Reinsurance has staged the event on the island. It was sponsored by Hamilton Insurance Group, AdvantageGo, Conyers Dill & Pearman, and supported by the BDA.