Legal and ethical constraints in the use of Artificial Intelligence ...

Legal and ethical constraints in the use of

Artificial Intelligence (AI) by Swiss health

insurance companies

Bachelor Project submitted for the degree of

Bachelor of Science HES in International Business Management

by

Kerim TRUNIGER

Bachelor Project Mentor:

Christophe COURBAGE, Professor

Geneva, 21st August 2020

Haute école de gestion de Genève (HEG-GE)

International Business Management

Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER i

Disclaimer

This report is submitted as part of the final examination requirements of the Haute école

de gestion de Genève, for the Bachelor of Science HES-SO in International Business

Management. The use of any conclusions or recommendations made in or based upon

this report, with no prejudice to their value, engages the responsibility neither of the

author, nor the author’s mentor, nor the jury members nor the HEG or any of its

employees.

Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER ii

Acknowledgements

This project has been completed with the support of various persons. I, therefore, wish

to thank them for their assistance and encouragement.

First, I would like to thank my mentor, Mr. Christophe COURBAGE, for his helpful

supervision, guidance, and advices throughout the entire preparation of this work and for

providing me with helpful sources and contacts.

Then I am also grateful to Ms. FLUECKIGER, Director New Technologies & Data at the

Geneva Association, Mr. IRIA, Head of Digital Transformation at La Mobilière and to the

three other persons working in the insurance field for accepting to provide me with

interesting and helpful answers to my interviews.

Finally, I also would like to thank Mr. KOCH, the Innovation Scout Outpost Silicon Valley

and the current Head of services, Underwriting and Legal Department at La Mobilière for

the discussions we had and for putting me into contact with various people of the

insurance industry, and Ms. MORALES, legal counsel at Barclays for her guidance

throughout the legal framework of this project.

Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER iii

Executive Summary

Experts in new technologies and economists seem to agree: robotics and deep learning

by machines will change most of the economic activities in the coming decades, at a

very fast pace. If in the past, technological changes mainly brought about gradual

increases of efficiency in human work; robotics and the availability of big data, used with

Artificial Intelligence (AI) will bring about exponential change to the working world of

humanity, in the very near future. It seems almost impossible to predict with certainty

what the social, economic, and psychological effect on humankind will be. Simply

learning from the experience of past technological evolutions is not enough. Many

experts foresee more of a disruption rather than a further, gradual development.

Among various other economic actors, health insurance companies may find many

benefits by introducing and developing digital technologies such as AI. However, they

must consider different limits and risks, which can restrain them in the development of

AI. One of those limitations is the complex legal framework that regulates data use and

protection, particularly when the available massive data and sources include the

personal data of individuals. Furthermore, in order to preserve their trustworthiness and

reputation, they must comply with ethical boundaries to retain their current customers

and acquire new ones.

Only then will health insurance companies be able to take full advantage of the benefits

of these new technologies and obtain valuable return for their sizeable investments in

such new tools.

This research has the ambition to produce recommendations to Swiss health insurance

companies on how to best achieve balanced, profitable returns through the gradual

development and use of AI, within the legal and ethical limits. It further considers the

important hurdles and barriers which the introduction and expansion of such new

technologies will meet, both internally, with their customers and in the large public, and

provide some recommendations which need to be taken into account in the change

management process.

A thorough background study and examination of empirical data obtained from insurance

experts and staff will provide a solid basis to formulate some main recommendations in

how to deal with the trust of customers, the education of the staff, with legal and ethical

limits and the necessary required adjustment of governance processes.

Legal and ethical constraints of the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER iv

Contents

Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss

health insurance companies ...................................................................................... 1

Disclaimer ..................................................................................................................... i

Acknowledgements ..................................................................................................... ii

Executive Summary ................................................................................................... iii

Contents ..................................................................................................................... iv

1. Introduction .......................................................................................................... 1

2. Background .......................................................................................................... 3

2.1 What is Artificial Intelligence? ................................................................. 3

2.2 AI in Swiss health insurance companies today ...................................... 4

2.1.1 Opportunities in the use of AI .................................................................... 6

2.1.2 Risks in the use of AI .................................................................................. 8

2.1.3 Legal limits ................................................................................................ 10

2.1.4 Ethical limits.............................................................................................. 15

3. Analysis .............................................................................................................. 18

3.1 Methodology ........................................................................................... 18

3.2 Interviews ................................................................................................ 18

3.3 Results .................................................................................................... 20

4. Discussion .......................................................................................................... 27

5. Recommendations for health insurance companies & conclusion ................ 29

Bibliography .............................................................................................................. 31

Appendix 1: Interview with Isabelle Flueckiger ....................................................... 36

Appendix 2: Interview with José Iria ........................................................................ 42

Appendix 3: Interview (Confidential) ........................................................................ 46



Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER v

Legal and ethical constraints of the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER 1

1. Introduction

From the 1990s onward, computer technology has transformed our world more than any

other force. Despite its massive impact, most politicians seem hardly able to comprehend

this innovation and are even less capable of controlling it. The repercussions of this

confusion for our political processes are disturbing. Just imagine a future where

governments must patiently wait for algorithms to give them the green light on their

budget or their tax-reform plans. Unfortunately, for many twenty-first-century politicians,

technological disruption is not at the top of the agenda. Ordinary people across the world

are feeling more and more irrelevant in this brave new world of artificial intelligence,

globalization, and machine learning. And this fear of becoming irrelevant has made them

desperate to wield whatever political power they still have before it becomes too late.

Throughout the twentieth century, common workers have worried about their labor being

exploited by economic elites. But these days, the masses are more afraid of losing their

economic status in a high-tech economy that no longer needs their labor at all. Although

most experts agree that robotics and machine learning will change nearly all lines of work

in the coming decades, we cannot predict how this change will look like. Will billions of

people find themselves economically irrelevant within the next twenty years, or will

automation result in wider prosperity and great new jobs for all? Many optimists point at

the industrial revolution in the nineteenth century, a time when the fear that new machine

technology would create mass unemployment was widespread. Since the start of that

industrial revolution, it was pointed out that the development of new technologies has

created a new job for each one it made obsolete. Unfortunately, there are good reasons

to assume that, in the twenty-first century, the impact of new technology on human

employment will be much more destructive. Just consider the fact that humans are

possessed of two sorts of abilities – cognitive and physical. In the previous industrial

revolution, humans experienced competition from machines largely in the realm of purely

physical abilities. Our cognitive abilities, meanwhile, remained far superior to machines.

Therefore, even as automation occurred in manual jobs within industry and agriculture,

there concurrently emerged new jobs that required the sort of cognitive skills particular

to humans – such as analysis, communication, and learning. But in this century,

machines are getting smart enough to compete for these cognitive-based jobs, too.

Recently, neuroscientists have discovered that many of our choices, preferences and

emotions are not the result of some magical human faculty, such as free will. Instead,

human cognition comes from our brain’s ability to calculate different probabilities in the

space of a split second. These neuroscientific insights raise a troubling question: Will

Legal and ethical constraints in the use of Artificial Intelligence (AI) by Swiss health insurance companies Kerim TRUNIGER 2

artificial intelligence eventually outperform people in professions requiring “human

intuition,” such as law and banking? It is highly probable. Computer scientists now know

that what looked like impenetrable human intuition was just our neural networks

recognizing familiar patterns and making fast calculations about probabilities. Therefore,

in the twenty-first century, computers might be able to make banking decisions about

whether to lend a customer money, as well as accurately predict whether a lawyer in a

court case is bluffing or not. In other words, in the years ahead, even the most cognitively

demanding jobs may not be safe from automation. (Harari 2018)

AI is growing nowadays exponentially and will soon be used in many areas of our society

such as insurance, healthcare, legal, marketing and advertising and security. Hundreds

of millions are being invested into AI and an increasing number of insurance companies

are starting to introduce it into their daily activities. Key players of the IT industry and

their implementing partners such as the big four and others, are currently massively

promoting this technology and have published numerous reports and project findings

about its implementation in the next decade. Insurers on their part realize that they could

benefit considerably from this new technology, which could help in reducing costs,

speeding up processes and thereby gaining time and efficiency, detecting fraud, and

reducing health costs. However, they also realize the limitations of these technologies.

The further artificial intelligence will be enhanced, the more it will hit legal and ethical

constraints that may affect the development of this technology, particularly in the health

insurance sector. (Michael 2019)

The use of AI technology requires important amounts of data. Health insurance

companies gather huge quantities of publicly available and private information about their

customers. They, therefore, have a high responsibility regarding the security of data and

protection of its privacy.

In addition to the traditional information stemming from questions asked to the insured,

insurance companies are now getting further data through the appearance of two new

sources: The first one is continuously generating information through our online activities

such as private information published on social media platforms, personal online

shopping habits gathered through e-commerce; the second one is data developed

through our private search behaviors and searching activities. (Keller, Eling, Schmeiser,

Christen, Loi 2018) (Michael 2019)

Insurers must therefore be fully transparent concerning their use of data and be

compliant with the data protection specifications in order to be able to keep their


customers and their trust. (Schmidt 2018) AI will have many positive impacts, but it will

also bring about important risks and challenges. (Nakonz 2019)

This study investigates the use of AI in the health insurance sector in Switzerland with a

particular focus on its potential and limits in the relationship with the insured.

Digital transformation and the current technological growth are bringing up many

debates. The question treated in this thesis is the following: What will be the impact of

the use of AI on Swiss health insurance market and what are the legal and ethical limits

of its use?

The main objectives of this research are therefore: to assess the general impact of the

application of AI enhanced computer programs on Swiss health insurance companies;

to determine the present and future legal and ethical limits of such developments, with

the purpose of providing a list of recommendations for Swiss health insurance providers

on how to best implement and develop AI into their processes. The first part of this study

presents the general background and information regarding the current sources,

researches, articles, and literature found about this subject. The analysis will then

describe how the different interviews have been conducted and will synthetize the

various answers. These will then be discussed, and finally, the report will end with some

recommendations for health insurance companies based on the findings with a general

conclusion.

2. Background

2.1 What is Artificial Intelligence?

Providing a plain definition of Artificial Intelligence is not simple since the scope is wide

and various viewpoints are the result of many different definitions. (Dignum 2019)

Consequently, the following definition has been established based on multiple sources:

AI is the ability for computer systems to learn over time from past data and algorithms,

improve themselves and adapt to new data and therefore achieve tasks that usually

require human know-how and intelligence. (Shaw 2018) (Panchal 2019) (Shroff 2019)

(Copeland 2019)

Various other definitions are provided in the book ‘Responsible Artificial Intelligence’ from

Ms. Dignum:

“One of the simplest definitions of an intelligent system is that of a system that ‘processes

information in order to do something purposeful’.


Another common definition explains AI as a computational artefact built through human

intervention that thinks or acts like humans, or how we expect humans to think or act.

This is the definition put forth by McCarthy, Minsky, Rochester and Shannon in the

classic ‘Proposal for the Dartmouth Summer Research Project on Artificial Intelligence’,

the founding document that established the field of AI in 1955: “For the present purpose,

the artificial intelligence problem is taken to be that of making a machine behave in ways

that would be called intelligent if a human were so behaving.”” (Dignum 2019)

Simple pieces of information may have little value alone, but when combined with others

they can tell much more. This is also true for datasets: often, their value becomes

apparent only when combined with other datasets. Trends can then be found in the newly

combined data that were not discoverable from the individual datasets alone. (Mayer-

Schönberger, Cukier 2013) For instance, a US traffic-analysis company called Inrix,

gathers real-time location data from cars, commercial fleets, and road sensors through

their smartphone app. Each one of these datasets is not of much use individually, but by

combining them, Inrix can create timely data on traffic flows and jams for its customers,

in return for a fee for their services. Moreover, any competent IT person can also make

the datasets show what s/he wants to, by taking into account some parameters and not

others. (INRIX 2020)

2.2 AI in Swiss health insurance companies today

The compulsory health insurance in Switzerland is based on the Law on Medical

Insurance, “Loi sur l’Assurance Maladie” (LAMal) where the insurance benefit is defined

by the government, the “Office federal de la santé” (OFSP). On the other hand, the

complementary health insurance is regulated by the Insurance Contracts Act, “Loi sur le

contrat d’assurance” (LCA). In comparison to many other insurance fields, the Swiss

health insurance sector is highly regulated, closed and protected. For instance, both in

the basic and complementary medical cover, insured can chose their insurer only among

those insurance companies which are accredited in the Swiss market. The principle of

solidarity for the basic part of the cover is intrenched in the relevant laws, norms and

rules, and loss adjustment is regulated to the most detailed level. This report will provide

information related to the use of AI with a focus on complementary Swiss health

insurances since its market is more liberal than for the basic cover.1

(TheSwissauthoritiesonline 2020)

1 Interview with Isabelle Flueckiger, Director New Technologies & Data at the Geneva

Association, Geneva, 29 April 2020 (cf appendix 1)


Digital technologies such as Big Data analytics, Artificial Intelligence, and the Internet of

Things are, nowadays, increasingly used by many important actors of our society. The

Swiss health insurance industry is benefiting from new data sources that become

available from such technological advances. The development of Artificial Intelligence in

the health insurance industry is triggering important changes in the type of risks covered

by the insurers and the way they underwrite, conduct, and manage claims. The large

amount of personal data available is also changing the relationship between the

insurance and the customers, as insurances tend to become even more client focused

through this new technology. The use of such technological advances has the ability to

produce important economic and societal advantages. However, it also leads to

challenges, risks and potential costs to the insurance industry, the insured and to the

overall society. (Schmidt 2018)

Using such technologies requires the sharing of huge quantities of personal data and it

therefore creates important tradeoffs, both for the insurer and the insured. Thus, it is

crucial for the health insurance industry to find the right balance between respecting

privacy when using personal data and permitting new technologies and innovation.

Indeed, not having concrete measures related to privacy protection may negatively affect

the insured and result in mistrust, while, having excessive limitations could prevent

society from taking advantage of the benefits resulting from technological advances and

data sharing. (Keller, Eling, Schmeiser, Christen, Loi 2018)

In Switzerland, there are various initiatives related to Artificial Intelligence that aim at

enhancing the use this technology. The ‘Swiss Alliance for Data-Intensive Services’ is

an initiative in which the group of experts “DataEthics” develop an ‘Ethical Codex for

Data-Based Value Creation’ that aims to strengthen trust in the use of AI. Another

initiative, called ‘SwissCognitive’, was launched by various industrial companies and by

the service sector that are entirely committed to AI. It is used as an exchange and

networking platform. Another initiative called ‘Digitalswitzerland’ aims to strengthen

Switzerland’s position as a leading innovation hub. And last, ‘Industrie 2025’ is a Swiss

AI-initiative that informs, raises awareness, and assists relevant stakeholders with

analyses and advices about the Industry 4.02. There are also various further specialized

groups and associations dedicated to research in digital technologies such as AI. (SEFRI

2019)

2 Industry 4.0 relates to a new era in the Industrial Revolution that centers around

computerization, machine intelligence, operational data, and interconnectivity.


2.1.1 Opportunities in the use of AI

The use of technological advances such as Artificial Intelligence by Swiss health

insurance companies is providing many advantages and benefits in several different

sectors. Thanks to the automation of the insurance procedures, underwriting is being

improved, becoming smoother and more efficient. Indeed, classic underwriting methods

do distinguish and select the different risks, but requires a lot of time, and creates high

administrative costs. With the new availability of data and sources, their faster

assessment and the current technological advances that enable a timely and efficient

data analysis, the underwriting processes length is reduced. The risk assessments are

faster, risk selection is improved and thus, allow more personalized rates. While,

insurance policies are often considered too expensive, this technology can produce

personalized tariffs according to the client’s parameters such as their lifestyle, financial

stability, illnesses and more. (Schmidt 2018) (Michael 2019)

Machine intelligence can analyze, identify, and extract information from various

documents and pictures. This “computer vision” enables the insurance company to

automate daily tasks that usually require manual human know-how and intelligence such

as extracting information from written documentation and pictures when underwriting or

processing claims. In underwriting, “computer vision” is utilized to check and approve the

information and documents provided by the client. It can, therefore, ensure an adequate

coverage rate and assist in the identification, prevention, and mitigation of insurance

fraud. Indeed, fraudsters tend to pursue similar patterns that can easily be spotted by AI

in a very short time span. “Computer vision” is also used in the processing and inspection

of claims to verify and authenticate the pictures and to retrieve data from documents

provided by the client. This enables an efficient classification of claims and

automatization of their related processes. Therefore, it speeds the procedures and

correspondingly reduces the risk of human error in the different processes of insurance.

For instance, an insurance company uses deep machine learning in its process of

recognizing the type of document and extracting information from different medical

records such as prescriptions, receipts, and other medical documents. The engine

(algorithm) extracts detailed information from medical bills such as the sum, billing

number, dates, fiscal identification and determines the medical care and examination. It

then, in a matter of seconds, pairs the data collected with the corresponding policy

coverage to the insured. Furthermore, AI is very useful to health insurance companies

as it can partially replace manual tasks that usually require important amounts of

workforce and time. For instance, the development of human language technology


enables machine intelligence to speak and communicate with human customers through

conversational agents such as chatbots. These online chats are operating 24/7 and can

recognize and answer to sophisticated client requests concerning health insurance.

They, therefore, enable an improved customer satisfaction and allow the insurer to be

more efficient and performant by handling significant amount of client inquires. Chatbots,

can have various functions such as, providing customers with the needed information,

guiding customers through insurance-based processes such as the submission of a

claim and can even execute and handle a commercial transaction. (Keller 2020)

Machine learning systems are proficient in identifying patterns and relationships among

sophisticated sets of data in manners that would be extremely complex or even

inconceivable for human beings. The detected patterns are used as ground in the

analytical processes, more specifically in essential tasks for the insurance industry such

as classifying, clustering or in the regression analysis. AI can generate very precise

predictions since it is able to assimilate complicated nonlinear correlations between large

sets of variables. The predictions provided by intelligent systems are, therefore, much

more accurate than those resulting from standard modelling methods which often are

based on linear models commonly used in insurances. Nowadays, this ability of AI is,

therefore, being used to assist human decision-making. For instance, by using AI as an

assistant, insurance salespersons can provide customer services that are much more

personalized. This increased personalization of the customer services is possible thanks

to the better understanding of the customer’s information such as their interactions,

products, claims and location data. The salespersons are then provided with suggestions

and opportunities to sell new or supplementary products and services to the existing

customers, although, the sales agents retain the decision to advise customers with

particular product offerings. These types of applications have shown great effectiveness

in the improvement of the insurance sales channel. (Keller 2020)

The promising benefits of using AI goes further than enhancing the insurance industry

and the insured. As an example, the use of AI can encourage the extension of the

insurance coverage to groups of customers such as recent and formerly underinsured

or uninsured individuals. It can also help in the expansion of the scope of risks that can

be covered, thus, enabling the enlargement of the range of risk pooling, which is one of

the essential socio-economic function of insurances. The cost of risk pooling can also be


reduced through the automation of particular tasks, the increased precision in the

assessments of risks and the reduction of adverse selection3 and moral hazard.4

Moreover, using AI could bring innovative risk understandings that may lead to a better

mitigation and prevention of risks. Risk reduction can be enhanced through an improved

alignment of premiums and risks. AI technology can help us consider every and all factor

or specific/smaller datasets to model further/possible needs. Furthermore, enhanced

data would help to establish developed risk management and early warning procedures

enabling timely actions, thus, reducing losses. Hence, employing Artificial Intelligence

could encourage the extension of the primary role of insurance which is pure risk

protection, shifting it to prediction and prevention of risks by developing real time models

with different scenarios specific to each individual. (Keller 2020)

2.1.2 Risks in the use of AI

Artificial Intelligence learns over time and is based on transaction of significant amounts

of data. Their ability to learn, therefore, depends on the quality of the data which they

use. This is risky because if there are any biased features in the data, the machine will

learn them and will therefore be biased itself. (Keller 2020) As an example, in 2016,

Microsoft published on Twitter a new chatbot called “Tay”. This chatbot assimilated the

language used by the users of the social network and within a few hours after being

released, “Tay” started to post racist and sexist statements. (Schwartz 2019)

Intelligent systems are very complex and challenging to understand. It is therefore very

complicated to figure out and explain why and how a system obtained a result or took a

specific decision. Moreover, similar to human beings, they can make mistakes, even if

the data used does not include any biased features and one error/fake trend could lead

to catastrophic forecasting. (Keller 2020)

Additional risks related to the storage, use and processing of personal information

include the potential to being hacked or criminally abused, the risk of misuse of private

data resulting from illegal operations or through unintentional transmission of data.

(Keller, Eling, Schmeiser, Christen, Loi 2018) The huge amount of data gathered for

3 Adverse selection arises when the insured has more information about its risks than the insurer.

In this case, high-risk persons purchase insurance whilst low-risk persons do not. As a result, insurance companies raise premiums and limit coverage.

4 Moral hazard arises when the insureds have an increased risk exposure because they know

that they will not suffer the cost of their risks.


enabling an efficient functioning of AI carries a risk of setting the basis for surveillance

regimes. Data security is very expensive as it requires focused attention and needs to

be continuously updated by specialists. (Nakonz 2019)

Further concerns arise from the risk of having well-established technological

corporations, such as Google, that might connect to the insurance industry and take

advantage of its monopolistic access to person-related information to obtain uneven

amounts of profits. This would result in having the insurance industry discriminated and

would dump its competitiveness. (Schmidt 2018)

As our ability to gather and asses data has developed, we have increasingly tried to use

data to improve many aspects of life. However, this ability does come with certain

potential threats. Being excessively data-driven can be problematic because we run the

risk of allowing data that is biased to influence our actions. The profusion of detail and

insight provided by big data creates a risk where the society could lose perspective and

become so fixated on data that it would neglect its limitations or fail to ensure its quality.

This would enable the data to govern in manners that could be more harmful than

beneficial. (Mayer-Schönberger, Cukier 2013)

A relevant example that highlights one issue about data quality is provided by a Geneva

based organization called “Women at the Table”:

“20th century drug trials, the design of safety features in cars, medical treatments, the

work equipment we wear, to name a few examples, are based on data that uses the

default of a “standardized male”, a default of flawed systems and cultural standards that

currently form the physical framework and infrastructure of how we live and work. These

defaults are so normalized we don’t even notice them, yet they have proven to harm and

lead to dangerously wrong consequences.” (Kraft-Buchman, Arian 2019)

The digital field is exploding with new and ever-growing technologies that are leading

and have led to a very quick increase in the volume of data available. Data is nowadays

considered as a new precious resource and is part of our new future. Yet, society should

be cautious and avoid becoming excessively data-driven since it could be risky. It might

be basing itself on biased data and, therefore, relying on inaccurate results. Note further

that big data only takes into consideration past events. (Mayer-Schönberger, Cukier

2013)


2.1.3 Legal limits

The use of AI as technology in Switzerland is not subject to any law or regulation.

However, in order to learn and operate correctly, such technology requires large

quantities of data. Swiss health insurance companies are, therefore, required to be

compliant with the regulations and laws related to the protection of data, when using

private information for their intelligent systems. (Gordon, Gurovits 2020)

Private customer information or personal data is defined by the Swiss Medical Network

as: “any information relating to an identified or identifiable natural person, such as a

name, an identification number, location data, an online identifier, date and place of birth,

professional information or to one or more factors specific to the physical, physiological,

genetic, mental, economic, cultural or social identity of that natural person.”

(SwissMedicalNetwork 2020)

In the basic health insurance, customers have the right to access and view their own

personal data obtained by the insurance company according to Art.47 of the Federal law

on the general part of social insurance law. (LPGA 2019) Customers have the same

rights in the complementary health insurance according to Art.8 of the Federal Act on

Data Protection (FADP)5 and according to Art.15 of the General Data Protection

Regulation. (GDPR Art.15 2019)

Swiss health insurance companies must be compliant with two different data protection

laws. The first and main data protection legislation is the Federal Act on Data

Protection (FADP) issued by the Federal Assembly of the Swiss Confederation in 1992.

This law derives from Art. 13 of the Federal Constitution that states:

“1 Every person has the right to privacy in their private and family life and in their home,

and in relation to their mail and telecommunications.

2 Every person has the right to be protected against the misuse of their personal data.”

(CC 101 Federal Constitution of 18 April 1999 of the Swiss Confederation 2020)

The FADP is only binding in Switzerland for either private entities or federal institutions

that process the data of natural or legal persons. It aims to safeguard the fundamental

rights and privacy of individuals when their personal information is processed.5

According to Art.3 e, the processing of data means: “any operation with personal data,

5 (CC 235.1 Federal Act of 19 June 1992 on Data Protection (FADP) 2019)


irrespective of the means applied and the procedure, and in particular the collection,

storage, use, revision, disclosure, archiving or destruction of data”. (FADP art. 3e) The

Ordinance on Data Protection Certification6 and the Ordinance to the Federal Act Data

Protection7 are two complementary standards and regulations to the FADP. The

processing of the individual’s personal information is further subject to rules and laws at

a cantonal and municipal level. (Commissioner (FDPIC) 2015a) The Federal Data

Protection and Information Commissioner (FDPIC) is the competent body that is

responsible for the supervision of the processing of private data by federal entities,

corporations, and private individuals. The supervision of data processing at a cantonal

or communal level is up to their own data protection agent. (Commissioner (FDPIC)

2015b)

Artificial Intelligence and Big data are directly linked. AI requires large quantities of data

to learn over time and operate accurately. Big data methods employ AI for the extraction

of value from major data records. However, AI and Big data were not considered when

the Swiss laws on data protection were established. The FADP applies only when private

data is processed and does not apply on computerized geographical data or anonymized

data (where no relationship to an individual can be determined). Nevertheless, Big data

incorporates massive volumes of data and, therefore, makes it easier to identify

individuals. The Swiss laws on data protection may therefore be applied, although, the

data processed has been anonymized at a certain moment. Once the FADP applies, the

processing of data must comply with the principles stated in art.4 of the FADP on the

processing of data which are the following:

“1 Personal data may only be processed lawfully.

2 Its processing must be carried out in good faith and must be proportionate.

3 Personal data may only be processed for the purpose indicated at the time of

collection, that is evident from the circumstances, or that is provided for by law.

4 The collection of personal data and in particular the purpose of its processing must be

evident to the data subject.

6 (CC 235.13 Ordinance of 28 September 2007 on Data Protection Certification (DPCO) 2016)

7 (CC 235.11 Ordinance of 14 June 1993 to the Federal Act on Data Protection (OFADP) 2012)


5 If the consent of the data subject is required for the processing of personal data, such

consent is valid only if given voluntarily on the provision of adequate information.

Additionally, consent must be given expressly in the case of processing of sensitive

personal data or personality profiles.”

(CC 235.1 Federal Act of 19 June 1992 on Data Protection (FADP) 2019) (Gordon,

Gurovits 2020)

The FADP has been revised and a new draft has been presented by the Swiss Federal

Council in September 2017. The revision has been drafted based on the General Data

Protection Regulation (GDPR) of the EU and is scheduled to be finalized at the end of

the year 2020. The main purpose of this revision is to align Swiss law on data protection

with the regulations of the European Union. (Wellens 2020) The revision is done to

increase the Swiss degree of data protection and to bring it up to the level of the GDPR

in order for Switzerland to still be accepted as a third country and thereby keep the ability

to exchange data with countries of the European Union. (Tischhauser, Georgieva 2019)

It will incorporate the ongoing technological developments and reinforce the protection

of private data for natural individuals. The main differences that are included in the

revised FADP are that the regulations only apply to natural persons and no longer to

legal persons. It also states clear penalties. An intentional violation of the regulations

from an individual would result in a fine up to 250’000 CHF. Moreover, new types of data

such as biometric and genetic data have been added and now fall under the regulations.

(PricewaterhouseCoopers 2020)

The second law with which certain Swiss companies must comply is the General Data

Protection Regulation (GDPR). It is the European Union law that sets regulations for the

gathering and processing of personal data from persons living in the EU. It was

introduced in May 2018 and applies to any Swiss companies that process personal data

if the company has an office located in the EU, whether a subsidiary or a local agency

acting for account of the Swiss holding corporation and processing personal data. It also

applies to Swiss corporations that are not settled or have a branch in the European

Union. For instance, if the company proposes online goods and services to a neighboring

EU country such as France, it would have EU customers and, therefore, the ability to

examine behavior patterns of EU individuals through website cookies. The Swiss

companies that fall under GDPR regulations without having an office located in the EU

are required to nominate a regional representative in the concerned countries.8 In certain

8 See exceptions in Art. 27 (2) GDPR https://gdpr-info.eu/art-27-gdpr/

https://gdpr-info.eu/art-27-gdpr/


circumstances, companies also need to nominate a data protection officer (DPO)9.

Corporations that fall under GDPR must comply with multiple standards and regulations.

(European General Data Protection Regulation GDPR 2019)

As with the FADP, the GDPR has in Art.5, a set of principles10 to respect regarding the

protection of data when processed. The six main principles are the following:

“1. Personal data shall be:

(a) processed lawfully, fairly and in a transparent manner in relation to the data subject

(‘lawfulness, fairness and transparency’);

(b) collected for specified, explicit and legitimate purposes and not further processed in a

manner that is incompatible with those purposes; further processing for archiving

purposes in the public interest, scientific or historical research purposes or statistical

purposes shall, in accordance with Article 89(1), not be considered to be incompatible

with the initial purposes (‘purpose limitation’);

(c) adequate, relevant, and limited to what is necessary in relation to the purposes for

which they are processed (‘data minimisation’);

(d) accurate and, where necessary, kept up to date; every reasonable step must be taken

to ensure that personal data that are inaccurate, having regard to the purposes for

which they are processed, are erased or rectified without delay (‘accuracy’);

(e) kept in a form which permits identification of data subjects for no longer than what is

necessary for the purposes for which the personal data are processed; personal data

may be stored for longer periods insofar as the personal data will be processed solely

for archiving purposes in the public interest, scientific or historical research purposes

or statistical purposes in accordance with Article 89(1) subject to implementation of

the appropriate technical and organisational measures required by this Regulation in

order to safeguard the rights and freedoms of the data subject (‘storage limitation’);

(f) processed in a manner that ensures appropriate security of the personal data,

including protection against unauthorised or unlawful processing and against

9 See Art. 31 GDPR https://gdpr-info.eu/art-31-gdpr/





accidental loss, destruction or damage, using appropriate technical or organisational

measures (‘integrity and confidentiality’).

2. The controller shall be responsible for, and be able to demonstrate compliance with,

paragraph 1 (‘accountability’).”(GDPR Art.5 2019)

Clear penalties are also stated in case of noncompliance with the GDPR. A noncompliant

company could be imposed penalties of up to 20 million Euro or up to four percent of the

company’s general annual revenue. Moreover, if companies do not comply with the

appropriate law or the law on the protection of data in their contracts, they may endure

contractual sanctions, early ending of contracts, compensations for damages and the

deprivation of their rights. (European General Data Protection Regulation GDPR 2019)

Under GDPR, companies in the healthcare industry are subject to further regulations due

to their processing of particular private information such as biometrical and genetical

data and data related to health.11 (Tischhauser, Georgieva 2019) In addition to the

challenges that FADP and the GDPR could create, a new legislation called ePrivacy

Regulation (ePR) that aims to complete the GDPR is still under discussion in Europe and

is expected to come into force within a few years. This regulation is similar to the GDPR

but is centered on the protection of e-communications and has a wider range since it is

also protecting legal entities. It has the same fines as the GDPR in case of

noncompliance and is expected to impact the digital plans of action for companies since

it adds another level of provisions to the GDPR. As with the GDPR, the ePR is expected

to apply to Swiss corporations. (Tischhauser, Georgieva 2019)

Further legal limits are imposed to basic health insurances by the “Office Fédéral de la

Santé”. (OFSP 2020) Complementary health insurances on their side are also

supervised and regulated by the Swiss Financial Market Supervisory Authority. (FINMA

2020)

It is hard to spend any time online these days without being presented with a lengthy

user agreement at some point. But do we actually read through them before agreeing to

the terms? Current privacy laws require that users get informed about what information

is being collected and for what purpose, and that we then give consent, which is why we

are bombarded with such requests. If the company then wants to share the data it

collects, it uses anonymization – the stripping out of any personal details to preserve the

privacy of the individuals – before publishing the data. The privacy laws prevent




companies from realizing secondary uses of data. Imagine that a company has collected

user data and later discovers a new and valuable use for it. Under the current legal

regulations, the company would need to seek approval from every user before adopting

the data for this new purpose. While the intent of the legislation is sensible, its application

in the Big data world may greatly hinder the benefits that could be realized. Additionally,

the greater details of big data may allow users to be reidentified from anonymized data,

potentially revealing sensitive information in the process. The current tools, either legal

or technical, are already proving to be ineffective, and as we move further down the road,

they may become obsolete. (Mayer-Schönberger, Cukier 2013)

2.1.4 Ethical limits

Technological advances are strongly shifting the function of data in the insurance

industry and the general role of insurance itself. They are promising to bring major socio-

economic advantages. However, they are also bringing challenges regarding

confidentiality, privacy of personal information, protection of data, and individualization

of insurance. (Keller, Eling, Schmeiser, Christen, Loi 2018)

Intelligent systems are omnipresent. Concerns about the ethical implications in the

privacy and data protection areas are, therefore, increasingly flourishing among analysts,

policymakers, and the society at large. Most of the concerns refer to the capacity or the

lack of ability of intelligent machines to make ethical decisions by nature. Ethical debates

relate to the assessment of what is good and bad, acceptable, or right and wrong, it also

relates to lawfulness, fairness, integrity, and social accountability. Ethical interpretations

are therefore defined as having the capacity to determine, evaluate, and establish ethical

assertions based on various viewpoints. Due to their rising competencies, autonomy,

and communication abilities, intelligent systems are increasingly expected to act as

moral agents, meaning that they can judge what is right and wrong and can be held

responsible for their performances. When interacting with intelligent systems, users

expect them to have identical obligations and responsibilities as when interacting with

human beings. This brings up concerns relating to accountability and questions the ability

of AI systems to perform in accordance with human principles and to adhere to human

rights. (Dignum 2019)

As mentioned earlier, AI works only if provided with large amounts of data. When

applying AI, insurance companies, therefore, use the personal data of their customers.

This may lead to concerns related to confidentiality and protection of data. The first

issues linked to these concerns relate to fairness and discrimination. Fairness can be


compromised in insurance due to the profiling of the insured and cause discrimination.

Here, the term discrimination refers to a situation in which the insured that should be

treated equally, are not. It also refers to a situation in which the insured that should have

a different treatment are treated the same. To avoid such discrimination, specific

characteristics such as gender, ethnic group, sexual preferences, etc. should be

disregarded. In the insurance context, the meaning of discrimination is an underlying

dilemma. When the insured may receive a treatment based on its personal risk, this may

result in disadvantages for protected groups, when having a greater risk than the

average. Contrarily, if the insured are not treated based on their personal risk, the

classification of risk may be seen as unfair since it treats all the insured equally, even if

their risk is not comparable or irrelevant as the personal factors are not taken into

consideration. As the World Health Organization states, health is a fundamental right.

(WHO 2017) This dilemma cannot be resolved just by excluding discriminatory data

characteristics such as gender or ethnic group etc., because any intelligent system would

quickly gather those characteristics from other sources such as online media, that freely

permits access to personal information. A further issue linked to the concerns about

privacy and data protection is about intrusiveness into the right of informational

independence. Privacy is generally seen as a fundamental right such as in Article 8 of

the European Convention on Human Rights. (Council of Europe 2013) Insurance

companies that massively process data and use computerized decision-making may be

intrusive in the right of self-determination of persons. The usual corporate models that

are based on the monitoring of computer science, reward various behaviors that are

considered good and penalize those considered bad according to the insurance

corporation. They, therefore, can decrease risks by provoking a change in behaviors or

choices in ways of living of individuals. However, they could be seen as intrusive and

impeding an independent decision-taking of persons. This may cause issues if persons

cannot bear the cost of insurance-coverage for high-risk profiles and are, therefore,

limited in their life choices. (Keller, Eling, Schmeiser, Christen, Loi 2018) Manipulation of

behaviors using personal information is a further ethical issue. The important amount of

exchanges and the profound knowledge about persons provided by AI and Big data

results in having the insured exposed to misleading and behavioral manipulations.

(Müller 2020)

Other ethical concerns are arising such as the individualization of insurance. With the

shifting function of data, the premium of the customers is now established based on their

individual risk profile and no longer upon their classification in a particular risk group.

This shift is raising different kinds of concerns. One of them is that high-risk individuals


might be rejected from insurances or may have to pay unaffordable premiums. Another

concern is related to the fundamental principle of insurance which is solidarity. This

principle may be deteriorated due to an increasing individualization of insurance that

might also jeopardize the main objective of an insurance which is risk pooling. (Keller,

Eling, Schmeiser, Christen, Loi 2018)

Last January, digitalswitzerland, a Swiss initiative and association that aims to enhance

the position of Switzerland as leader in innovation, launched the Swiss Digital Initiative

(SDI) in Davos. With the fast evolution of computer science and intelligent data-based

systems, the objective of this initiative is to assist the development and application of

guidelines, facilities, and procedures for the implementation of ethical principles for

companies on a worldwide scale. (DFF 2020) In Switzerland there are no legally binding

ethical limits to be respected. However, for the good of their label, companies that use

private data from their customers may maintain and promote a good reputation by

respecting an ethical code of conduct provided by external experts when processing

data, as for example at La Mobilière with the Ethical Code for Data-Based Business by

the Swiss Alliance for Data-Intensive Services. 12 (Ethics - Innovation - data-service-

alliance.ch 2020)

12 Interview with José Iria, Head of Digital Transformation at La Mobilière, Geneva, 08 July 2020

(cf appendix 2)


3. Analysis

3.1 Methodology

This research used diverse approaches to gather sufficient data to reach the set

objectives. It used both primary and secondary data.

First, the current and future AI potential in this sector, as well as it’s legal and ethical

limits, have been analysed and compiled based on extensive review of available, second

source literature. Secondary data have been further gathered from literature and

guidelines produced by personal data protection specialists and institutions.

This compilation provided a general understanding of the different current knowledge

around this topic and enabled the structuring of the background that highlights the

existing use of AI in the insurance industry and the advantages, risks, legal and ethical

limits related to the use of such a technology.

In the second part of the research, primary qualitative data has been gathered by means

of a questionnaire, from five managers working in major Swiss insurance groups. These

written interviews included Isabelle Flueckiger, Director New Technologies & Data at the

Geneva Association, José Iria, the Head of Digital Transformation at La Mobilière, the

Regional Business Manager of a large Swiss healthcare organization, an Account

Manager of a Swiss insurance Group and the Head of Investment Management in a

Swiss major health partner group.The answers provided allow to have a snapshot of the

current understanding, knowledge and use of AI in the Swiss insurance industry.

The legal framework has been reviewed and discussed with Carla Morales, a legal

counsel working at Barclays. This meeting allowed confirmation of the completeness of

the information found about the subject. Additionally, several findings of this research

have further been discussed with Stefan Koch, the Innovation Scout Outpost Silicon

Valley and the current Head of Services, Underwriting and Legal Department at La

Mobilière.

3.2 Interviews

To obtain relevant primary data, a survey has been conducted by means of a standard

questionnaire. It has been circulated among major health insurance companies in

Switzerland, to assess current attitude, planned investments, legal and ethical limits. The

written interview has been done and sent by email due to sanitary restrictions during the

current pandemic crisis. The persons interviewed preferred to reply in writing, to avoid


being pushed into questions to which they would not want to, or were not be able to,

reply. Several questions were not answered in the written interviews, either because of

confidentiality or because the question was not applicable.

We may also assume that knowledge and statements from insurance staff about AI has

been limited and probably also simply absent, because development of such systems is

company secret, for various reasons such as:

• Companies are developing such toolboxes secretly because of their sensitivity and the legal aspects and limitations

• They develop them with the aim to reduce staff costs, in marketing, back office, loss adjustment and fraud prevention and detection. But as the technology still is in an exploratory stage, communications about such developments are kept confidential

• Costs of AI introduction are very considerable – currently hidden mostly in the admin and information technology budget lines

As several of the questions are sensitive in terms of business secrecy, it has been

necessary to conduct the research in a way that allowed full protection of the sources

who accepted to contribute. To do so, at the beginning of the questionnaire, the person

interviewed was asked if he/she was willing to have his/her name and the name of his/her

current employer kept confidential or he/she did not mind to have them mentioned. The

answers of those willing to keep confidentiality were anonymised, but a description of

their position and company has been added at the beginning of the interview to enable

the reader to have an idea of where the answers came from.

The written interview has been formed based on the current trends of the use of AI in the

health insurance industry in Switzerland through the analysis of secondary data,

researches, articles, and books. The questionnaire was then split into four different parts

in order simplify the understanding of the answers. The first part is about the integration

of Artificial Intelligence in the health insurance and aims at establishing in which area of

the health insurance AI is being the most used currently and where it is expected to be

increasingly used. The second part is about the advantages and constraints (apart from

legal and ethical limits) of the use of AI in this industry and aims at determining if the use

of such a technology is seen as an advantage from both the insured and the insurer. The

third part of the questionnaire refers to where the company of the interviewee is currently

focusing the introduction and use of AI. The objective of the questions in this part is to

assess in which operations the company currently focuses the use AI, if they are

developing and deploying this technology through external consulting firms or through

inhouse data scientists and what are the benefits and challenges of their current use of


AI. It also attempts to determine what the strategy of the company related to the

introduction of AI in the next 2-5 years is, and what the constraints are that prevent the

company from developing the use of AI even further. The fourth and final part is about

the legal and ethical limits to the use of AI. The objective in this part is to see what the

legal and ethical limits of the use of AI are, according to the interviewee and in the eyes

of the insured. Moreover, the final questions are aimed to identify how the company is

dealing with those limits and if they expect them to rather harden or soften in the coming

years.

3.3 Results

In this part of the analysis, the answers provided by different members of the insurance

industry are described. Each answer is reported in one of the four different parts as in

the questioners found in the appendixes.

Integration of Artificial Intelligence (AI) in the health insurance sector at large

In this first part of the interview, the majority of the interviewees have answered that

nowadays, the area of Swiss health insurance where AI is most often used, is in sales

and marketing and in the detection and prevention of fraud. Most of them expect an

increase of the use of AI in marketing, but also in direct customer interaction (such as

the use of chatbots), in the detection and prevention of fraud and in pricing. Regarding

the question on whether there are differences of trends in the use of AI between smaller

and larger groups, the answers were quite similar. Most of them replied that a

development of the use of AI is more likely to happen in large insurance groups because

the use of this technology requires significant amounts of data and resources which

might be difficult for smaller insurance groups to gather. One answer however, provided

by Ms. Flueckiger, highlighted the fact that the development of AI did not necessarily

depend on the size of the health insurance company, but more on how advanced the

executive management and board of directors are in their familiarity with AI. She has

observed in Switzerland, to the one, small companies with many AI driven processes,

and to the other, leading Swiss health insurance companies that are hardly using AI in

their operations.13


Association, Geneva, 29 April 2020 (cf appendix 1).


Advantages and constraints in the introduction of AI in the health insurance sector

(apart from legal and ethical limits)

The first question in this part is to know whether in the point of view of the interviewees,

the insured consider the integration of AI in the health insurance to their advantage. Here

the answers differed; two persons answered that it depends on the area where AI is

implemented. According to Ms. Flueckiger, the integration of AI is seen as an advantage

only in certain automated areas, such as customer services for instance, in chatbots or

medical advice. In all other areas where personal data is processed, such as risk

classification, pricing or fraud detection, she assumes that the insured are concerned

and fear unfair treatment and discrimination.16 Another answer provided from a

confidential interview states that the insured perceive the use of AI as an advantage only

if it is correctly used by the insurance company. As per this interviewee, it is important

that the client is informed that he is dealing with AI and that he must be provided with

reasons for this use of technology.14 A further answer states that generally, customers

are very skeptical regarding the use of personal data and that they do not fully trust the

company. The person also added that it is still a long way to go, and that the company

must first prove that it can add value through the use of the customer’s data.15 The

second question is whether the health industry view the use of AI by Health Insurance

companies to their advantage. Answers are mainly positive. It is stated that AI can add

value particularly in automation and in the interaction with customers15 and that it has a

great potential for the insurance industry but it also carries some risks.14 According to

Ms. Flueckiger, the health industry mainly does not see the use of AI by health

insurances to their advantage. She mentions that in general, health service providers

such as doctors, consider that health insurance companies do not trust their expertise

and give more trust to their own AI system and that they come up with their own

conclusions based on the data generated by their AI algorithms, regarding the

appropriate treatment to be provided. Moreover, it demands some increased

administrative efforts for health service providers, since they must document each detail

of their decisions, whereas AI algorithms are perceived as a “black-box”. She also states

that life sciences and pharma firms fear discrimination on certain drugs and treatments

and that health insurances might refuse to pay for them and put pressure to find

14 Interview with a Regional Business Manager of a large Swiss healthcare organization, Geneva,

12 April 2020 (cf appendix 3)

15 Interview with an Account Manager of a Swiss insurance Group, Geneva, 30 April 2020 (cf

appendix 4)


alternative treatments and drugs based on their own analysis of the data.16 The third

question refers to the main constraints faced by health insurance companies when using

AI, apart from the legal and ethical limits. Ms. Flueckiger asserts that the main purpose

of an insurance is risk pooling. However, with the introduction of AI, health insurance

companies tend to be willing to have details on the risks of each individual, thus, adapting

individual risk premiums or not insuring high risk patients, which contradicts the main

purpose of insurance. She also added that with the use of AI, individual treatments are

increasingly being scrutinized and, therefore create a debate about who his rightful about

decisions. In her view, the right to take decisions is being removed from health service

providers while they still have the full medical responsibility for patients.16 Two further

short answers, related to customer trust17 and to the quality of data used in the AI

algorithm.18 Another answer stated that health is a very personal topic and that the

customer must not lose confidence because of AI. He should be able to decide for himself

when he wants his personal data to be processed into AI. “The art is to show the

customer where he can benefit from AI without giving him the feeling that the company

wants to save costs or that his data is not safe”19.

Current focus of your company regarding the introduction and use of AI

The first question is to find out on what processes the company of the interviewee, is

currently focusing the use of AI. The most common answers were related to customer

interactions. One person answered that the current main AI focus is on customer services

such as chatbots and recommendations in the company’s applications.19 Another person

answered that in her company, AI is currently used in marketing, risk assessment and

pricing.18 A further answer provided by Mr. Iria states that the current main use of AI at

La Mobilière is in the automation of back office processes, related to contract

management and customer services. He added that some AI components are also being

sprinkled into the development of applications in sales and loss adjustment processes.

This is empowering the company’s sales agents and claims experts. He finally added

that there is a development in the use of AI in some internal actuarial processes, but that



17 Interview with an Account Manager of a Swiss insurance Group, Geneva, 30 April 2020 (cf

appendix 4)

18 Interview with the Head of Investment Management in a Swiss major health partner group,

Geneva, 20 April 2020 (cf appendix 5)




it is still in the early days due to tight regulations.20 The next question intends to establish

whether the interviewee’s company is using consulting firms or inhouse data scientists

for the development and deployment of this technology. One answers that the company

has one internal Data Scientist who is also trying to undertake different projects with

external experts.17 Another answer is that they only use inhouse data scientists18 and

some others use both consulting firms and inhouse data scientists. The following

question intends to determine the benefits of the current use of AI in the company. Mr.

Iria answers that the main benefits are in the speeding up of the execution of tasks

through (semi-) automation, enabling their experts to avoid spending their time on

manual and repetitive tasks and, instead, can focus on applying their expertise. He adds

that it translates into an increased cost effectiveness and an increase in the quality of

services provided to customers.22 Another answer suggests that the company is

benefiting from the use of AI to target marketing.22 A further answer states that the

benefits are mainly resulting in a better understanding of the company’s data and

possibilities.23 The last answer states that the current benefits of using AI are that his

company can efficiently handle redundant customer requests, is now available 7/24, can

generate insights from customer inquiries and can play out tailored content to the

customer.21 The next question is about the main challenges encountered with AI

regarding the company’s processes and staff. A first person answers that one of the

challenges is that the results are still of limited quality and value.17 A further answer is

that the main challenges are accuracy, the promotion of trust regarding AI and its

integration into the existing products and services.21 Finally, Mr. Iria states that like any

change, the introduction of AI into the company’s operations requires change

management. The purpose of the next question is to understand the strategy of the

organization with regards to the introduction and use of AI in the coming 2-5 years. The

first person answered that the company is continuously developing the chat offer,

introducing AI in further applications, and using voice recognition.20 The second person

stated that the company’s strategy is to improve the quality of data.22 A further person

stated that their strategy is to invest into AI technologies and build more resources

internally, in order to be able to implement more cases with added value for the


(cf appendix 2)






customer.23 Finally, Mr. Iria stated that La Mobilière is reshaping governance around

Data and AI, educating their staff on the benefits and risks of using AI, and raising their

technical expertise in AI by hiring Data Scientists. The goal is to monitor the new AI

trends in the insurance industry, and make AI a part of their software development

process.24 The final question relates to the business constraints that may prevent the

company from developing the use of AI even further. The general answer to that question

is that the main constraints are legal compliance, the data protection regulations, data

quality, transparency, and the customer perception regarding the use of this technology.

Mr. Iria states that he believes it is currently not cost effective to use AI in some

processes, due to current limitations of AI when applied to specific domains. He adds

that for La Mobilière, being personal and close to the clients is a fundamental part of the

way they do business, therefore, in some key processes AI will never replace humans

(by choice).23

Legal and ethical limits to the use of AI

The first question in this part is about the legal limitations from the perspective of the

insured. As reported by Ms. Flueckiger, the first limit is transparency. For instance, how

can an insured person reject certain AI-based decisions and who is legally liable for it.

She added that further limits are data privacy protection and prohibited discrimination.

Regarding discrimination she questions how it could be proven, since AI is a “black box”,

and what would be enough evidence for the insured to reject potential discrimination.25

The next question relates to the legal limits of the health insurance industry. Ms.

Flueckiger replies that the health insurance industry would like to apply AI as much as

possible but that they are confronted by various limitations. The first one is the difficulty

to explain AI and the need to acknowledge it to regulators and customers. She adds that

other limits are the high cost of resources, people and infrastructure and that companies

do not have sufficient knowledge and experience and are, therefore, dependent on

external consultants. Another limit in her view is the data itself, missing data

management and data governance. She states that often, data is not available in the

quality needed, or it is not permitted to use it (e.g. data of genomic tests). Additionally,

the market share in Switzerland is so small that not enough data is available resulting to

23 Interview with an Account Manager of a Swiss insurance Group, Geneva, 30 April 2020

(cf appendix 4)


(cf appendix 2)




fragmented data and data systems. For instance, only about 20-25% of the data in a

company can be accessed within a useful timeframe and on a regular basis.24 Another

answer, states that the regulations from the Federal Office of Public Health (FOPH) and

the Swiss independent supervisory authority (FINMA) imposes guidelines that are

sometimes stricter than the law on data protection.26 The last answer relating to this

question is provided by a colleague of Mr. Iria. He states that he has no knowledge on

whether there are specific regulations related to the use of AI. However, since AI

generates and processes data, the activity should be compliant with the Federal Act on

Data Protection (FADP) and the General Data Protection Regulation (GDPR). He added

that the use of sensitive data such health, race, religious and political affiliation, is legally

restricted. If they want to process this kind of data, they need to obtain the prior consent

of the insured and must inform him about how the data will be used and why. However,

if the data is anonymized, there are no legal restrictions.27 The next question relates to

the current ethical limits linked to the use of AI. The first answer provided by Mr. Iria’s

colleague, mentions that La Mobilière is applying an ethical code of conduct regarding

the processing of data. This is based on the principles of a group of experts from the

Swiss Alliance for Data-Intensive Services.27 (Ethics - Innovation - data-service-

alliance.ch 2020) Another answer provided by Ms. Flueckiger states that the ethical

limitations are discrimination, unfairness, inexplicability and the use of AI to adverse the

patients and customers.28 A further answer states that customers are not confident about

the use of their data. They fear it might be misused and that they would then be provided

with an insurance that does not correspond to their needs. Additionally, they worry that

it contributes to growing a two-tier society, in which healthy and sick individuals are

differentiated, thus, getting away from the social principle of a health insurance.26 The

next question relates to how the company deals with the legal and ethical limits of AI.

One person answers that in the company, the legal and external advisors are included

from the start in the projects and developments.29 Another person states that in his

company, new scenarios are discussed and examined to see if they comply with

regulations and if the company is ready to take a certain risk, to get a step further into


(cf appendix 4)


(cf appendix 2)






the future.29 Finally, Ms. Flueckiger says that besides the report, by the Geneva

Association about responsible AI (published in January 2020), they will be doing some

research in 2021 about Ethical Insurance Conduct when using new technologies.28 The

final question intends to determine if the person is expecting the current legal and ethical

limitations to rather harden or soften in the coming years. The answers differ

considerably. Ms. Flueckiger states that before Covid-19, she was expecting the

limitations to harden, and more limitations put in place by the regulators. But now, in the

aftermath of the Covid-19 shock, she rather thinks that the limitations will soften as

tracking and AI based diagnostics are becoming standard and even imposed by

governments, pushing for the use of more data and AI technologies.28 Another person

replies that the European General Data Protection Regulation (GDPR) is definitely the

first sign that the legal side will harden. However, the Covid-19 crisis will increase the

customers awareness about the positive aspects of sharing data and thus, potentially

softening the legal and ethical limitations.30 Another interviewee, states that the

limitations will stay the same in the coming years.31


(cf appendix 4)




4. Discussion

The various answers to this questionnaire are a snapshot of the current knowledge,

trends and development of AI in Swiss major insurance and health insurance companies.

These are only answers from a few major insurance companies; the overall result cannot

be considered as fully representative for the entire sector. However, they indicate that

the use of AI is still at an early stage. Whether it is about the introduction of this

technology into the company’s processes or the knowledge about its potential and risks,

we can notice in the diversity of the answers, that the understanding of this technology

and its potential need to be further developed.

One important point that was identified comparing different interviews, is the difference

between the basic health insurance and the complementary health insurance. One

person states that the use of AI in the basic insurance is very limited or not possible.

Data linkage between basic and supplementary insurance is not allowed and the link to

medical data is not possible.32 Moreover, as Ms. Flueckiger clearly explains, the basic

health insurance is defined by the governmental Office Federal de la Santé Publique

(OFSP), where the key driver to get new customers in only the price. In this insurance

coverage, AI is not used in the marketing processes. The other point is the

complementary health insurance that is offered by private insurance companies. In this

business area, AI is heavily used for customer segmentation, customer targeting,

customer lifetime value calculations and marketing.33 Another important point highlighted

by Ms. Flueckiger, which confirms the findings in the background, is the negative effect

that AI has on one of the basic principles of insurance. Indeed, one of the primary

principles of a health insurance is risk pooling, implying that the combination of the

healthcare costs of all the insured in a specific segment allows to compute premiums

and to compensate higher or lower average costs of the healthier and less healthy

individuals within the same pool. (Williams 2017) But with the enhanced risk profile

provided by AI for each individual, insurances may adapt the risk premium of each

individual, or even decide not to insure a person, based on the level of risk. This is in

contradiction with the principle of solidarity and it renders risk pooling very difficult.






By comparing the different answers, we can notice that each person has a different point

of view and knowledge on AI, its current and future use and, its legal and ethical limits.

As seen in the interview with the Account Manager of a Swiss insurance Group, rules

and regulations are expected to be further refined, notably due to the revision of the

GDPR and the new ePrivacy regulation (ePR) and additionally, due to the revision of the

Swiss regulations such as the FADP, that have recently been adapted to the GDPR.

However, as seen in the answers of two different interviews, with the current Covid-19

pandemic, governments are pushing for an increased use of digital technologies such

as AI to contain and track as much as possible the virus. For health and economical

purposes, this may be of advantage for health insurances; however, the regulations will

not change. Health insurances companies must, therefore, be very cautious regarding

their compliance with regulations when using data and digital technologies.


5. Recommendations for health insurance companies & conclusion

In the coming years, Artificial Intelligence is expected to be much more autonomous

and will have an increasingly important role in the decision-making process of more

and more sectors. (Keller 2020)

The deep changes being brought through the introduction of AI in various processes

will obviously result into an increasing number of stricter rules and regulations.

Moreover, confusion and misconception about Artificial Intelligence truly could prevent

such technologies to develop and benefit the overall society. (Tabate 2016)

The overall findings and information gathered in this thesis allows the listing of the

following recommendations for Swiss health insurance companies:

The development and integration of these new technologies is unavoidable

Big data has been available for many years already and is not a temporary phenomenon.

It is not likely to disappear naturally or to be stopped by legal or technical intervention.

To the contrary: big data will be increasingly available and the capacity to create

information by developing new algorithms will grow at a fast pace, or even exponentially.

Considerable investment of internal and external expertise and financial resources into

such technology by health insurance companies appears unavoidable, in order to remain

competitive in the market.

Adapt governance of the company

In the past, insurance companies were steering and overseeing their IT developments

through committees of mostly technical experts. Future expert groups which pilot

development of big data and AI require to be enhanced with crucial, additional expertise,

such as legal and compliance experts, data protection officers, ethical experts, social

and marketing specialists, and others. Such steering committees must be integrated and

become part of the overall risk management framework of the company.

Customer focus

AI and use of big data should be further developed, with the primary aim to offer a best

possible service to the insured customer. Customers therefore need to be sensitised to

their advantages, such as reduction of administrative costs, an increase of efficiency in

loss adjustment, the prevention of fraud which ensures fair and equal treatment of all


insured, premium savings, more convenience, transparency and reliability in the

customer’s dealings with the insurance company.

Expand knowledge about AI to all staff

Management staff of health insurance companies may not have enough knowledge

about AI, to understand how it could be implemented into their processes to create value,

and its legal and ethical limitations. Those health insurance companies which develop AI

in their systems should therefore invest in staff training in parallel, in order to increase

awareness and knowledge about and support for this technology within the organisation.

This will then facilitate the understanding of the potential of the use of AI that is currently

still developing in the Swiss health insurances.

External audit and certification

Insurance companies are currently obliged to publish their legal code on use and safety

of personal data, on their Websites. Based on the findings, and in view of the complexity

and technical sophistication, it is advisable that health insurers periodically task third

party experts to audit their use and safety of data and certify compliance with the general

legislation and their own legal code. Companies may also develop, publish, and get

audited on an ethical code of conduct regarding the use and protection of data. This is

not compulsory but promoting transparency and compliance with legal and ethical norms

is essential in creating and maintaining a solid, trusted reputation.

In conclusion, the integration of a technology such as AI will have a great potential for

Swiss health insurances. It can benefit the overall society through various advantages,

but still also carries some risks. It will, therefore, require significant efforts, prerequisites,

and investments from both sides, the insurance companies, and the insured. To be

implemented and further developed, insurance companies should gain trust among

customers through the promotion of their responsible use of data and AI. They should

pay particular attention to their reputation by being transparent regarding the processing

of customers’ personal data and by promoting compliance to the legal regulations and to

ethical codes of conducts. Additionally, Swiss health insurance companies should raise

awareness among customers about this technology, by showing them the advantages

of its use, that they will remain fair, they won’t discriminate and they won’t use personal

data to adversely impact the customers.


Bibliography

CC 101 Federal Constitution of 18 April 1999 of the Swiss Confederation, 2020. [online]. [Viewed 14 August 2020]. Available from: https://www.admin.ch/opc/en/classified-compilation/19995395/index.html

CC 235.1 Federal Act of 19 June 1992 on Data Protection (FADP), 2019. [online]. [Viewed 16 July 2020]. Available from: https://www.admin.ch/opc/en/classified-compilation/19920153/index.html

CC 235.11 Ordinance of 14 June 1993 to the Federal Act on Data Protection (OFADP), 2012. [online]. [Viewed 16 July 2020]. Available from: https://www.admin.ch/opc/en/classified-compilation/19930159/index.html

CC 235.13 Ordinance of 28 September 2007 on Data Protection Certification (DPCO), 2016. [online]. [Viewed 16 July 2020]. Available from: https://www.admin.ch/opc/en/classified-compilation/20071826/index.html

COMMISSIONER (FDPIC), Federal Data Protection and Information, 2015a. Data Protection - Switzerland. [online]. 2015. [Viewed 16 July 2020]. Available from: https://www.edoeb.admin.ch/edoeb/en/home/der-edoeb/links/datenschutz---schweiz.html

COMMISSIONER (FDPIC), Federal Data Protection and Information, 2015b. Task. [online]. 2015. [Viewed 16 July 2020]. Available from: https://www.edoeb.admin.ch/edoeb/en/home/der-edoeb/auftrag.html

COPELAND, B.J., 2019. Artificial intelligence | Definition, Examples, and Applications. Encyclopedia Britannica [online]. 19 November 2019. [Viewed 10 December 2019]. Available from: https://www.britannica.com/technology/artificial-intelligence

COUNCIL OF EUROPE, 2013. European Convention on Human Rights. COE [online]. 2013. [Viewed 11 August 2020]. Available from: https://edoc.coe.int/fr/convention-europenne-des-droits-de-l-homme/5579-european-convention-on-human-rights.html

DFF, Département fédéral des finances, 2020. Swiss Digital Initiative. [online]. 2020. [Viewed 11 August 2020]. Available from: https://www.efd.admin.ch/efd/fr/home/themen/Digitalisierung/swiss-digital-initiative.html

DIGNUM, Virginia, 2019. Responsible Artificial Intelligence: How to Develop and Use AI in a Responsible Way [online]. Cham: Springer International Publishing. [Viewed 24 June 2020]. Artificial Intelligence: Foundations, Theory, and Algorithms. ISBN 978-3-030-30370-9. Available from: http://link.springer.com/10.1007/978-3-030-30371-6


Ethics - Innovation - data-service-alliance.ch, 2020. [online]. [Viewed 29 July 2020]. Available from: https://data-service-alliance.ch/innovation/ethics

European General Data Protection Regulation GDPR, 2019. S-GE [online]. [Viewed 24 July 2020]. Available from: https://www.s-ge.com/en/european-data-protection-regulation

FINMA, Eidgenössische Finanzmarktaufsicht, 2020. Supervision in the insurance sector. Eidgenössische Finanzmarktaufsicht FINMA [online]. 2020. [Viewed 14 August 2020]. Available from: https://www.finma.ch/en/supervision/insurers/

GDPR ART.5, 2019. Art. 5 GDPR – Principles relating to processing of personal data. General Data Protection Regulation (GDPR) [online]. 2019. [Viewed 15 August 2020]. Available from: https://gdpr-info.eu/art-5-gdpr/

GDPR ART.15, 2019. Art. 15 GDPR – Right of access by the data subject. General Data Protection Regulation (GDPR) [online]. 2019. [Viewed 14 August 2020]. Available from: https://gdpr-info.eu/art-15-gdpr/

GORDON, Clara-Ann and GUROVITS, András, 2020. AI, Machine Learning & Big Data Laws and Regulations | Switzerland | GLI. GLI - Global Legal InsightsInternational legal business solutions [online]. 2020. [Viewed 13 July 2020]. Available from: https://www.globallegalinsights.com/practice-areas/ai-machine-learning-and-big-data-laws-and-regulations/switzerland United Kingdom

HARARI, Yuval Noah, 2018. 21 Lessons for the 21st Century. 1st ed. London: Jonathan Cape/libri. ISBN Trade paperback 9781787330870.

INRIX, 2020. About. Inrix [online]. 2020. [Viewed 18 August 2020]. Available from: https://inrix.com/about/

KELLER, Benno, ELING, Martin, SCHMEISER, Hato, CHRISTEN, Markus and LOI, Michele, 2018. Big Data and Insurance: Implications for Innovation, Competition and Privacy. [online]. March 2018. P. 48. Available from: https://www.genevaassociation.org/sites/default/files/research-topics-document-type/pdf_public/big_data_and_insurance_-_implications_for_innovation_competition_and_privacy.pdf

KELLER, Benno, 2020. Promoting Responsible Artificial Intelligence in Insurance. Artificial Intelligence [online]. 2020. P. 24. Available from: https://www.genevaassociation.org/sites/default/files/research-topics-document-type/pdf_public/ai_in_insurance_web_0.pdf


KRAFT-BUCHMAN, Caitlin and ARIAN, Renée, 2019. Deadly Data Gap: Gender & Data. Deadly Data Gap: Gender & Data [online]. 21 November 2019. [Viewed 17 August 2020]. Available from: https://www.womenatthetable.net/blog/deadly-data-gap-gender-data

LPGA, 2019. RS 830.1 Loi fédérale du 6 octobre 2000 sur la partie générale du droit des assurances sociales (LPGA). [online]. 1 October 2019. [Viewed 14 August 2020]. Available from: https://www.admin.ch/opc/fr/classified-compilation/20002163/index.html

MAYER-SCHÖNBERGER, Viktor and CUKIER, Kenneth, 2013. Big Data: A Revolution That Will Transform How We Live, Work, and Think. Reprint edition. Eamon Dolan/Houghton Mifflin Harcourt.

MICHAEL, 2019. How Artificial Intelligence Is Improving Insurance. [online]. 22 September 2019. [Viewed 8 November 2019]. Available from: https://hackernoon.com/how-artificial-intelligence-is-revolutionizing-insurance-9n5e230zp

MÜLLER, Vincent C., 2020. Ethics of Artificial Intelligence and Robotics. In: The Stanford Encyclopedia of Philosophy [online]. Fall 2020. Metaphysics Research Lab, Stanford University. [Viewed 7 July 2020]. Available from: https://plato.stanford.edu/archives/fall2020/entries/ethics-ai/

NAKONZ, Jonas, 2019. AI “principles” need refinement. digitalswitzerland [online]. 14 June 2019. [Viewed 8 November 2019]. Available from: https://digitalswitzerland.com/2019/06/14/ai-principles-need-refinement/

OFSP, Office fédéral de la santé publique, 2020. Assurance-maladie : L’essentiel en bref. [online]. 2020. [Viewed 14 August 2020]. Available from: https://www.bag.admin.ch/bag/fr/home/versicherungen/krankenversicherung/krankenversicherung-das-wichtigste-in-kuerze.html

PANCHAL, Shubham, 2019. Types of Artificial Intelligence and examples. Medium [online]. 1 November 2019. [Viewed 10 December 2019]. Available from: https://medium.com/predict/types-of-artificial-intelligence-and-examples-4f586489c5de

PRICEWATERHOUSECOOPERS, 2020. New Swiss Federal Act on Data Protection: 5 changes to remember. PwC [online]. 2020. [Viewed 19 July 2020]. Available from: https://www.pwc.ch/en/insights/fs/data-protection-switzerland.html

SCHMIDT, Christian, 2018. Insurance in the Digital Age. [online]. September 2018. P. 20. Available from: https://www.genevaassociation.org/sites/default/files/research-topics-document-type/pdf_public/insurance_in_the_digital_age_01.pdf


SCHWARTZ, Oscar, 2019. In 2016, Microsoft’s Racist Chatbot Revealed the Dangers of Online Conversation - IEEE Spectrum. IEEE Spectrum: Technology, Engineering, and Science News [online]. 2019. [Viewed 4 May 2020]. Available from: https://spectrum.ieee.org/tech-talk/artificial-intelligence/machine-learning/in-2016-microsofts-racist-chatbot-revealed-the-dangers-of-online-conversation

SEFRI, Secrétariat d’État à la formation, à la recherche et à l’innovation, 2019. Intelligence artificielle. [online]. 13 December 2019. [Viewed 10 August 2020]. Available from: https://www.sbfi.admin.ch/sbfi/fr/home/bfi-politik/bfi-2021-2024/transversale-themen/digitalisierung-bfi/kuenstliche-intelligenz.html

SHAW, Jonathan, 2018. Artificial Intelligence and Ethics. Harvard Magazine [online]. 6 December 2018. [Viewed 8 November 2019]. Available from: https://harvardmagazine.com/2019/01/artificial-intelligence-limitations

SHROFF, Raj, 2019. How Are Insurance Companies Implementing Artificial Intelligence (AI)? Medium [online]. 5 October 2019. [Viewed 10 December 2019]. Available from: https://towardsdatascience.com/how-are-insurance-companies-implementing-artificial-intelligence-ai-aaf845fce6a7

SWISSMEDICALNETWORK, 2020. Swiss Medical Network - Data protection. [online]. 2020. [Viewed 14 August 2020]. Available from: https://www.swissmedical.net/en/data-protection/

TABATE, Tom, 2016. History | One Hundred Year Study on Artificial Intelligence (AI100). [online]. 2016. [Viewed 10 August 2020]. Available from: https://ai100.stanford.edu/history-1

THESWISSAUTHORITIESONLINE, 2020. Health insurance in Switzerland: basic insurance, health insurance companies, supplementary insurance, coverage - www.ch.ch. [online]. 2020. [Viewed 18 August 2020]. Available from: https://www.ch.ch/en/health-insurance/

TISCHHAUSER, Stephanie and GEORGIEVA, Evelina, 2019. GDPR, Swiss DPA & ePrivacy – what Swiss companies should know. Pryv [online]. 20 November 2019. [Viewed 26 July 2020]. Available from: https://www.pryv.com/2019/11/20/gdpr-swiss-dpa-e-privacy/

WELLENS, Patrick, 2020. Ethics and Compliance Switzerland » Revision of the Swiss Federal Act on Data Protection: Lawmakers establish first cornerstones of the new Data Protection Act. [online]. 27 February 2020. [Viewed 19 July 2020]. Available from: https://www.ethics-compliance.ch/2020/02/27/revision-of-the-swiss-federal-act-on-data-protection-lawmakers-establish-first-cornerstones-of-the-new-data-protection-act/


WHO, 2017. WHO | Health is a fundamental human right. WHO [online]. 10 December 2017. [Viewed 21 August 2020]. Available from: http://www.who.int/mediacentre/news/statements/fundamental-human-right/en/

WILLIAMS, Joeff, 2017. Risk Pooling: How Health Insurance in the Individual Market Works | American Academy of Actuaries. [online]. 2017. [Viewed 30 July 2020]. Available from: https://www.actuary.org/content/risk-pooling-how-health-insurance-individual-market-works-0


Appendix 1: Interview with Isabelle Flueckiger

(Director New Technologies & Data at the Geneva Association)

Questionnaire on the use of Artificial Intelligence (AI) by Swiss health insurance companies

I ensure that this interview will be conducted in a way that allows full protection of the sources.

O I want my name and the name of my current employer to be kept

confidential

● I do not mind if my name and the name of my current employer is

mentioned in the thesis

1. Integration of Artificial Intelligence (AI) in the health insurance

sector at large

1.1 In which areas of Swiss health insurance is AI most often used, and

how is it used?

▪ Marketing

One has to differentiate between the basic health insurance (KVG,

en français: Loi fédérale sur l'assurance-maladie (LAMal)), and

supplemental health insurance according VVG (en français: Loi sur le

contrat d’assurance, LCA)

In KVG, the insurance benefit is defined by the government / BAG

(en français: office federal de la santé OFSP), and the key driver for

getting new customers is only the price. There typically, no AI is used

in marketing.

It is different in the supplemental health insurance that can be offered

by health insurance companies as well as private insurance

companies. On one side, several different insurance benefit contracts

exists, and one has a free market. There AI is heavily used for

customer segmentation, customer targeting, customer lifetime value

calculations.

▪ Risk assessment & pricing

Most products in the Swiss market are still based on and are

assessed by classical actuarial procedures. There are few direct

insurer that use AI, but still struggle with the lack of transparency and

explainability and therefore, they are very cautious to apply that only

on a stand-alone basis.


▪ Loss adjustment

See also others. As part of the automated background processing of

bills, AI is used. When it comes to loss adjustments, they are either

made by an expert, or if there are serious and complex cases, even

with the help of layers. The AI is used for either: no large deviations

from the expectations, and thus payment, or deviations and then an

expert has to do it. The reason behind that are again regulatory

requirements as well as it need to be bullet-proved because

otherwise there will be a lawsuit.

▪ Detection and prevention of fraud

See also other.

AI is used in the automated background checking of bills for finding

inconsistencies, as well as for looking patterns over certain time

periods, taking into account bills from the same insured person, or

bills from the same health benefit provider e.g. if certain benefits are

billed twice on different bills and so on.

▪ Product development

This is only valid for the supplemental health insurance. Currently, still the classical actuarial product development methods

are used and not necessarily AI. The reason is that is more difficult to

find a product type that people will buy because they see a need for it

than finding individual demand by AI.

▪ Other?

In the health insurance industry, one has automated background

processes for checking the incoming bills on consistency (“shadow

processing”). If no inconsistency is found, the bill is just paid. Only

bills that seems to be inconsistent e.g. strange SwissDRG or Tarmed

codes, or quantities, will then go to an expert, sometimes to medical

experts, where the bill is reappraised. This is one of the most

important process in a health insurance company and thus, most

automated and with most sophisticated AI. I.e. 100% of the incoming

bills are assessed in that way.

Personalization of the insurance pricing based on wearable data e.g.

fitbits, fitness trackers, genetic tests and so on. There is a lot of

ethical and data privacy concerns but the trend is there to use these

data with AI for personalized insurance and finally, the determination

of the premium.


1.2 In which of the above areas do you expect an increase of use of AI?

Marketing, pricing, fraud incl. automated background checking, and in

pricing based on wearable data e.g. fitbits and so on.

1.3 Do you expect a difference of these trends between smaller and

larger insurance groups?

It is not necessarily dependent of the size of the health insurance

company, or insurance company.

It depends how advanced the executive management and board of

directors are in the familiarity with AI. I have seen very small

companies with a lot of AI driven processes, and leading Swiss health

insurance companies, that are doing nearly nothing with AI.

The ones that implement have different approaches. Larger ones

build it up in-house, whereas smaller ones have standardized

products for AIaaS (AI as a service).

Direct insurers that have a supplemental health insurance line of

business, are typically leveraging the knowhow of the other business

and building it up by themselves.

2. Advantages and constraints in the introduction of AI in the health

insurance sector (apart from legal and ethical limits)

2.1 Do the insured view the integration of AI to their advantage?

It depends on the area where it is integrated.

If it is integrated in customer services e.g. chatbots, medical advice,

then it is considered as an advantage. But in all other areas, if data

are used for risk classification, pricing, fraud detection, the insured

have concerns and are expecting unfairness and discrimination.

2.2 Does the health industry view the use of AI by Health Insurance

companies to their advantage?

Mainly not. Hospitals or benefit providers (e.g. doctors) in general have

concerns because of the analyses and results based on AI and that

there are much more rejections for paying bills on the health insurer

side. Or because health insurer come up with their own conclusion of

what could be the best treatment based on the data and generated by

AI algorithms. Benefit providers say that the health insurance

companies do in general not trust to them and trust more to their own

AI than to the experts.


All of this also leads to more administration effort of the benefit provider

because they have to document each detail of their decision, whereas

the AI algorithms are a black-box.

Life sciences and pharma firms fear that certain treatments and drugs

will be replaced by the health insurance companies because they will

not pay them and put pressure to alternative treatments. They fear a

discrimination of certain drug and treatments.

2.3 What are the main constraints which the health insurance industry

faces when using Artificial Intelligence apart from the legal and ethical

limits?

Medical treatment and advice is not black nor white and individual

treatments become much more scrutinized. And if there is a deviation

from the average there is a debate who is now right with the decision,

AI on the health insurance side or a doctor. There is more decision-

rights taken away from the benefit providers but they still have the full

medical responsibility for a patient.

Further, the purpose of insurance is risk pooling, but health insurance

companies wants to know the detailed individual risk and thus, details

individual risk premium or not insure the patient at all which is in

contradiction of the purpose of insurance.

3. Current focus of your company regarding the introduction and use

of AI

3.1 If any, on what processes is your company currently focusing AI

technology?

We are a global think tank, and we are doing research on the topic

but not implementing it.

(At the former company, a global consulting firm, we implemented it in

internal processes, like HR, IT support, and so on.)

3.2 Is your company using consulting firms or inhouse data scientists for

the development and deployment of this technology?

N/A – we are a think tank for the insurance industry

3.3 What are the benefits of the use of AI currently?


3.4 What are the main challenges you encounter with AI regarding your

processes and staff?



3.5 What strategy does your company have with regards to the

introduction and use of AI in the coming 2-5 years?


3.6 What are the business constraints that may prevent your company

from developing even further the use of AI?


4. Legal and ethical limits to the use of AI

4.1 What are the legal limitations in the use of AI?

4.1.1 From the perspective of the insured

Transparency, e.g. how can an insured person veto against

certain AI-based decision? Who is responsible?

Data privacy protection, and not permitted discrimination, but

how to prove it, if AI is a black box? And what are enough

evidence that an insured person can veto against potential

discrimination?

4.1.2 From the perspective of the health insurance industry

The health insurance industry would like to apply AI as much

as possible. Limitations are:

• Non-explainability of AI, but needed to explain it towards

the regulators, customers, and so on

• Talents, skills and capacities to build the knowledge and

tools to apply it in a proper way – Google, IBM, or banks

are more attractive employers with higher salaries

• High costs of resources, people and infrastructure and

not sufficient knowledge and experience (or

dependence on external consultants)

• Data and data input: often data are not available in the

quality needed, or data are not permitted to use e.g.

data of genomic tests, or the market share in

Switzerland is to small that not enough data is available,

or very fragmented data and data systems (typically,

only about 20-25% of the data in a company can be


accessed within a useful timeframe and on a regular

basis (missing data management and data governance)

4.2 What are the ethical limitations linked to the use of AI?

Discrimination, non-fairness, no explainability, use to the adverse of

the patient / customer

4.3 How does your company deal with the legal and ethical limits of using

Artificial Intelligence?


We are researching these topics. Besides the report about

responsible AI (published in January 2020 by the Geneva

Assocation), we will do research in 2021 about Ethical Insurance

Conduct when using new technologies.

4.4 Do you expect the current legal and ethical limitations to rather harden

or soften in the coming years?

Before Covid-19 I was expecting the limitations to harden and more

limitation will be put in place by governments. But now, with Covid-19

and the aftermaths, I rather think it will be soften. Because tracking and

AI based diagnostics are becoming standard and imposed by the

government. Thus, the government pushes in using more data and AI

and softens limitations.


Appendix 2: Interview with José Iria

Head of Digital Transformation at La Mobilière

Questionnaire on the use of Artificial Intelligence (AI) by Swiss health insurance companies


O I want my name and the name of my current employer to be kept

confidential

X I do not mind if my name and the name of my current employer is



sector at large


how is it used?

▪ Marketing


▪ Loss adjustment



▪ Other?


I expect that insurance companies will mainly increase their use of AI

in "Risk assessment & pricing" and "Detection and prevention of fraud",

while benefiting from developments in the use of AI in "Marketing"

across all industries. There will be some use of AI in "Sales and

Distribution" and "Loss Adjustment" as well. Less so in "Product

Development"




No reply.




No reply.



No reply.



limits?

No reply.


of AI


technology?

Currently, our main use of AI is for automating back office processes related to contract management and customer services.

There are also some AI components being sprinkled into the development of applications in sales and loss adjustment processes, empowering our sales agents and claims experts.

Finally, we start seeing the use of AI in some internal actuarial processes, but it is still early days due to tight regulation.




Both


Mainly speeding up the execution of tasks through (semi-)automation, avoiding that our experts spend their time doing manual and repetitive work and can focus on applying their expertise.

This translates into increased cost effectiveness and increase in the quality of our service to the customer.



Like any change, it needs change management. Overall, I would say

that most collaborators are positive about AI.



We are reshaping governance around Data and AI, educating our staff

on the benefits and risks of using AI, raising our technical expertise in

AI by hiring Data Scientists, monitoring the new trends in AI in the

insurance industry, and making AI a part of our software development

process.



Besides the constraints related to the usage of Data (see below), I

believe that it is currently not cost effective to use AI in some

processes, due to present limitations of AI when applied to certain

domains. Moreover, for La Mobilière being personal and close to their

clients is a fundamental part of the way we do business, therefore in

some key processes AI will never replace a human (by choice).



Je n'ai pas connaissance de limitations spécifiques sur l'utilisation de l'IA; cependant bien attendu, comme l'IA utilise et génère des données, l'activité doit respecter les lois suisse LPD (loi sur la protection des


données) et européenne RGPD (règlement sur la protection des données).

Les données pour lesquelles il y a des restrictions légales sont les données sensibles sur la personne (ex: santé, opinions religieuses / politiques, appartenance à une race, …). Si l'on veut travailler avec des données sensibles (récolter, communiquer, exploiter…), il faut obtenir l'accord de la personne et l'informer sur l'utilisation que l'on va en faire. Cependant, si les données sont anonymisées, il n'y a plus de restriction légale.


No reply.


No reply.


La Mobilière applique aussi un code éthique sur l'utilisation des données; rien de particulier sur l'aspect IA; la Mobilière s'appuie sur les principes de ce groupe d'experts:

https://data-service-alliance.ch/innovation/ethics

4.3 How does your company deal with the legal and ethical limits of using

Artificial Intelligence?

Je n'ai pas connaissance de règles particulières à l'IA en particulier ; comme mentionné, c'est l'aspect données qui est soumis à des restrictions.

4.4 Do you expect the current legal and ethical limitations to rather harden or soften in the coming years?

No reply.




Appendix 3: Interview (Confidential)

Interview with a Regional Business Manager in a large Swiss Healthcare Organization providing health and accident insurance

Questionnaire on the use of Artificial Intelligence (AI) by Swiss health

insurance companies


x I want my name and the name of my current employer to be kept

confidential

O I do not mind if my name and the name of my current employer is



sector at large


how is it used?

▪ Marketing


▪ Loss adjustment



▪ Other?

Customer service: Answering or triage of customer enquiries


Marketing



More increase in larger insurance groups. They have more resource

to develop and invest.





If AI is used correctly, it is perceived as an advantage by the insured.

It is important that the client understands that he is dealing with AI

and is given reasons why it is being used.



AI offers a lot of potential for the insurance industry, but also carries

risks. It is important to assess where the use of AI makes sense.



limits?

Client must not lose confidence because of AI. Health is a very

personal topic. Therefore, the customer must be able to decide for

himself if and when he wants to be confronted with AI.

The art is to show him where he can benefit without giving him the

feeling that you want to save costs or that his data might not be safe.


of AI


technology?

Customer service / Chatbot → Recommendations in our applications



Both


Efficient handling of redundant customer requests

7x24 availability

Generate insights from customer enquiries


Playing out content tailored to the customer



Accuracy

Promoting trust in AI

Integration into existing products and services



Continuous development of the chat offer, use of AI in further

applications, use of voice recognition



Data protection Transparent customer



No reply.


No reply.


No reply.


No reply.

4.3 How does your company deal with the legal and ethical limits of using Artificial Intelligence?

No reply.


4.5 Do you expect the current legal and ethical limitations to rather harden

or soften in the coming years?

No reply.



Interview with an Account Manager in a Swiss Group made of the leading health insurers in Switzerland


insurance companies


X I want my name and the name of my current employer to be kept

confidential




sector at large


how is it used?

▪ Marketing


▪ Loss adjustment



▪ Other?

Currently, the main usage is in sales & marketing (optimization of the sales process, personalization of services and interaction). Coming more and more is the usage in process automation (for example self-services like chatbot, automated E-Mail responses) and detection and prevention of fraud as another atomization case.


In the direct customer interaction




The amount of data you need for fraud detection or prevention has to

be quite big, not sure if small insurances will be able to build a model

on that. Smaller insurance groups will probably not invest as much in

AI and do more with simple management of data-based rules, getting

probably already 50-75% of the possibilities out of it.




No, generally, our customers do not really trust us, this is still a long

way to go. So, they view the usage of data very skeptical. We first have

to provide proof that we can add value with using the customer’s data

so they will become more open to sharing their data with us.



Definitely, this is a field that can add value especially in automation and

in the interaction with the customer.



limits?

As mentioned under 2.1 this is trust.


of AI


technology?

Customer interaction



We have one Data Scientist internally and try to do projects in a

tandem with external experts as well.


A better understanding of our data and the possibilities.




Everyone thinks they know what AI is, but in fact they have all different

opinions. Thankfully, we do not encounter a lot of skepticism.



Invest in it and build up more resources internally, to be able to

implement more cases with added value for the customer.



Customer perception and definitely also legal and compliance.




As long as the insured allows us the usage of their data, there are no

limitations.


The BAG and FINMA give us some guidelines, which are sometimes

stricter than the data law.


Customers are not sure if we could misuse their data and then not

provide them with the insurance they need.

Additionally, it gets more into a two-tier society, differentiating more

between healthy and sick, getting away from the social principle of

health insurance, as data makes everything transparent.

4.3 How does your company deal with the legal and ethical limits of

using Artificial Intelligence?

With new cases, they are discussed if they compile with regulations

and if our group as a company is ok with taking a certain risk if

necessary, to get a step further into the future.


4.4 Do you expect the current legal and ethical limitations to rather

harden or soften in the coming years?

GDPR is definitely the first sign that the legal side will harden. However,

crisis like Corona will make customers more aware of the positive

aspects of sharing data, maybe softening their ethical view as well on

the other side.



Interview with a Head of Investment Management in a major health partner group in Switzerland


insurance companies


X I want my name and the name of my current employer to be kept

confidential




sector at large


how is it used?

▪ Marketing


▪ Loss adjustment



▪ Other?


Marketing



Yes





Depending on the area yes



Yes



limits?

Data quality


of AI


technology?

Marketing, Risk assessment & pricing



Inhouse


Target Marketing



Results are still of limited quality and value



Improve data quality




Regulatory, data quality



There are no “perspectives” with respect to legal limitations, but there

is a difference between basic insurance and supplementary

insurance. In basic insurance a lot of fields above are simply not

possible. Data linkage between basic and supplementary is not

allowed. Link to medical data is not possible.


No reply.


No reply.


In the basic insurance, this is addressed by the legal context. In

complementary, the insured should not be in worse situation as if

there was no AI.

4.3 How does your company deal with the legal and ethical limits of

using Artificial Intelligence?

Legal department and external advisors are included from the start in

the projects and developments

4.4 Do you expect the current legal and ethical limitations to rather harden or soften in the coming years?

It will stay the same.

Legal and ethical constraints in the use of Artificial Intelligence ...

Documents