Description: Artificial intelligence (AI) models have become increasingly popular in various industries, including finance. Such models can help lenders assess the risk of loan applicants, identify potential defaults and make better decisions about granting credit. However, the complexity of many machine-learning and, particularly, deep-learning models makes it difficult to understand how they arrive at their decisions. This lack of interpretability can be a significant problem in the legal and ethical realms, as it hinders transparency and accountability. Join us to learn interpretability techniques that are essential to address the challenges of using complex AI models in credit risk modeling. These techniques aim to provide a better understanding of how an AI model arrived at a particular decision or prediction. Techniques include surrogate models, which aim to provide a simpler model that approximates the behavior of the original model, and decision trees, which show the decision-making process in a more transparent way. However, applying interpretability techniques to AI models can be a challenging task. For example, some techniques can reduce the performance of the model, which can be unacceptable in applications where high accuracy is critical. Additionally, some techniques may only be effective for certain types of models or datasets, which limits their applicability. Despite these challenges, interpretability techniques are becoming increasingly important in the field of credit risk modeling. Financial regulators have recognized the importance of interpretability, and are starting to require financial institutions to use models that can be explained and validated. In addition, interpretability techniques can be used to improve the performance of credit risk models. By understanding how a model makes decisions, it is possible to identify areas where the model can be improved or where additional data may be needed.

Description: Artificial intelligence brings immense opportunities for productivity gains—and valid concerns. It can both support innovations in insurance products and in the customer experience as well as enhance our profession’s work. Actuaries share a responsibility for caution and professional judgment to use it without jeopardizing the fair and equitable treatment of individual consumers and the principals we serve.AI creates significant challenges for consumers: opacity, reliable accuracy, privacy, and verification of personal data. The Academy and its members have been working to educate and inform stakeholders, insurance commissioners, and regulators toward improving the transparency of algorithms that underlie the use of AI by insurers, to explain how the algorithms can be used to achieve equitable treatment, and to identify the issues and trade-offs related to ensuring consumer protections. We must amplify the actuarial perspective on how adequate disclosures, documentation, and governance of AI can help protect users. This session will examine how we can govern and support the evolution of better practices. We will discuss how AI is addressed via standards such as ASOP No. 56, Modeling, and other guardrails that are being developed to best utilize AI’s real-world potential and fulfill our professional responsibilities.

Description: This session includes a new idea based on dynamic treatment regime work (OWL) and integrating cost into this work. This will enable identifying the best treatment for a patient based on both outcomes and cost. This is novel for actuaries because in previous analysis we have used cost only in the after-analysis, whereas here we use it in the actual analysis and treatment identification. Moreover, this analysis provides a way to consider multiple treatments for a patient (including the no treatment option) and choosing among those multiple treatments, for both outcomes and cost. This session will share the background theory and set up a case study. The case study will parallel what is in the analysis in the book by Ian Duncan 'Predicting Hospital Readmission with Case Management Recommendation.' This analysis will then go further to compare a situation where there are multiple potential case managers. Also, this will potentially explore care pathways and choosing the best care pathway for a patient based on health outcomes and cost combined.

Description: The expanded use of predictive models, algorithms and external consumer data has drawn the attention of the life insurance industry to the matter of fairness and bias in underwriting, including proposed regulation in states like Colorado and New York and a draft NAIC Model Bulletin. Insurers will benefit taking a proactive approach to verifying that their practices, both current and future, do not result in unfair discrimination. This session will help attendees in understanding potential requirements and reporting insurers must address, techniques and tools for evaluating bias, and challenges in doing so. At the conclusion of the session, attendees will be able to: o Understand proposed regulations and practices regarding fairness and bias o Identify tools and techniques for evaluating bias in their processes o Discuss the challenges the industry will face in addressing fairness in underwriting

Description: Pandemics are a main driver of losses for both life and non-life (re-)insurance. Until COVID-19, risk assessment mostly relied on a few historic examples, most importantly the Spanish flu. The COVID-19 pandemic has shown that other pathogens (namely the coronaviruses) need to be taken into account. Also, modern societies and governments drastically respond to novel threats and implement measures such as social distancing, mask mandates, and lockdowns. Besides the properties of the pathogen itself, these reactions and the medical progression (namely the development of vaccines) are key influences on the course of a pandemic, and thus also the resulting outcomes. To account for the new insights, join us as we take a look at a stochastic, age-structured compartmental model that is calibrated on recent pandemics with several parameters sets to model different levels of reaction speed and stringency between groups of countries. Other important factors also must be defined in close collaboration with experts. Get ready to explore the development of a model and framework that can provide potential outcomes to be used as the basis to derive both summary and tail statistics to assess pandemic risk and help price complex deals.

Description: Artificial Intelligence (AI) has become a key component of the insurance landscape and getting full value from an AI depends on much more than simply building the model. Please join us to hear the challenges and potential solutions around successfully deploying an AI-driven initiative within an organization. Hear from panelists as they describe the human impacts of model implementation, how to assess a model for effectiveness and (lack of) bias, and detail the journey of implementing AI-driven underwriting models at a large insurance company.

Description: Further explore the pipeline of large claims approaching in the future in this webcast. Some of these upcoming claims include Cell and Gene Therapies, as well as high medical costs. In order to prepare for the multitude of claims, this session provides the insurance industry with an overview of what these large claims consist of. You can also expect important recommendations on how to properly prepare for them to succeed in this market.

Description: Web3, Crypto, and Generative AI are no longer just buzzwords as they reshape the tech landscape. Viewed by many as unconventional technologies, their potential to now revolutionize the insurance industry cannot be ignored. Join us for an enlightening session that delves into the transformative power of these technologies and their current applications within the insurance sector, as well as discover how forward-thinking insurance companies are embracing these tools to unlock innovation and challenge the insurance status quo.

Description: Attend if you are an organization interested in utilizing various kinds of client data to improve the whole insurance sales process, and to develop new applications. Interested parties may include distribution sales force, underwriting, claims, actuarial and finance departments within a typical insurance company, online distribution partners, health care providers and other consulting service providers. The research looks into various use cases, which are typically main issues in modeling in the insurance industry. The research also helps benchmark performance using different languages that may serve as a good reference when best to use a certain language to enhance the development cycle.

Description: ALM and hedging can reap significant benefits from innovation in computation and operations research. Explore applications of static and dynamic optimization in two different ALM and hedging use cases. Insurers are increasingly moving offshore by establishing Bermuda captive reinsurers. In calculating the Bermuda Best Estimate Liability ('BEL'), reinsurers will need to segment assets to support liabilities in each treaty. A Linear Programming ('LP') problem can be formulated to tactically select assets in a way that minimizes the BEL and consequently the total asset requirement. Learn more about what US Variable Annuity Insurers face with the regulatory requirement to reflect 'future hedging strategy' in their statutory funding calculation. The traditional numerical method of finding the optimal hedging target and then allocating assets may encounter the curse of dimensionality with expanding product and economic environment complexity. A Reinforcement Learning algorithm may well fit this problem and provide insurers a solution for dynamic hedging.