Artificial intelligence has started to occupy a unique place in our lives. In fact, no other technology has the kind of far-reaching implications as AI because it touches human lives across industries and geographies.
Beyond the relatively simple fear that AI applications will become knowledgeable enough to control humans, AI poses a number of less instantly dramatic conundrums with an impact just as profound.
There are numerous instances of AI being notoriously prone to biases, including racism and sexism. From my experiences working with L&T Infotech’s enterprise customers, I believe comprehensive corporate guidelines are the only sure way to govern AI to behave ethically.
AI in the insurance industry
AI and robotics applied to data offers promising benefits for the global insurance industry. The insurance industry
basically takes customer data, adds insurance expertise (read as “risk expertise”) and then churns out insights (read as “pricing and risk estimates”). The more varied and granular the input data is, the better the insights to the extent of hyperpersonalization, which could lead to better pricing and customer satisfaction. What makes it happen?
Lots and lots of data and AI, including sophisticated algorithms. All good!
In many cases, use of AI is fairly objective — understanding claims, detection of fraudulent activities, underwriting decisions and compliance checks, among other activities. However, when AI is applied to data about human beings, such as behavioral patterns, ethnicity and preferences, there could be potential issues involving policyholders and risks, and pricing and the best possible customer experience.
If the training data — which guides the AI-based decision-making — is not diverse enough, it may lead to discrimination against certain people.
AI delivers powers, such as image and voice recognition, predictive analytics and fraud detection, previously the exclusive domain of humans; and enables widespread automation of many functions, with all the advantages and disadvantages that entails. It is truly disrupting the complete value chain and opening doors for newer business models. Vehicle insurers started using machine learning on telematics data from connected cars to accurately price usage-based insurance premiums and better manage underwriting risk. Manufacturing companies are not just selling machines, but “machines as a service,” where customers don’t mind paying extra insurance premium fees if there is guaranteed quality and predictive maintenance of products.
Double-edged sword
The use of AI is a two-way street. Given enough data, an insurer can surely self-insure its property for natural calamities; an individual with favorable genetics can opt to forgo life coverage. However, one needs to be absolutely sure about biases in the data and algorithms baked into AI solutions. We are not dealing with movie recommendations where a wrong suggestion leads only to misspent money and two hours wasted. We are dealing with financial implications, coverage and associated perils, and there needs to be enough transparency about why the AI model recommended what it recommended.
At a broad level, three areas of concern relate to ethical implications of AI in insurance:
Price and claims optimization: In this digital economy, abundance of data is a given. Insurers have access to data on all factors influencing the risks to be covered, and about a policyholder’s lifestyle and purchase habits. When they start mashing up these datasets, they can infer many insights, such as which policyholder might be prepared to pay for insurance, or if the claimant would be happy to receive a quick but reduced claims settlement. From an insurer’s perspective, the price optimization business use cases are perfect; however, from a policy holder’s perspective, these could be unfair.
Profiling and bias: Now that technology has allowed for the collection of abundant data on everything, it also creates a serious concern regarding the usage of these data points. Why so? When the number of factors influencing risk ratings are suddenly increased, risk-related correlations pop up in all sorts of surprising places. If based on these correlations it is determined to whom to sell, what to price and what to cover, insurance firms may be getting into unfair means territory. Also, there are ethical implications as underwriting progresses toward automation: There’s an increased risk of discriminatory outcomes experienced by consumers.
Algorithmic accountability: Insurers have always been accountable for how they run their businesses. With sophisticated algorithms at play, insurers will find it difficult to maintain that accountability.
How would you understand what a self-learning algorithm, which is probabilistic by nature, is doing? And, what happens during the transition from largely human decision-making, to largely algorithmic decision-making? A significant gap is emerging between the algorithm’s behavior and the sense of accountability for the outcomes generated.
Delivering on AI’s potential, ethically
Scientists are still struggling to understand how our brains actually work, yet we’ve developed ethical frameworks, laws and policies to govern what we, as humans, do. The same philosophical scaffolding is now urgently needed to support the growth of AI. But the question to ponder will be whether development of ethics should be solely within the purview of humans or if machines can also play a role, as someday the “intelligence” manifested by machines will surpass the cognitive abilities of its makers.
How do we keep AI safe from evil intentions? The more powerful a technology becomes, the more possibilities it opens for malicious intent. What if AI agents become so focused on achieving their goals, that they recommend and implement actions that may bring disastrous consequences? For example, what if the goal of an AI system is to find solutions for cancer, and after careful considerations of numerous diagnosis results, root causes, treatment plans and effectiveness of medicines, it realizes that the most effective and best way of solving the cancer problem is to kill everybody on the planet? From a machine’s point of view, it has found the solution. From a human point of view, it is catastrophic. Hence, we must ensure that there are enough checks and balances in place before we start using AI systems across all spheres of our life.
For the insurance sector, getting experts together from various related fields of medicine, law, risk and regulations, technology, social policy, economics and ethics would be a strong foundational point for the industry to review and rebuild its ethical foundations. There is need to provide time, resources and support to pilot thought experiments to address bigger problems of turnaround time claim processing, revenue leakage and more.
Robust ethics, including legal, intellectual and corporate frameworks, will keep developing as the industry evolves. First, organizations need to develop a data-driven culture. Subsequently, they need to be mindful of regulatory and ethical considerations, and steer clear of dangerous myths. Finally, they need to foster a continuous learning laboratory for AI capabilities.
The goal of achieving a generally agreeable transparency level to establish a commonly acceptable set of ethical standards requires a tremendous attention to detail: Organizations implementing AI-based solutions need to detail the kind of data used, the data sources, the parameters used in the development of prediction models, the accuracy levels achieved, and statistics on false-positives and false-negatives. How the industry proceeds is up to these individual organizations — but the pace and magnitude of this change will be far greater than anything experienced before.
Soumendra Mohanty serves as executive vice president and chief data analytics officer at L&T Infotech, a global technology consulting and solutions firm. Contact him at info@lntinfotech.com.
