CIA-3

INSURANCE LAW

TITLE- Analysis: Usage of Artificial Intelligence in Insurance Sector.

SUBMITTED BY

NAME: KHUSHAL MITTAL

REGISTER NUMBER: 1650424

CLASS: 7 BBA LLB- A

Introduction

“Whatever is created will be destroyed”, the theory of insurance is based on this particular
principle. Human life is full of danger, uncertainties and risk. The safe remedy against it is
insurance. The thing in itself is not protected by insurance but the financial loss caused by its
destruction. A pool is created through contribution made by persons seeking to protect them
from such financial loss. The premium collected is used by the insurer in case of any uncertainty
is use to pay the loss. It is a vindication against all types of losses.1

The development in mankind and economy has led to widespread sharing of risk. Artificial
Intelligence has become the need for developing economy to cope up with the dynamics. It is a
faster way of performing activities. It is a decision making program which operates in the real
world and make decision more appropriate and accurate. It can be said to be the future of
insurance law in India. It is a complex digital process vis an infinity loop. It will not be possible
to turn back to manual process once the dependence has been on artificial intelligence.2

The complexity of Artificial intelligence is very high. There are functions of the number of
linked process of hardware and software to perform uniformly. The complexity in forming
claims report needs advancement of technology. This will give decision making process to
machine completely and it will be out of reach of human process. The complex artificial
intelligence which is been developed by humans can make life miserable for the humans who are
expected to operate the system. It is true how much better information internet may provide it
cannot compete with the growing informing in the market3

People don’t necessarily mean a particular approach or well define school for learning new ways
to perform a task. It give them ideas to perform business in an more efficient and economical
method. We can foresee in case of insurance business the amount of premium to be charged for a
particular insurance taking into consideration the various elements of risk involved. This
includes different kinds of future events. Artificial intelligence can be said to be an economic
way to customize the insurance coverage as per the requirement of the customer/insured

1
DR. SACHIN RASTOGI, LAW AND PRINCIPLES, 39, FIRST EDITION, 2014.
2
Shoshana Zuboff, In the Age of the Smart Machine
3
8. Anne Knowles, InterAp Assigns Intelligent Agents to the Web, PC W eek, June 12, 1995, at 42, 47 (quoting Bob
Johnson of Dataques
Impact of AI in Insurance:

Artificial intelligence has made changes in every sector considering it construction, banking,
manufacturing, marine. The wonted between them is the risk involve, the growth in the sector
requires a more productive and faster way for performing the task.

 Improving the pace of carrying out work- The use of robotic process to perform the set
of repeated task from operation team by trained and coded AI models lifts the burden
from the team to work in another productive manner.

 Optimising the service- Insurers can provide to the different set of stakeholders based
on their relationship and previous contact more efficient form of service and a lower
quoted rate.

 Analysis- Artificial intelligence is a coded program to perform a particular task. It will

not miss even a minute detail or step while analyzing the insights. It can adjust,
optimize, pricing of premium and services provided and mange risk. The process of
underwriting is reduced to a few seconds as the majority of underwriting is automated
and supported by a combination of machine and deep learning models built within the
technology stack. These models are powered by internal data as well as a broad set of
external data accessed through application programming interfaces and outside data and
analytics providers. Information collected from devices provided by mainline carriers,
reinsurers, product manufacturers, and product distributors is aggregated in a variety of
data repositories and data streams. These information sources enable insurers to make
ex ante decisions regarding underwriting and pricing, enabling proactive outreach with
a bindable quote for a product bundle tailored to the buyer’s risk profile and coverage
needs. Regulators review AI-enabled, machine learning–based models, a task that
requires a transparent method for determining traceability of a score (similar to the
rating factor derivations used today with regression-based coefficients). To verify that
data usage is appropriate for marketing and underwriting, regulators assess a
combination of model inputs. They also develop test policies for providers when
determining rates in online plans to ensure the algorithm results are within approved
bounds. Public policy considerations limit access to certain sensitive and predictive data
 (such as health and genetic information) that would decrease underwriting and pricing
flexibility and increase anti selection risk in some segments.4

 Fundamental changes in daily activity- The growth in the sector needs large amount of
labor hour. The use of trained artificial intelligent bot can replace it from tradition small
scale industry. It can manage large data base in a more faster and efficient way then
humans.

 Reducing Fraud- Frauds arising out of identity theft are common in the insurance
industry. AI has the ability to identify gaps in the process that leave the claims process
vulnerable to fraud. Machine learning algorithms can scan through historical data and
find out if there is anything out of place. Certain patterns can be highlighted, resulting in
alerting against a fraudulent activity. It is not plausible to expect such an intense level of
scrutiny just from the human eye, but the human eye coupled with AI can work
wonders.

 Risk identification and classification use risk- Machine Learning and Deep Learning
techniques can be applied to historical data sets to help identify clients who might be at
risk of generating “large loss” scenarios for car insurance or home insurance. Given that
an insurance company is likely to have a considerable body of data associated with large
loss incidents, putting this data to work and feeding it to a machine learning algorithm
that will learn to detect a client who poses a risk as a large loss candidate is an intriguing
use case for adopting AI, augmenting existing methods of identifying risk. Machine
learning techniques might be probabilistic or decision tree based and arrive at a level of
accuracy worthy of including as part of this analysis. There could be a considerable
number of parameters in a data set that might indicate patterns of large loss scenarios that
just aren’t detectable or intuitive when using traditional data analysis techniques.

4
https://www.mckinsey.com/industries/financial-services/our-insights/insurance-2030-the-impact-of-ai-on-the-
future-of-insurance(last accessed on 10th September, 2019)
 Applying deep learning to learn the “encoding” of data is another potential technique that
can be applied to detect and categorise large loss scenarios accurately. It’s possible that
deep learning techniques such as using auto encoders could deliver better accuracy than
machine learning techniques with larger, more complex data sets that require more
compute power. As with many of the use cases discussed here, incorporating non-
traditional data sources such as weather or travel as extra dimensions to traditional data
sources and models can help assess and classify risk.

The rise in the expectation of the economy has seen has early switch from traditional to
machines. The urgency is because the insured wants their interaction with insurance company to
be fast, smart and convenient. Artificial intelligence helps everyone as a call centre handle of
grievance, underwriter, customer.5 The year 2019 will be the turning point where real-life
implementations achieving quantified business benefits will replace the marketing hype around
AI. There is still much work to be done around this, especially since the technology is always
improving and we are finding more ways in which to use it.

5
https://www.kdnuggets.com/2018/05/poll-tools-analytics-data-science-machine-learning-results.(last accessed
on 10th September, 2019)
Conclusion:

 Data

AI is built with data and is data centric. There are two aspects to AI solutions when you consider
the data aspect. The first is building your AI. The second is the deployment of the AI model into
production and building the AI pipeline. When it comes to building AI, you need to ensure you
gather and collect your data in a passive way. Typically, you will build AI with data at rest. It
may reside in a data warehouse, or a big data Hadoop platform, or a combination of both. It is
essential that interrogation of data happens in a non-intrusive way. A key consideration might be
“do I want to run a large ad-hoc query on my production enterprise data warehouse”. The likely
answer to this is “no” and the solution might be to create a separate mirror system that is isolated
from production. When it comes to deploying AI, the next consideration is how you integrate it
into the existing data flow. Sometimes, AI will run ad-hoc or batch jobs based on demand.
However, some AI will involve near real time analysis and be engineered into the existing data
pipeline. An example of this is the “Data First Method”.6

 Infrastructure and technology development

Infrastructure and Technology Innovation As previously mentioned, AI is a data centric

workload. Developing AI includes the manipulation of data, building a machine learning model
and training it with data. This can be a time-consuming business. It’s important to have a
combination of the right skilled resources working with the right software on the right platform.
Data gravity is a key consideration for where the workload lives.

If you are training a machine learning model using a large dataset that is hosted on premise, it
would make sense to host the training in the same location. The same is true if the data is located
in the cloud – it makes sense to train your AI in the Cloud. Cloud increasingly means a
capability. Services can be hosted on a private cloud using infrastructure in a client’s data centre.
They can also be hosted on a public cloud. It is prudent to consider a multi cloud model, with
different cloud vendors and different on premise and hosted solutions based on requirements like
data security, access and control. It seems that most Insurance companies have strategies that
span both Private and Public Cloud. Therefore, it might make sense to build AI applications

6
https://www.ibm.com/analytics/datafirst( laat accessed on 10th September,2019)
using data hosted on premise, whilst publishing and deploying the AI on the cloud for
consumption. This can help data scientists and administrators by distributing and prioritising
jobs, whilst sharing a common platform and includes tooling designed to analyse and optimise
deep learning jobs

 Skills

AI can be built by a cast of many. Building and deploying an AI application could include: • A
Data Engineer: Somebody skilled in manipulating data using tools like python, or enterprise ETL
tools like Datastage. • A Data Domain Expert: Somebody who understands the data features.
This person would be closely related to the person consuming the AI or even the same person. •
A Data Scientist: Somebody who understands statistics and algorithms and data modelling
techniques. • A Developer is somebody who integrates the AI into an application. • A Business
Analyst: Somebody who can measure the value and impact on the business of the developed AI
application. • An IT Architect: Somebody who will collaborate with the above team and define
the requirements for the best platform and software available to the team that will host the
development and deployment of the solution. There is a vibrant and varied community of
businesses who specialise in Data Science and AI. They can help with all stages from
workshopping ideas (the art of the possible), building minimum viable products to assisting
clients to build their data science team.