(GIST OF YOJANA) Artificial Intelligence: Challenges and
Opportunities for India
[FEBRUARY-2020]
Artificial Intelligence: Challenges and
Opportunities for India
Introduction:
- Artificial Intelligence can be described as a system’s ability to learn
and interpret external data via software/algorithms or machines/devices for
problem solving by performing specific roles and tasks currently executed by
humans.
- The term AI has been used interchangeably with other closely related
terms such as expert systems, decision-support system, knowledge-based
systems, machine learning, natural language processing, neural networks,
pattern recognition, recommender systems and text mining.
Background:
- Although the origin of the term AI can be traced back to early 1950s,
the relatively recent advancement in information technology (such as big
data, improved computing, storage capability and super-fast speed of data
processing machines) and robotics has enabled AI to gain significant
momentum in terms of its development, application and use within public and
private sector organizations.
- The recent developments in AI offer the potential for significant
opportunities for industry, governments and society, but there are many
challenges and subsequent risks as Al-based systems are adopted for an ever
increasing range of tasks and duties. In this article, we aim to briefly
outline the opportunities and challenges, particularly focusing on elements
of policy that could act as a major roadblock for development and further
diffusion of AI-based systems.
Opportunities and Applications:
- A multitude of opportunities have been presented for the application and
use of Al-based systems in various domains particularly to assist where
structured decision making is needed.
- The ability of AI to the computationally intensive, intellectual and
perhaps creative limitations of humans opens up new application domains
within manufacturing, law, medicine, healthcare, education, government,
agriculture, marketing, sales, finance, operations and supply chain
management, public service delivery and cyber security.
- Within the education sector, AI can be deployed to improve teacher
effectiveness and student engagement by offering capabilities such as
intelligent game-based learning environments, tutoring systems and
intelligent narrative technologies. Schmelzer suggested that AI can impact
education in three ways.
- Firstly, Al-enabled hyper-personalisation helps in developing student
specific learning profiles and in developing customised learning
environments based on ability, preferred mode of learning and experience.
- Secondly, the use of smart assistants (Amazon Alexa, Google Home, Apple
Siri, and Microsoft Cortana) and associated technologies offer significant
potential to help students. Universities are already using voice assistants
to help answer common questions about campus, student schedules and courses.
- Thirdly, AI systems can assist educators with secondary tasks such as
grading activities, providing personalised responses to students, handling
routine and repetitive paperwork and dealing with logistics-related matters.
Al-based analytics can help with academic research within various
disciplines and potentially transform library processes and staffing
requirements with aim to provide a richer user experience.
- AI technology can be used within several other sectors for enhancing
both efficiency and effectiveness. Specifically, AI can help in achieving
good health and well-being goals within rural and remote areas in developing
countries where access to medical care is limited. In such scenarios,
Al-based systems can be utilised for conducting remote diagnosis supporting
doctors to help improve health service delivery.
- Al-based systems can also help achieve the “Zero Poverty and Zero
Hunger” (SDG 2) by assisting in resource allocation for predicting adverse
environmental conditions, diagnose crop diseases and identify pests in a
timely manner to mitigate the risk of catastrophic agricultural events.
Similarly, Al-based systems can be used to predict energy and utility demand
to help in achieving SDGs such as “Clean water, sanitation” and “Affordable
clean energy”.
Application of AI in India:
- Within the Indian context, a number of key indicators from health,
education and agriculture sectors are important to highlight as AI is
further adopted. India has 0.8 per thousand doctor-to-patient ratio (UK:
2.8, Australia: 5, China: approximately 4). This low ratio implies a heavy
workload on Indian doctors. In India, doctors spend just 2 minutes per
patient, whereas in the US it is close to 20 minutes. AI could be a valuable
assistive tool for doctors in helping reduce their workload and assisting in
diagnosis.
- Al-assisted diagnostics can provide access to quality healthcare for
people in remote areas. The per hectare cereal productivity in India is
almost half that of China and UK (3000 kg/ha vs. over 6000 kg/ha). There is
a significant loss of productivity due to pests and diseases.
- The Tamil Nadu e-Governance Agency has partnered with Anna University to
launch a Tamil smart assistant called `Anil”. This NLP-based smart assistant
provides a step-by-step guide to people in helping them apply online for
scores of critical government services. The Tamil Nadu Government has been
one of the pioneers in using AI for public service delivery.
- The agency has recently launched an Al-based agricultural pest and
disease identification system and made it available to over half a million
farmer families through a mobile app. The farmer clicks an image of diseased
crop or a pest and the system processes the image through an AI algorithm to
identify the pest or disease and sends a message to the farmer advising the
remedial measure. This system is gaining a good field response in which
nearly 400 fanners are posting identification requests and getting answers
every day.
- The Tamil Nadu Government is implementing an innovative use of AI
through face recognition for recording attendance. The system is saving more
than 45 minutes per day and is freeing up extra time for core educational
activities in schools. Within healthcare, AI solutions such as radiographic
diagnostics like “detection of internal bleeding in the brain from CT scans”
are being tried to assist doctors and increase their reach to serve remote
areas of India.
Challenges and Shortcomings:
- There exists a number of challenges and limitations of successfully
implementing and utilising AI in both public and private sector
organisations. Some of the key challenges are briefly outlined here.
Lack of explain ability:
- Generally AI operates effectively as a black-box-based system that does
not transparently provide the reasoning behind a particular decision,
classification or forecast made by the systems. This is a major limitation
of this technology as it has direct impact on transparency, hence trust and
confidence of using decisions made.
Lack of contextual awareness and inability to learn:
- Al-based systems are good at performing with given parameters and rules.
However, they still have major limitations in terms of making decisions
where context plays a critical role. Unlike humans, Al-based systems cannot
learn from their environment. This limits the application of AI to specific
types of domains.
Lack of standardization:
- Al-based systems that may have utilised different types of
technologies/techniques are increasingly being embedded in a variety of
products and services (for example, smart assistants, modules for enterprise
products, widely available cloud libraries and bespoke data science-driven
applications).
- This poses a critical question: how can the inferences delivered by
different AI components be integrated coherently when they may be based on
different data and subject to different ecosystem conventions (and the
associated quality differences)? Furthermore, organisations face challenges
on how to ensure AI and human work together successfully.
Job losses:
- Increasing automation will lead to significant job losses particularly
at operational and lower skill levels for repetitive tasks. This critical
consequence of AI Use will continue to impact all sectors and countries
across the world but particularly developing economies where employment
opportunities are already limited.
- This emphasizes the need for strategic management of AI transition
requiring organizations' to carefully consider a number of major challenges:
how to select tasks for automation; how to select the level of automation
for each task; how to manage the impact of Al-enabled automation on human
performance and how to manage Al-enabled automation errors.
Lack of competency and need for re-skilling and up-skilling workers:
- A large number of organisations still lack in-house competency to
successfully develop and implement Al-based systems. In such a scenario,
organisations utilise specialised consultancy firms which can be very
resource intensive.
- But this restricts organisations having limited resources in using such
systems. Similarly, using or working with Al-based systems requires workers
to be equipped with a new and advanced set of skills, which is a challenge
for government, organisations and individuals.
Lack of trust and resistance to change:
- Due to the above mentioned issues and negative media coverage on the
consequences of AI, people are generally apprehensive about its
implementation. This poses a major challenge on how to establish trust among
workers and stakeholders in the management of resistance to change in
adopting AI systems.
- Public policy is facing unprecedented uncertainty and challenges in this
dynamic world of AI. The velocity and scale of impact of AI is so high that
it creates an interesting dynamics in terms of the need to predict its
impact and inability to draw boundaries. We have identified six key public
policy challenges of AI.
Ethics:
- Ethics for machines has been an area of immense interest for the
researchers. However, defining has proven to be problematic and difficult to
make it computable.
- To tackle this, we need to deal with ethics purely from an AI
perspective. There are two dimensions of ethics in AI:
- Privacy and data protection and
- Human and environmental values.
(i) Privacy and Data Protection:
- Privacy is possibly the top-most concern while using AI systems. Users’
sensitive and highly granular data is likely to be stored and shared across
the AI network (for example, a person’s location for the day based on face
recognition and CCTV feeds, food habits, shopping preferences, movies, music
etc.).
(ii) Human and Environmental Values:
- Any AI system has to conform to the human value system and the
policymakers need to ask: Has the AI system been sensitised to human values
such as respect, dignity, kindness, compassion, equity or not? Does the
system know that it has a preferential duty towards children, elderly,
pregnant women, sick and the vulnerable? An important aspect which needs to
be built into AI systems is the overall cost of their decisions on the
society.
Transparency and Audit:
- In the future, many of the Al-based systems could be interacting with
humans in fields such as finance, education, healthcare, transportation and
elderly care. The technology providers must explain the decision-making
process to the user so that the AI system doesn’t remain a black box.
- There exists a legal need to explain the decision taken by such systems
in case of litigation. These AI systems must provide an audit trail of
decisions made not only to meet the legal needs but also for us to leam and
make improvements over past decisions.
Digital Divide and Data Deficit:
- Since the entire AI revolution has data at its foundation, there is a
real danger of societies being left behind. Countries and governments having
good quality granular data are likely to derive maximum benefit out of this
disruption. Countries where the data is of poor quality or of poor
granularity would be left behind in harnessing the power of AI to improve
lives of its citizens adversely affecting low-resource communities.
- AD can disrupt social order and hierarchy creating new social paradigms,
which could damage the social fabric exposing people lower in the bargaining
hierarchy with a real threat of exploitation and unfair treatment.
- This could lead to commoditization of human labour and chip away human
dignity. An AI system designed with equity as a priority would ensure that
no one gets left behind in this world. Another key need for autonomous
systems is fairness. They must not exhibit any gender or racial bias and
they must be designed to stay away from ‘social profiling’ (especially in
law enforcement, fraud detection and crime prevention areas). The recent
reports questioning the neutrality of AI systems used by police to identify
crime-prone individuals has brought this issue out in sharp focus.
Accountability and Legal Issues:
- Without AI, any system designed by a human is only a machine under the
control of the operator. Therefore, accountability has not been an issue.
- Almost all civil and criminal liability laws of the world fairly
unanimously attribute accountability to the operator, owner and manufacturer
of the machine in varying degrees depending upon the facts of the ease.
However, once machines are equipped with AI and take autonomous decisions,
the question of accountability becomes very hard to answer, more so when the
algorithms are unknown to the designer.
Misuse Protection:
- This possibly is the toughest of all six questions. How do we insulate
every new technology to prevent it from being twisted for achieving
destructive goals?
- An ease in point - how internet proliferated across the globe
benefitting billions but also carried along with it a wave of cybercrime,
malware, viruses and violent online games which resulted in loss of innocent
lives of teens around the world. Autonomous AI systems must be designed for
misuse protection. It cannot be an afterthought.
Conclusion:
- AI as a technology holds tremendous potential for a country like India,
which is data rich and has the requisite technological capability to create
AI solutions for many of its problems. States like Tamil Nadu have already
started deploying AI systems at scale for addressing sonic of the key
challenges in health, education and agriculture sectors.
- Public roll-out of AI systems needs to address issues of ethics,
transparency, audit, fairness, equity, accountability and misuse prevention.
An effective public policy framework for AI along with a practical scorecard
would be needed to make this AI revolution work towards an equitable
prosperity.
