Generative AI – ongoing evolution or market revolution?

ChatGPT launched in November 2022 and overnight the technology became available globally. It sparked a huge amount of excitement and speculation about how generative AI might transform our personal and professional lives.

It also prompted a lot of reminiscing about the seemingly change-making inventions and tech trends of the past.?

Blockchain, for example, came on strong, but eventually became the hammer in search of a nail, which couldn’t be better serviced by a simple, cheap database. Low Code/No Code (or as I call it, “Some Code/No One Wants to Maintain”) has had a much greater practical impact on tech and digital initiatives than blockchain and has definitely democratized and shifted-left tasks that were previously part of the dark arts of coding, but has not, for a variety of non-technical reasons, crushed whole industries and obsoleted the skillsets and efforts of millions of coders worldwide.

Different than blockchain, Low Code/No Code, or even cloud, ChatGPT was a consumer-ready product that offered a simple user experience, and instant gratification.?

The nerd world was set ablaze with immediate applications and implications.

A gradual sobering

Generative AI (and all of its ecosystem and hangers-on) is different to those preceding it in the hype cycle. ?It seems impactful, and it seems applicable to a variety of real business problems and personal convenience needs today.? Its seeming simplicity (for end users at least) and ubiquitous applicability put it in a different class.?

On the surface, the search, research and assistant uses make sense: you get an answer rather than a series of links to potential answers.? And at a deeper level, despite early-days complexity, the tech basically makes sense.? Generative AI seems to have all the makings of a bona fide game changer.?

But as the initial blush of coolness and ease of personal use wore off, the more serious aspects of really understanding not just the potential uses, but the actual complexities and risks began to balance the effervescent exuberance.?As covered in an excellent article in The Information, while there remains much excitement and not a small amount of hype, sober thinking has engaged. Engineers and thinkers across industries and in academia are more seriously considering how to properly productize this new tech, while at the same time keeping data safe, and protecting both consumers and business from the scourges of bias, toxicity, drift, and evildoers looking to exfiltrate data for their nefarious purposes.

While slowing down the “joy train” a bit, this deliberate sobriety is good, and a natural progression in the tech world, but one which normally plays out over years, not weeks and months.?

Everything GenAI is fast.? Everything GenAI is changing and evolving quickly.? And as the benefits become better understood, so do the (potential) risks and costs.

How to responsibly harness the power of generative AI is a focus across all industries, and within the insurance industry.? Tokio Marine is no different. In the coming months and years, its importance to business and the wider insurance industry will, indeed, grow dramatically, making it all the more important that we keep up with understanding the risks, the tech and the costs (both fiscal and, yes, environmental). ??

So, exactly how will generative AI impact and enhance carrier, agent, and policyholder experience in the near term? Where will it add value to carrier operations and customer outcomes?

Sorting impact from fiction

While there’s enormous interest in the potential of generative AI, there’s also growing understanding that the technology isn’t yet mature or robust enough to be given acapella decision-making responsibility. There are enough risks and nuances that human-guided and supervised use will remain the norm (and the requirement for regulated industries) for quite some time. It’s useful to think of GenAI as a co-pilot or assistant that can do a lot of the heavy lifting and provide incredibly valuable insight and support without having the final say on business decisions.

While the perception is that GenAI is always learning, always incorporating, and never forgets, the reality is different.? GenAI does have memory issues (and hallucinations).

Generative AI is not really intelligence, but rather extremely detailed and deep pattern detection and prediction.?

As such, it is deficient in basic cognitive functions such as reasoning, comprehension, and strategic planning—competencies that are well-refined in humans. On the other hand, GenAI can access an extensive repository of stored knowledge, rapidly analyze new information, and generate new content, which are capabilities more complementary to humans. ?The symbiotic combination of human cognitive skills and GenAI's data processing proficiencies can yield highly beneficial and efficient outcomes.

So even its “assistant mode” – which is a very appropriate, readily applicable use – needs supervision.? As one of my colleagues, Stephen Haldis, recently coined:

“Co-pilot is not autopilot.”

Flight safety

The diverse, global nature of Tokio Marine Group means we can explore the use of generative AI in different markets and territories, and then share the learning from these pilots across our portfolio of businesses.

Tokio Marine Group is experimenting with the technology across myriad proofs of concept and pilots in markets on five?continents. These efforts range from using generative AI to create summaries and first drafts for different business documents in the UK to reviewing the completeness of the third-party contracts held by policyholders to ensure compliance with underlying policy requirements, industry obligations, and regional/national legislation for one of our larger carrier companies in the US.

We’re exploring how generative AI and integrated capabilities can enhance our customer service operations, from truly human-like multilingual chatbots in Brazil, to real-time customer service advisory and sentiment analysis in Japan and the US. ?

We are also exploring how generative AI might speed and inform certain claims reserving activities, still in the form of advice to real human actuaries who are accountable. ?And in Japan, we launched an internal “Tokio GPT” for 14,000 users in our flagship Tokio Marine Nichido Fire business, which has been exploring the potential of generative AI since 2022, even before ChatGPT was launched to the public.

These pilots need to be detailed, in-depth, and rigorous. Insurance is a highly regulated and carefully audited market. Carriers must be able to show their work and detail precisely – both now and in the future – the calculations underpinning their decisions on pricing, liability, and claim settlements. ?In the next couple of years, the industry will be humming with stories about all sorts of generative AI pilots and initial forays into production use. But it won’t be until after these trials have completed that we begin to see projects aggregating into something that has hands-off production impact at the enterprise level for insurers. ?????

In the first instance, projects will focus on internal processes such as content generation and synthesis, summarization and extraction, content search and analysis, and code generation and conversion.

These uses are less risky generally, falling into areas of less concern for regulators, and many are purely internal.? As the technology and the governance surrounding it matures, GenAI will find its way into more client-facing operations and processes, but these will take more time to develop, refine, and implement.?

It’s tempting to believe that new technologies will supersede everything that’s gone before them. But the experience of the past three or four decades highlights that those who are the most thoughtful in their adoption of technology tend to generate the best results for their organizations and their clients.