Mastering a new AI-driven insurance world

The accident involving a self-driving Uber car that killed a woman in Arizona in March 2018 has made headlines around the world. The vehicle was in autonomous mode when it hit the woman, who was walking outside the crosswalk and later died at a hospital.

“What we can learn from the Uber incident is that there will not be 100 percent security in autonomous driving even in an ideal world where all vehicles are autonomous,” says Michael Bruch, head of emerging trends at Allianz Global Corporate & Specialty (AGCS).

“An environment with zero accidents or fatalities is unrealistic. Imagine a child running directly in front of an autonomous car, the hardware would not be able to brake quickly enough to avoid an accident.”

Likewise it appears that even if a human had been in control of the Uber car, the accident could not have been avoided because the woman was too close to the approaching car. Nevertheless, the fatal accident has shocked the public and Uber has since suspended the testing of the technology.

The AI impact on insurance

Despite triggering some suspicion in parts of the public, artificial intelligence (AI) is expected to significantly improve safety in the mobility sector. It is estimated that it could help reduce the number of road accidents by as much as 90 percent, according to an Allianz report titled The Rise of Artificial Intelligence: Future Outlook and Emerging Risks.

For insurers, a lower risk of human error means that the frequency of claims is likely to be reduced, according to a SCOR newsletter on intelligent machines published in March.  Perhaps paradoxically, it could make some aspects of pricing more difficult as historical data loses its relevance and becomes scarcer, the newsletter says. The remaining causes of accidents will probably be machine features leading to errors in human decision-making or unsafe human behaviour that machines did not foresee, the reinsurer notes.

The progressive introduction of innovative safety technologies and processes in aviation, for example, has helped reduce passenger fatality rates by up to 10 percent per year over the past 20 years, according to SCOR. This has led to lower re/insurance rates as well as lower plane ticket prices; the motor industry could now benefit in the same way.

However, risks can become more complex and more expensive to insure because of AI. Increasing interconnectivity means the vulnerability of automated, autonomous or self-learning machines to failure or malicious cyber acts will only grow, as will the potential for larger-scale disruptions and losses, particularly if critical infrastructure is involved, the Allianz report noted.

SCOR added that a high dependence on data service providers such as GPS, communication networks or live data feeds means that single points of failure could bring a whole operation to a halt. In addition, an identical dysfunction such as a programming mistake or an inadequate software update in mass-produced machines could generate simultaneous and repeated deficiencies. Furthermore, one defective machine could autonomously execute the same erroneous activity numerous times.

As a result, the costs of claims may severely increase, owing to the growing complexity of machines and higher interconnectivity of the machinery that will impact the overall sequence of accidents, according to SCOR.

AI, which can be described as the ability of a computer program to think and learn like a human, has been spreading quickly around many industries in areas as diverse as medicine, finance, manufacturing, agriculture and marketing. From chatbots to autonomous cars, more widespread implementation of AI applications is expected to transform industries and society as it enables increased efficiencies, new products and replaces repetitive tasks.

The changing risk landscape

The spread of AI in the economy will affect and change the insurance industry in many ways. New risks associated with AI must be assessed, quantified, insured and mitigated against. As the decision-making is shifted towards machines, the risk is for example moving from personal insurance towards product liability, but how the claims process will be structured in future remains unclear, as the case of autonomous cars exemplifies.

“The car manufacturer itself may be accountable, but then also its suppliers and partners,” says Bruch. “It will become more difficult to trace what went wrong in a claims event. The network provider could for example be responsible for incorrect GPS information, or it could be the hardware or the software supplier of the autonomous car.”

The insurance industry, in case of an accident caused by an autonomous vehicle, will need to decide who is liable: the user, the manufacturer or the creator of the algorithm behind the technology, according to Allianz. The sector also needs to understand the consequences of an AI bug or an AI cyber attack. Insurers will need to manage the shift in their risk profile due to the impact of AI on biometric, property, casualty, financial, operational and strategic risks.

The deployment of intelligent machines—which SCOR describes as able to analyse and react to information about their surroundings, and self-manage over a long time with no human intervention—will result in making value chains more intertwined, the reinsurer predicts. Intelligent machines will blur the boundaries between personal liability, product liability and professional liability, SCOR says, adding that determining the cause and assigning responsibility for accidents will therefore become more challenging.

There are different options currently being discussed around how transparency can be introduced to data flows, Bruch notes. One difficulty may be to develop such a structure without breaching data privacy issues, but the aim would be to develop cross-industry data provider platforms to allow access to the necessary data and therefore a quick resolution of the liability issue, Bruch explains. Despite the advanced application of AI in the motor industry, here the insurance industry has yet to find a solution for the liability issue.

“Some argue that the car manufacturer should be the first body to be approached after an accident and then be responsible for claiming back cost from suppliers,” Bruch says. “It is, however, too early to make a definitive decision on the regime. The ultimate goal should be to ensure that the process of settling claims does not get too complicated or lengthy.”

SCOR believes that, in future, in the case of self-driving cars, various parties can be liable if an accident occurs. The responsibility of a car crash could for instance be attributed to the driver, for not correcting the course of the car; the manufacturer, who failed to implement sufficient warning signals; the software provider, if the software was unable to detect the obstacle; the repair workshop, for not replacing defective sensors such as radar or cameras; the digital map provider, who did not show the obstacle on the map; or the hacker, for manipulating software or data in the car.

Legislation which regulates cohabitation  and interaction between humans and intelligent machines is currently being introduced, as the EU General Data Protection Regulation (GDPR) and the changes to the Vienna Convention on Road Traffic,  and German motor regulation  illustrate, SCOR notes.

Laws changing the liability rules and liability insurance regulation may follow afterwards. Inside the vehicle, the notion of driver and passenger will disappear in favour of the notion of occupant. All victims on board a vehicle will be compensated regardless of whether the driver is potentially at fault, according to the reinsurer.

The ways risks are likely to change are manifold and present many challenges to the insurance industry. While in transportation concerns exist around who is liable in case of accidents, in healthcare, the use of advanced AI for care of elderly people and childcare is subject to risk of psychological manipulation and misjudgement, according to the Allianz report. Misuse of AI applied in security and defence may increase risk of cyber attacks if malicious hackers train AI to attack, including the utilisation of autonomous weapons, such as drones.

When AI turns evil

While some AI applications are already showing the benefits of the technology, some tests also show how much can go wrong. Take the Microsoft Tay AI experiment for example, where an AI bot named Tay was kicked off Twitter the same day it launched “for becoming a sexist, racist monster”. It showed how a chatbot can be ‘deceived’ into learning bad human behaviours, according to Allianz.

Meanwhile, Facebook recently  shut down an experiment when two chatbots developed their own language to talk to each other.

Major ethical concerns and societal implications will rise as machines gain more autonomy, according to SCOR. One major uncertainty is the potential ability of a machine to take a decision by itself without human intervention and what it would mean for liability.

The importance of ethical questions for automated machines becomes clear in a scenario where a self-driving car faces a potential collision with pedestrians and it is too late to brake. The only alternative is to change course, which would lead to a collision with an obstacle. The car needs to decide between continuing its route and driving into the pedestrians, potentially killing all of them, or deviating into the obstacle and avoiding the pedestrians, but possibly killing the car passengers.

Defining rules to frame the learning process and decision-making of intelligent machines will be difficult, all the more since ethics are different from laws and related to the cultural background, SCOR noted.

AI penetrates insurers

Potential failures are, however, unlikely to deter industries to continue developing and applying the technology, including the insurance sector itself.

AI will support underwriters with analysis of data and assessment of risks, the Allianz report states. There are many areas—reputation, cyber, supply chain and economic and climate risk scenarios—where machine learning could help companies better understand their risks.

AI could also work alongside other new technologies, such as blockchain, to enable new, faster and more customised services. For example, sensors on shipping containers are already providing data on the location and condition of cargo which, once analysed, can trigger insurance cover or mitigation measures if the goods are damaged, the Allianz report claims.

Insights gained from data and AI-powered analytics could expand the boundaries of insurability, extending existing products, as well as giving rise to new risk transfer solutions in areas such as non-damage business interruption and reputational damage, according to the Allianz report.

AI may help insurers to sort through and analyse customer information and provide accurate customer profiles to create successful individualised marketing campaigns.

AI applications can also support the process of recommending new products to prospective customers, Allianz predicts.

The manual processing and screening during a policy purchase can be automated and sped up through the application of AI, the report notes. Chatbots can assist customers of in-force policies on a 24/7 basis while policy adjustments such as portfolio diversification and risk profiling can be automatically performed by AI-based algorithms.

Robots gain control

Where tasks are repetitive they will be replaced by predictive analytics or data streams pulled in from outside the organisation, Bruch says.

“AI will not replace processes and underwriters completely, but it will free up underwriters’ time for client interaction, especially in the industrial insurance sector with tailored solutions at the end. For the industrial insurance business there will always be a portion of tailored solution and client interfaces,” Bruch notes.

The availability and timeliness of data on economic, demographic, environmental, and market conditions offers a vast potential for more refined definition of risk. AI supports interpretation of risk data to provide actuaries with cutting-edge models for efficient risk management, the Allianz report states.

“You can see that already happening with cyber insurance,” says Bruch. Allianz has, for example partnered with an IT security company which gives it insight into client performance in terms of cyber risk, IT and management, into how the IT architecture of a client looks, and how fast a client is able to restore the system after a cyber attack.

“This information is helping our underwriting and risk assessment to become much faster in a predictive analytics way and to assess and predict client preparedness,” Bruch says.

As AI allows for deeper insight into risk, insurers could become more selective in the risks they take on their balance sheet. For clients this could theoretically mean that some risks may become more expensive or even uninsurable. Bruch finds such a scenario unrealistic.

“Insurers may end up putting more risk on the balance sheet, more tailored risk, while offering consulting services to the client in loss prevention,” Bruch says. He points to cyber as an example, where insurers advise clients seeking higher limits as to how the clients’ risk and IT management can be made more robust.

“Around 70 percent to 80 percent of losses in cyber can be prevented by having a proper system and the right processes in place,” Burch says.

A shift in parameters

Eventually, developments in AI and data collection stand to completely alter the asymmetry of information between insurers and insureds, representing a game-changer for the re/insurance sector, according to SCOR CEO Denis Kessler.

Historically, the parties to an insurance contract—the insurer and the insured—have always had different sets of information, Kessler explained in a March article distributed by SCOR.

The asymmetry leads to strategic behaviours. The insurer will attempt to extract the maximum amount of data through questionnaires, observations and statistics, in order to infer how the insured will behave. The insured, for his or her part, may endeavour through strategic positioning to underestimate the risk, to maximise the value of the claim and to manipulate the price system to his or her advantage.

Developments in AI and data collection will completely alter this asymmetry by bringing comprehensive and dynamic observability to the insurance transaction, Kessler argues. Whereas information was previously incomplete, static, fragmentary and delayed, the new era of big data enables access to information that is comprehensive, accessible from multiple sources, ranked by quality and available on a real-time basis. Furthermore, the formerly high costs of obtaining information have been reduced to a minimal expense.

“Information is becoming a commodity, and AI will enable us to process all of it,” Kessler wrote.

AI will enable the re/insurance industry to improve the customer experience and to enhance efficiencies in underwriting, claims processing, risk analysis and product development. Tasks that once took months to finish can be completed in a matter of minutes, opening the gate for insurers to reap sizable cost savings.

Further, in enabling re/ insurance professionals to focus on value-added tasks and by alleviating administrative and process-related burdens, AI will augment re/insurers’ capabilities to analyse risk and design new products.

As a result, re/insurers will know their customers and risks more thoroughly, price and underwrite more accurately, better identify fraudulent claims, and detect and monitor evolving risks. They will be able to tailor products and services to the exact needs of their customers, when and as those needs appear and evolve.

Although re/insurers might have an early advantage given that they have greater means and more tools and, for the time being, more data, AI will transform both sides of the insurance transaction in the long run. All parties to the insurance ecosystem, be they risk carriers, brokers, or customers, will use AI tools, Kessler predicts.