Module 4

7a) List and explain any four Trusted AI principles.

Trusted AI principles are guidelines to ensure that Artificial Intelligence systems are
ethical, safe, and beneficial for [Link] include the following:
1. Responsible
AI systems should protect human rights and privacy. This means handling data
securely, using it ethically, and ensuring AI does not cause harm or discrimination.
2. Accountable
Organizations and developers must take responsibility for AI decisions and outcomes.
There should be mechanisms to monitor AI systems, receive feedback, and correct
errors or biases.
3. Transparent
AI systems should be understandable to users. This involves explaining how AI
makes decisions or recommendations so users can trust and appropriately use the
system.
4. Empowering
AI should support economic growth, job creation, and improved productivity. It
should help people—customers, employees, and society—rather than replace or
disadvantage them unfairly.
5. Inclusive
AI must respect social and cultural values and benefit everyone affected, not just its
creators. This includes designing AI that is fair, accessible, and unbiased across
different groups.

8b What is expert system? Explain three components of expert system.

An expert system is an AI-based computer program that solves complex problems in a
specific domain using expert-level knowledge. It mimics the reasoning process of a
human expert and provides intelligent solutions, suggestions, and decisions.
• First expert system appeared in 1970
• Uses facts + heuristics
• Does not replace experts but assists them.
Main Components of an Expert System
1. User Interface

 Acts as a bridge between the user and the expert system.

 Users enter queries in understandable language.

 The system processes the query and displays results in a readable or audible
format.

 It also explains how a conclusion was derived, proving system credibility.

2. Inference Engine (Brain of the System)
o Considered the brain of the expert system.
o Applies rules to the knowledge base to draw conclusions or generate new
information.
o Handles conflicts when multiple rules apply by selecting the best one.
o May include explanation and debugging features.
o Types:
 Deterministic: Uses rules and facts to give definite conclusions.
 Probabilistic: Makes decisions based on probabilities, allowing
uncertainty.

o Reasoning Strategies:
 Forward chaining → “What will happen next?”

 Backward chaining → “Why did this happen?”

3. Knowledge Base
• Stores all domain knowledge needed for decision-making.
• Contains both factual knowledge (universally accepted) and heuristic knowledge
 (based on experience).
• Knowledge is stored mostly in If-Then rules.

c. Relate the role of ethics in AI.

Role of Ethics in Artificial Intelligence
 Ethical AI ensures decisions are right, fair, unbiased and justified.
 Example of unethical AI:
Microsoft Tay chatbot (2016) learned toxic behaviour from users due to biased inputs.
Ethical AI must:
 Use unbiased data
 Use appropriate models
 Continually monitor results
AI is used in sensitive areas: Healthcare, Law enforcement, Hiring, Autonomous vehicles
and Defense
Unethical AI can: Violate privacy, Produce biased decisions, Make decisions that cannot
be justified and can Cause harm without clear liability
Examples:
• US hospitals’ AI favoured white patients.
• Amazon hiring tool biased against women.
Goal of Ethical AI:

1. Support human autonomy

2. Technical robustness and safety
3. Protect data privacy
4. Transparency and explainability
5. Fairness and nondiscrimination
6. Societal & environmental wellbeing
7. Accountability
8a) Explain the working of an expert system taking any example.
An expert system is a computer-based system that mimics the decision-making ability of a
human expert in a specific domain. It does not replace human experts but assists them in
solving complex problems.

Main Components

1. User Interface (UI)

It allows the user to interact with the expert system by entering queries and receiving
results in an easy-to-understand form.

2. Knowledge Base (KB)

It stores domain-specific knowledge in the form of facts and rules (usually if–then
rules). This knowledge is collected from human experts.

3. Inference Engine
It is the brain of the expert system. It applies logical rules to the knowledge base to
derive conclusions. It uses techniques like forward chaining or backward chaining.

Working of an Expert System (Example: MYCIN)

MYCIN is a medical expert system used to diagnose bacterial infections and recommend
treatment.

Step 1: Knowledge Base Creation

The knowledge base is created using expert knowledge about bacterial infections, their
symptoms, causes, and treatments.

Step 2: User Input

A doctor enters patient details such as symptoms, age, medical history, and test results
through the user interface.

Step 3: Inference Process

The inference engine applies if–then rules from the knowledge base.
Example:
If the patient has fever and high white blood cell count, then there may be a bacterial
infection.
Step 4: Reasoning
MYCIN uses backward chaining to confirm possible diseases by checking whether the
given facts support a particular diagnosis.
Step 5: Output
The system displays the diagnosis and suggests suitable antibiotics along with an explanation
of how the conclusion was reached.

8b. AI could be programmed to do something beneficial, but the method used to

achieve its goal can be highly destructive, Explain why?

Ethical Use of Artificial Intelligence

• An ethical AI system must be unbiased, fair, transparent and explainable.
• Black box AI = AI system whose decisions cannot be explained.
Industries like finance and healthcare need explainable AI for regulatory reasons.

Ethical AI must be:

• Explainable
• Monitorable
• Reproducible
• Secure
• Human-centred
• Unbiased
• Justifiable

AI can be dangerous in two cases:

1. AI programmed to cause harm
Example:
• Autonomous weapons → risk of AI war if misused.
• Hard to “turn off.”
2. AI programmed for good but misinterprets its goal
Example:
• A self-driving car told “reach fast” may break rules and cause havoc.
Other Risks:
• Hacking of AI-controlled devices (cars, drones, hospital equipment).
• No clear legal responsibility for AI actions.
• Job losses due to automation (RPA, robotics).

8c) What is Artificial Intelligence of Things (AIoT). Explain how Does AIoT Work?

AIoT:-
• Artificial Intelligence of Things (AIoT) is the combination of Artificial Intelligence
(AI) technologies with Internet of Things (IoT) infrastructure.

• AI provides intelligence (learning, decision-making).

• IoT provides connectivity (devices, sensors, data transfer without human help).
• Together, AIoT makes systems more:

• Efficient

• Smart
 • Automated

• Data-driven

Examples of IoT devices:

• Heart-rate monitor, smart doorbell, self-driving cars, smart meters, wearables.

Why AIoT?
AI improves IoT by analysing massive sensor data and enabling automated, accurate
decisions.

How AIoT Works?

1. AI is embedded inside IoT devices (chips, firmware, programs).
2. All devices are connected through an IoT network.

3. Devices generate & collect data.

4. Data is exchanged using pre-written APIs → fully automated.

5. AI algorithms analyse the data → take decisions in real-time.

6. Using Edge AI, data is processed near the device to:

• Reduce delays

• Save bandwidth

• Improve real-time response

Example:
Autonomous vehicles have AI-powered chips + IoT sensors → gather surroundings data
→ AI decides instantly how to drive.

[Link] is Neuromorphic Computing? Explain its architecture, features

resembling the human brain, and how it contributes toward achieving Artificial
General Intelligence (AGI).

Neuromorphic Computing
• Neuromorphic computing is a modern computing approach where both hardware
and software are designed by imitating the human brain and nervous system. It
combines ideas from computer science, biology, mathematics, electronics, and
physics.
• The main goal is to build artificial neural systems that behave like the human brain,
meaning they can:
• Learn and store information
• Make logical decisions
• Acquire new knowledge
• Work like a cognition machine
Neuromorphic systems can achieve both high speed and low power usage.
Features of Neuromorphic Systems:-
Neuromorphic computers resemble the brain because they are:
 • Massively parallel – handle many tasks simultaneously
• Event-driven – respond only when changes occur
• Power efficient – consume very less energy
• Flexible – adapt and generalize to new problems
• Fault tolerant – still work even if some parts fail
Key Concepts and Mechanics of Neuromorphic Systems:-
Spiking Neural Networks (SNN)
Neuromorphic systems use neurons and synapses like the human brain.
• Neurons process information
• Synapses connect neurons and transfer signals
• Unlike traditional neural networks (digital), neuromorphic chips use spiking neural
networks (SNNs) that send signals in the form of electrical pulses (spikes).
• This makes them behave more like real brain communication.
Neuromorphic Computing & AGI
• Artificial General Intelligence (AGI) refers to AI with intelligence equal to humans.
• Researchers believe neuromorphic computing may help achieve AGI because it
mimics real brain-like learning.
Example:-Human Brain Project (HBP) developed the SpiNNaker neuromorphic
supercomputer, which simulates brain functioning.
A machine is considered AGI if it can:
• reason and judge in uncertain situations
• plan
• learn
• communicate using natural language
• represent knowledge
• combine skills to achieve goals
AGI could even become self-aware, and tests like the Turing Test can help evaluate it.

7b) Describe the concept of AI as a Service (AIaaS). Outline two advantages and two
challenges of using AIaaS in organizations.
Artificial Intelligence as a Service (AIaaS) :-
• AI as a Service (AIaaS) refers to using ready-made AI tools hosted on the cloud,
instead of building in-house AI systems.
• Makes AI affordable, scalable, and accessible even for small businesses.
• Works similar to Software-as-a-Service (SaaS) — AI services are delivered over cloud
computing platforms.
Why AIaaS became popular?
• Cloud computing made AI tools widely accessible.
• Organizations can now store and analyze large amounts of data
• No need to build private infrastructure for AI development and testing.
Benefits of AIaaS
1. Focus on core business, not on becoming AI experts.
2. Reduced infrastructure & operational cost -No need for GPUs, parallel machines, large
AI teams.
3. Lower investment risks
4. Short deployment time
Challenges of AIaaS
1. Reduced data security-Data must be shared with third-party cloud vendors, raising
 privacy and security concerns.
2. Vendor dependence (Reliance)-Organizations depend on external providers for
processing, which may cause delays.
3. Reduced transparency-AIaaS works like a black box — companies don’t know which
algorithms or processes are used.
4. Data governance issues-Organizational data policies may restrict usage of third-party
cloud platforms.

7c. List and explain the risks associated with Artificial Intelligence and their
societal impact.
Risks of AI:-
1. Job Loss (Immediate biggest risk)
• AI and automation are quickly replacing people in simple, repetitive jobs.
Examples:
• Robots working as waiters in restaurants
• Automated systems in BPOs
• Warehouse robots
• Retail automation
Studies show:
• 36 million U.S. jobs are at high risk (Brookings Institution, 2019).
• Even white-collar jobs may be affected.
• The African American workforce is predicted to be hit hardest (McKinsey,
2018).
2. Biased Algorithms (GIGO Problem)
• GIGO = Garbage In, Garbage Out
• If AI is trained on biased data, it will give biased results.
Example:
• A facial recognition billboard in Utrecht was switched off after public outrage
because it collected unnecessary personal data.
Biased AI can:
• Discriminate
• Make unfair decisions
• Cause social problems
3. Privacy Loss
• AI and IoT devices constantly collect huge amounts of data.
Facts:
• 2.5 quintillion bytes of data are created every day.
• 90% of the world’s digital data was created in the last two years.
• AI-powered cameras can detect: Age,Gender,Ethnicity,Emotions,Gestures
Example:
• In China, police officers wear glasses with facial recognition to spot
“suspects”.
Once data is collected, we cannot know:
• Who is using it
• How it is used
• For what purpose
4. Everything Becomes Unreliable (Fake Media)
• AI can easily create: Fake images,, Fake videos, Fake voices, Fake tweets,
Fake news, AI-generated faces of people who do not exist
Example:
• Deepfake software can put someone's face onto someone else’s body in a
video.
• Even normal citizens are blackmailed with morphed images and/or videos.
This is often known as Faceswap videoblackmailing
Consequences:
• Celebrities suffer through fake videos
• Normal people are blackmailed using morphed videos
• Political fake videos are used to spread propaganda
Case:
• Cambridge Analytica used data from 87 million Facebook users to influence
U.S. elections (Trump campaign).

[Link] AI Bias. Describe the sources of bias, real-world examples, and methods to
mitigate bias in AI systems

AI and Bias
AI Bias = Skewed/unfair outcomes towards a group.
Sources of Bias:
1. Data Bias
• If training data is biased → output becomes biased.
• Example: Voice assistants struggle with non-American accents.
2. Algorithm Bias
• Algorithms may amplify data bias.
• Example: Image search for “teacher” shows mostly women.
3. People (Human Bias)
• Developers’ unconscious bias appears in systems.
• “Bias is not the machine’s fault, but the human’s.”
Examples of bias:
• “Doctors” shown as men; “nurses” shown as women.
• Vision systems misidentify darker skin.
• All voice assistants use female voices.

Towards Ethical and Trustworthy AI

Ways to achieve ethical AI:
1. Regulations
• EU (2021): First legal AI framework based on risk levels.
• Bans harmful systems like social scoring or manipulation.
2. Organizational Actions
• Cultural training, audits, risk assessments.
• Prevention–Detection–Response frameworks.
3. Rights & Activist Groups
• Push for human-centric AI.
• Question necessity and fairness of systems.
4. Ensuring Data Privacy
• Encrypt data
• Prevent algorithms from learning outside the dataset
• Secure computation
• Prevent reverse engineering
5. Diversifying AI Teams
• Reduce bias by including:
• Scientists
• Lawyers
• Sociologists
• Ethicists
• Use tools like Bias Analyzer.