AI Question bank

Q) What is planning in AI? Discuss partial order planning and Hierarchical planning in detail.

 Planning in AI involves the decision-making process undertaken by robots or computer programs

to achieve specific goals. A plan is essentially a sequence of actions, each with preconditions that
must be met before the action can be executed and effects that can be either positive or negative.
There
are two main types of planning: Forward State Space Planning (FSSP) and Backward State Space
Planning (BSSP).

Forward State Space Planning (FSSP) involves progressing from an initial state to a target state by
performing necessary actions along the way. This method faces the challenge of a large branching
factor but is sound in its approach

Backward State Space Planning (BSSP), on the other hand, starts from the target state and regresses
backward to identify the sub-goals necessary to achieve the final goal. While BSSP has a smaller
branching factor compared to FSSP, it may encounter inconsistency issues.

Partial order planning is an approach to planning that maintains a partial ordering between actions
and only commits to a specific order when necessary. It does not dictate the exact sequence of
actions, allowing flexibility in decision-making. For instance, in a partial order plan for baking a cake,
the order of obtaining ingredients from the store is not specified, allowing the agent to adaptively
gather items.

Hierarchical planning is a method that breaks down complex tasks into smaller, more manageable
sub-tasks, forming a hierarchy of actions. Each level of the hierarchy represents a different level of
abstraction, with higher levels focusing on overarching goals and lower levels dealing with specific
actions. Hierarchical planning helps in organizing and simplifying the planning process, making it
more efficient and scalable.

In summary, partial order planning allows for flexibility by maintaining a partial ordering between
actions, while hierarchical planning breaks down complex tasks into simpler sub-tasks to streamline
the planning process.

Q) Write a short note on:-

1) Language models of Natural Language Processing

2) AI applications in healthcare

3) Robotics

4) AI applications in Banking

1) Language models of Natural Language Processing:

Language models in Natural Language Processing (NLP) are algorithms designed to understand and
generate human language. These models are trained on large amounts of text data and learn
patterns in language, enabling them to perform tasks such as text generation, machine translation,
sentiment analysis, and speech recognition.
One of the fundamental concepts in language modeling is predicting the probability of a word or
sequence of words given a context. This is often done using techniques such as n-grams, recurrent
neural networks (RNNs), and transformer models like GPT (Generative Pre-trained Transformer) and
BERT (Bidirectional Encoder Representations from Transformers).

Language models have revolutionized various NLP applications, including chatbots, virtual assistants,
language translation services, and text summarization tools. They play a crucial role in enabling
computers to understand and generate human-like text, thereby enhancing communication
between humans and machines.

2) AI applications in healthcare:

Artificial Intelligence (AI) has made significant advancements in healthcare, revolutionizing various
aspects of medical diagnosis, treatment, and patient care. AI-powered systems analyze large
volumes of medical data, including patient records, medical images, and genomic data, to provide
insights and support decision-making by healthcare professionals.

AI applications in healthcare include disease diagnosis and prognosis, personalized treatment

planning, drug discovery and development, medical imaging analysis, and remote patient
monitoring. Machine learning algorithms can detect patterns and correlations in medical data,
leading to more accurate diagnoses and predictions of disease progression.

Moreover, AI technologies such as natural language processing (NLP) enable the extraction of
valuable information from unstructured medical texts, such as clinical notes and research articles, to
support evidence-based medicine and clinical research.

Overall, AI has the potential to improve healthcare outcomes, enhance patient experience, and
optimize healthcare delivery by enabling more precise diagnosis, personalized treatment, and
efficient healthcare management.

3) Robotics:

Robotics is a field of engineering and computer science that deals with the design, construction,
operation, and use of robots. Robots are autonomous or semi-autonomous machines capable of
performing tasks in the physical world, often in collaboration with humans or independently in
various domains such as manufacturing, healthcare, agriculture, and exploration.

Artificial Intelligence plays a crucial role in robotics by enabling robots to perceive and interact with
their environment, make decisions, and adapt to changing conditions. AI techniques such as machine
learning, computer vision, and natural language processing enable robots to understand and
respond to human commands, recognize objects and obstacles, navigate complex environments, and
learn from experience.

Robots are used in various applications, including industrial automation, logistics, healthcare
assistance, search and rescue operations, exploration of hazardous environments, and household
chores. Advances in AI and robotics are driving the development of more capable and intelligent
robots that can perform increasingly complex tasks and interact more seamlessly with humans.
4) AI applications in Banking:

Artificial Intelligence (AI) is transforming the banking industry by automating processes, enhancing
customer experiences, and improving decision-making. AI-powered systems analyze vast amounts of
financial data to detect patterns, identify trends, and make predictions, enabling banks to offer
personalized services, manage risks, and optimize operations.

AI applications in banking include fraud detection and prevention, credit scoring, customer service
automation, chatbots and virtual assistants, personalized financial recommendations, and
algorithmic trading. Machine learning algorithms analyze transaction data to detect anomalous
behavior and identify fraudulent activities in real-time, reducing financial losses and protecting
customers' assets.

Moreover, AI-powered chatbots and virtual assistants provide instant customer support, answer
inquiries, and assist with banking transactions, enhancing customer satisfaction and reducing service
costs. Personalized financial recommendations based on AI algorithms help customers make
informed decisions about investments, savings, and budgeting.

Overall, AI is revolutionizing the banking industry by enabling banks to offer more efficient and
personalized services, improve risk management, and enhance customer experiences.

Q) What do you understand by forward chaining and backward chaining? Explain in detail.

 Forward Chaining and Backward Chaining in AI:

In artificial intelligence, the inference engine plays a crucial role in deriving new information from
known facts using logical rules. Two common modes of inference employed by inference engines are
forward chaining and backward chaining.

Forward Chaining:

- Definition: Forward chaining, also known as forward deduction or forward reasoning, is a reasoning
method that starts with known facts in the knowledge base and applies inference rules (such as
Modus Ponens) in a forward direction to derive new information until a goal is reached.

- Algorithm: The forward-chaining algorithm iteratively triggers rules whose premises are satisfied by
the available facts and adds the conclusions to the known facts. This process continues until the
desired goal is achieved.

- Properties:

- Down-up approach: Starts from known facts and moves towards the goal.

- Data-driven: Relies on available data to infer new information.

- Used in expert systems, business rule systems, etc.

Backward Chaining:

- Definition: Backward chaining, also known as backward deduction or backward reasoning, is a

reasoning method that starts with the goal and works backward through the inference rules to find
known facts that support the goal.

- Algorithm: The backward-chaining algorithm begins with the goal predicate and attempts to prove
it by finding rules whose conclusions match the goal. It recursively breaks down the goal into sub-
goals until it reaches known facts in the knowledge base.

- Properties:

- Top-down approach: Starts from the goal and works towards known facts.

- Goal-driven: Determines which rules to apply based on the goal.

- Often uses depth-first search strategy for proof.

Horn Clause and Definite Clause:

- Horn Clause: A clause with at most one positive literal.

- Definite Clause: A clause with exactly one positive literal.

- These forms of sentences enable more restricted and efficient inference algorithms.

Example:

Consider the task of proving the statement "Robert is a criminal" using both forward chaining and
backward chaining approaches. Here's how we can do it:

1. Forward Chaining:

- Start with known facts.

- Apply inference rules (e.g., Modus Ponens) to derive new information.

- Continue until the goal is reached.

2. Backward Chaining:

- Begin with the goal predicate "Criminal(Robert)."

- Work backward through the inference rules to find supporting facts.

- Break down the goal into sub-goals until reaching known facts.
Conclusion:

Forward chaining and backward chaining are both essential reasoning methods in artificial
intelligence, each with its own advantages and use cases. While forward chaining starts with known
facts and derives new information, backward chaining starts with the goal and works backward to
find supporting facts. These techniques are fundamental to various AI applications, including expert
systems, automated theorem proving, and inference engines.

Q) Explain the concept of Reinforcement learning.

 Reinforcement Learning:

Reinforcement Learning (RL) is a type of machine learning where an agent learns to make decisions
by interacting with an environment. Unlike supervised learning, where the model is trained on
labeled data, and unsupervised learning, where the model finds patterns in unlabeled data,
reinforcement learning operates based on feedback received from the environment.

Key Concepts:

1. Learning through Feedback:

- In reinforcement learning, the agent learns through trial and error. It takes actions in an
environment and receives feedback in the form of rewards or penalties based on the consequences
of its actions.

- The agent's goal is to maximize the cumulative reward it receives over time by learning which
actions lead to favorable outcomes and which do not.

2. Dynamic Conditions:

- RL systems adapt to dynamic conditions in the environment. As the environment changes or

evolves, the agent must adjust its behavior to achieve its objectives.

- This adaptability allows RL systems to handle real-world scenarios where conditions may be
unpredictable or subject to change.

3. Performance Evaluation:

- RL systems evaluate their performance based on the feedback received from the environment.
Rewards are used to reinforce desirable behaviors, while penalties discourage undesirable
behaviors.

- The agent continuously assesses its actions and updates its strategy to improve its performance
over time.
Examples:

1. Self-Driving Cars:

- Self-driving cars are a prominent example of reinforcement learning in action. The car's AI agent
interacts with the environment (roads, traffic, pedestrians) and receives feedback (rewards or
penalties) based on its driving actions.

- By learning from this feedback, the AI agent adjusts its driving behavior to navigate safely and
efficiently.

2. AlphaGo:

- AlphaGo, the chess master algorithm developed by DeepMind, is another notable example of
reinforcement learning.

- AlphaGo learns to play the game of Go by playing against itself and receiving feedback on its
moves. Through this iterative process, it improves its gameplay and ultimately becomes a world-
class Go player.

Conclusion:

Reinforcement learning is a powerful paradigm for training intelligent agents to make decisions in
dynamic environments. By learning from feedback and adapting to changing conditions, RL systems
can achieve complex objectives and excel in tasks ranging from game playing to real-world
applications like autonomous driving.