Unit -2 [Overview of Artificial Intelligence (AI)]

Philosophy of AI:
The philosophy of AI explores deep questions about the
nature of intelligence, whether machines can achieve
human-like understanding, and the ethical implications of
creating artificial entities with cognitive abilities.
For example, one philosophical question is: can AI systems
truly understand the information they process, or are they
merely manipulating symbols without comprehension?
Philosophers like John Searle have debated this issue, using
thought experiments such as the Chinese Room, which
questions whether AI can ever possess "true" understanding
or consciousness.

AI Technique:

Artificial Intelligence techniques refer to a set of methods

and algorithms used to develop intelligent systems that can
perform tasks requiring human-like intelligence. Some of the
widely used ones are:

 Machine Learning.

 Natural Language Processing.

 Computer Vision.

 Deep Learning

 Data Mining

 Robotics.
Machine Learning:

This approach involves the building of algorithms to learn

patterns in data and make predictions based on it.

1. Unsupervised machine learning -AI systems analyse

unlabelled data, where no predefined outcomes are
provided. The objective is to uncover inherent structures or
patterns within the data without any prior knowledge. For
instance, it can group similar customer behaviour data to
identify customer segments for targeted marketing
strategies.

2. Supervised learning - A combination of an input data

set and the intended output is inferred from the training
data. AI systems learn from a labelled dataset, where each
data point is associated with a known outcome. For instance,
it enables email spam filters to distinguish between spam
and legitimate emails based on learned patterns.

3. Semi-supervised learning - It is a method that uses a

small amount of labelled data and a large amount of
unlabelled data to train a model. The goal of semi-
supervised learning is to learn a function that can accurately
predict the output variable based on the input variables,
similar to supervised learning. However, unlike supervised
learning, the algorithm is trained on a dataset that contains
both labelled and unlabelled data.

4. Reinforcement learning - In RL, the data is

accumulated from machine learning systems that use a trial-
and-error method to learn from outcomes and decide which
action to take next. After each action, the algorithm receives
feedback that helps it determine whether the choice it made
was correct, neutral or incorrect. It performs actions with the
aim of maximizing rewards, or in other words, it is learning
by doing in order to achieve the best outcomes.

Natural Language Processing:

Natural Language Processing involves programming

computers to process human languages to facilitate
interactions between humans and computers.

However, the nature of human languages makes Natural

Language Processing difficult because of the rules involved
in passing information using natural language. NLP leverages
algorithms to recognize and abstract the rules of natural
languages, converting unstructured human language data
into a computer-understandable format.
Applications of Natural Language Processing can be found in
IVR systems and applications used in call centres, language
translation applications like Google Translate, and word
processors such as Microsoft Word to check the accuracy of
grammar in text.

This AI technique has paved the way for virtual assistants,

chatbots, and language translation tools, making
communication between humans and machines more
seamless than ever.

Deep Learning:

Deep learning is the branch of machine learning which is

based on artificial neural network architecture. An artificial
neural network or ANN uses layers of interconnected nodes
called neurons that work together to process and learn from
the input data.
In a fully connected Deep neural network, there is an input
layer and one or more hidden layers connected one after the
other. Each neuron receives input from the previous layer
neurons or the input layer. The output of one neuron
becomes the input to other neurons in the next layer of the
network, and this process continues until the final layer
produces the output of the network. The layers of the neural
network transform the input data through a series of
nonlinear transformations, allowing the network to learn
complex representations of the input data.

The main applications of deep learning can be divided into

computer vision, natural language processing (NLP), and
reinforcement learning.

Fundamental Problems and Challenges

Here's a detailed explanation of these major AI challenges:

1. Realism

Challenge: Creating realistic simulations or responses that

mimic human behavior and understanding the intricacies of
human interaction, intuition, and behavior.

Details: AI systems, especially those in virtual

environments, strive to simulate real-world physics, decision-
making, and human-like behavior. However, producing
realistic outputs—whether it’s in generating human-like text,
creating authentic virtual simulations, or replicating
emotions—is extremely complex. The nuances in human
emotions, context, and dynamic behaviors often make it
hard for AI to appear genuinely realistic. AI needs to go
beyond basic patterns to understand subtleties like sarcasm,
irony, and cultural context.

Example: Chatbots attempting to hold meaningful

conversations often struggle with nuances. For example, if a
user expresses frustration or anger in subtle ways, the
chatbot may not pick up on it, thus failing to respond in a
truly realistic and empathetic manner.

2. Brittleness

Challenge: Many AI systems are "brittle," meaning they

perform well only within the limited range of scenarios they
were trained on and fail or make mistakes outside of this.

Details: AI systems, especially those based on machine

learning, are often brittle because they are trained on
specific data and may not adapt well to new situations or
environments. When exposed to conditions or inputs outside
of their training data, these systems can fail unpredictably,
making them unreliable in complex real-world applications.

Example: An image recognition model trained to identify

objects in daylight might fail to accurately identify objects in
low-light conditions. For instance, a self-driving car’s AI
might struggle to recognize obstacles in foggy conditions if it
was not adequately trained on such data.

3. Scalability

Challenge: As AI models grow in complexity, scalability

becomes an issue in terms of computation, data storage,
and energy consumption.
Details: Building large, complex models like GPT-4 or
AlphaGo requires massive amounts of data, computational
power, and energy. Scaling these systems to handle more
data, more parameters, or to be more widely distributed is
not just a technical challenge but also an economic and
environmental one.

Example: Training a large language model requires

immense computational resources and can take weeks or
even months on specialized hardware. The energy costs are
also considerable, raising sustainability concerns for large-
scale AI models.

4. Real-Time Requirements

Challenge: Some AI applications, such as autonomous

driving or real-time medical diagnostics, need to process and
act on data immediately.

Details: Real-time processing demands low latency, high

computational speed, and efficient algorithms to make quick,
accurate decisions. Achieving real-time responses is
challenging, especially in complex scenarios where multiple
variables are at play, and mistakes can have significant
consequences.

Example: In autonomous driving, AI needs to process a

continuous stream of sensor data to recognize and react to
road hazards instantaneously. Delays in processing could
lead to accidents, requiring the AI to be highly optimized for
real-time decision-making.

5. The Frame Problem

Challenge: Determining what is relevant in a situation and

deciding which information to focus on when making
decisions.
Details: The Frame Problem refers to an AI's difficulty in
understanding context or relevance without specific
programming. Unlike humans, who naturally know what to
focus on in a given situation, AI systems struggle to identify
the boundaries of relevance. This makes it challenging for AI
to operate in dynamic environments where only certain
details are important.

Example: In a household robot, it may need to understand

that if a person is resting, it should avoid making noise. This
requires the robot to frame the context of a "quiet
environment" as relevant to someone resting—a concept
that isn't directly programmable and must be inferred.

6. Common-Sense Knowledge and Reasoning

Challenge: Developing AI with an understanding of basic

facts about the world and the ability to reason about them.

Details: Common sense includes basic knowledge like

"water is wet" or "fire is hot" and is crucial for AI to interact
naturally with humans. However, encoding this knowledge is
difficult because common sense varies greatly in complexity
and is often implicit. AI struggles to understand and apply
these intuitive facts without specific training or
programming.

Example: A customer service chatbot may misunderstand a

user’s intent if it lacks common-sense reasoning, such as
interpreting “My laptop won’t start” as a request for
troubleshooting rather than taking it literally.

Ethical and Society Related Implications, Laws for AI

in India, and Best Practices
The rapid advancement of AI brings several ethical concerns,
including:

- Job Displacement: AI-driven automation threatens to

replace many jobs, especially in manufacturing and
customer service. For example, factories increasingly use
robots instead of human workers for repetitive tasks.

- Privacy Issues: AI technologies, such as facial recognition,

raise concerns about surveillance and the potential for
misuse. In countries like China, AI-powered cameras are
used to monitor citizens in public spaces.

- Bias in AI Systems: AI systems may inherit biases present

in their training data. For example, some AI models used in
hiring processes have been found to favor male candidates
over females due to biased historical data.

- Ethical AI in India: India has started taking steps to regulate

AI with frameworks like the National AI Strategy, which
promotes AI development while ensuring ethical use. Laws
such as the Personal Data Protection Bill seek to govern data
privacy in the AI landscape.

- Best Practices: Transparency, accountability, and fairness

are some of the best practices being promoted for the
ethical use of AI. For example, AI systems in healthcare
should be transparent about how decisions are made,
ensuring that patients understand why a diagnosis was
suggested.

Applications of AI
AI has wide-ranging applications across industries. Some
examples include:

- Transportation: AI is used in autonomous vehicles to

navigate roads, make real-time driving decisions, and avoid
accidents. Tesla’s self-driving cars use AI to detect lanes,
pedestrians, and other vehicles.

- Home/Service Robots: AI-powered home assistants like

Amazon’s Alexa or vacuum robots like Roomba help with
household tasks. These robots can perform cleaning, provide
reminders, or answer questions based on voice commands.

- Healthcare: AI assists doctors in diagnosing diseases and

recommending treatment. For example, IBM Watson Health
uses AI to analyze medical data and help doctors identify
cancer treatments.

- Education: AI-driven tools like personalized learning

platforms adapt to students' needs, offering customized
lessons based on their progress. Platforms like Coursera use
AI to recommend relevant courses to learners.

- Banking: AI helps detect fraud by analyzing transaction

patterns and flagging unusual activity. For instance, AI
systems are used in credit card fraud detection to prevent
unauthorized transactions.

- Entertainment: Streaming services like Spotify and Netflix

use AI to recommend music and shows based on users’
viewing and listening history.
- Public Safety and Security: AI is used for surveillance and
crime prediction. For example, AI systems in police
departments analyze crime data to predict where crimes are
likely to occur and allocate resources accordingly.

Malicious Use of AI: Prevention and Mitigation

While AI offers significant benefits, it can also be used

maliciously. Examples include:

- AI-generated Deepfakes: AI can be used to create realistic

but fake videos of people, often used for disinformation. For
example, deepfake videos have been created to make
politicians appear to say things they never said.

- Autonomous Weapons: AI can be exploited to create

autonomous drones or robots used in warfare, posing a
significant threat to global security.

- Prevention and Mitigation: To counter malicious use of AI,

security measures include robust authentication
mechanisms, regular monitoring, and AI models that can
detect and mitigate cyber threats in real-time.
- Security Domains: In digital security, AI is used to detect
cyber threats such as malware or phishing attacks. In
physical security, AI is used in surveillance systems to
monitor public spaces and detect suspicious activity. In
political security, AI is used to identify fake news and
disinformation campaigns.
- Factors Affecting AI and Security Equilibrium: Governments
and industries need to work together to create regulations
and frameworks that balance the potential of AI with its
security risks. Regular audits and ethical AI standards can
help ensure that AI is used safely and responsibly.