Module 4

(Text Book 1: Chapter 26)

Philosophical Foundations
Artificial Intelligence raises foundational questions about the nature of minds, consciousness, and ethics.
Philosophers ask: Can machines think? Should machines think? What are the consequences if they do?
WEAK AI: CAN MACHINES ACT INTELLIGENTLY?
Key Points
• Weak AI claims that machines can behave as if they are intelligent and simulate cognitive functions,
without necessarily having real understanding.
• The field of AI was founded assuming weak AI is possible — machines can simulate every aspect of
learning and reasoning.
• Turing rephrased the question “Can machines think?” into a practical test — the Turing Test —
based on observable behaviour rather than internal consciousness.
• Objections to Weak AI include:
o Argument from Disability: Claims machines can’t perform distinctly human abilities such as
kindness, humour, or creativity. Yet machines already outperform humans in many cognitive
domains.
o Mathematical Objection: Based on Gödel’s incompleteness theorem; critics argue humans
surpass formal computation, but this lacks evidence.
o Argument from Informality: Human behaviour cannot be captured by explicit rules. AI has
responded with learning, probabilistic reasoning, and embodiment-based approaches.
Core takeaway: Weak AI is pragmatically accepted and proven useful, though philosophical debates
persist.
Weak AI (Narrow AI) Examples
▪ These systems are designed to perform specific tasks and do not possess consciousness or
general understanding.
▪ ChatGPT – generates text based on prompts
▪ Spam filters in email systems
▪ Recommendation systems – Netflix, YouTube, Amazon product suggestions
▪ Autonomous robots used in manufacturing
▪ Face recognition on smartphones
▪ Chess-playing programs like Deep Blue

1
STRONG AI: CAN MACHINES REALLY THINK?
Key Points
• Strong AI asserts machines can possess genuine minds, consciousness, and subjective
understanding—not just behaviour.
• Critics argue simulation ≠ real thinking; Turing responds that we do not verify human consciousness
either—we assume it socially.
• Central issues:
1. Mind–Body Problem: Are mental states physical or separate? Physicalism supports
possibility of Strong AI.
2. Functionalism: Mental states are defined by causal roles, not materials. Replacing neurons
with electronic equivalents challenges biological objections.
3. Chinese Room Argument (Searle): Syntax ≠ semantics; running a program doesn't
guarantee understanding. This remains a central challenge to Strong AI.
4. Consciousness & Qualia: Machines could mimic behaviour yet lack subjective
experience—this is the explanatory gap.
Core takeaway: Strong AI faces unresolved debates around consciousness, intentionality, and subjective
experience.
Strong AI (General AI) Examples
▪ Hypothetical systems that would possess human-like consciousness, reasoning, and self-
awareness. They do not exist yet.
▪ Possible/Futuristic Examples:
▪ A machine that can think, understand, and learn like a human in any domain
▪ A robot that can independently make decisions, understand emotions, and develop its own
goals
▪ AI capable of creative thinking and common-sense reasoning like humans

Note: There are no real-world examples of fully developed Strong AI today — it remains a theoretical
concept.

(Ref. Book 1: Chapter 8 – 8.1 & 8.2)

THE ETHICS AND RISKS OF DEVELOPING ARTIFICIAL INTELLIGENCE
Key Points
AI presents societal, legal, and existential risks beyond technical challenges. Major concerns include:
• Job displacement: Automation may eliminate human roles, though historically AI has created new
opportunities.
• Loss of human uniqueness: AI capabilities challenge our identity and autonomy, similar to earlier
scientific disruptions.
• Undesirable uses of AI: Autonomous weapons, mass surveillance, and power concentration pose
real dangers.
• Accountability and legal liability: If AI make wrong decisions (e.g., medical diagnosis),
responsibility becomes unclear.

2
 • Existential risk / Singularity: Ultra intelligent machines could surpass humans, potentially making
harmful decisions or evolving unintended goals.
• Asimov’s Laws and Friendly AI research attempt to embed human-aligned behaviour into intelligent
agents.
Core takeaway: Ethical design, transparency, and controlled evolution of AI systems are essential to
safeguard future human society.

I. Current Trends in Artificial Intelligence and Ethical Concerns (Section 8.1)

A. AI and Ethical Concerns: Definitions and Frameworks
Ethics defines the discipline dealing with the moral obligations and duties of humans. Correspondingly, the
ethics of AI deals with ethics specific to robots and other artificially intelligent beings.
AI ethics can be divided into two main groups:
1. Roboethics: Considers the moral behavior of humans as they design, construct, use, and treat
artificially intelligent beings.
2. Machine ethics: These ethics manage the moral behavior of artificial moral agents (AMAs) that
play an important role in delivering artificial general intelligence (AGIs) within existing legal and
social frameworks.
Ethical Use of AI (8.1.1):
• AI systems offer wide-ranging functionality but their use raises ethical questions.
• Ethical AI is the theory behind developing computer systems that perform tasks requiring human
intelligence.
• It focuses on how right, how fair, and how just the AI’s output, outcome, and impact are.
• An ethical AI system must be in synchronization with the ethical standards and norms of our society.
• An ethical system is designed to solve a particular problem, uses unbiased data, is trained using
the right learning model, and is monitored to evaluate if the results produced are right and fair.
• An AI system that works on biased data is considered unethical.
B. Trustworthy AI Principles and System Features
A strong understanding of ethical issues is critical to the success of AI technology. Guiding principles aim to
ensure AI is doing good for the planet.

Figure 8.1 (Trusted AI Principles)

3
Core Concerns Posed by AI (Figure 8.1):
1. If AI generates human-like output, can it also make human-like decisions?
2. If AI makes human-like decisions, are these decisions human-like?
3. If AI takes decisions (e.g., sanctioning or rejecting a bank loan), is that decision justified?
4. If AI decides on college enrollment, is it taking a fair decision?
5. If AI makes a human-like decision, is it human-like trustworthy?
Features of a Responsible/Ethical AI System (Figure 8.2):

Figure 8.2 Features of Responsible AI

• A responsible AI system is explainable.
• A responsible AI system is monitorable.
• A responsible AI system is reproducible.
• A responsible AI system is secure.
• A responsible AI system is human-centered.
• A responsible AI system is unbiased.
• A responsible AI system is justifiable.
Trustworthy AI Principles (European Commission High-Level Expert Group):
1. AI systems should support human autonomy and decision-making.
2. AI systems should be technically robust and safe to use, including having a fallback plan for when
things go wrong.
3. AI systems should ensure data privacy by protecting user data and providing mechanisms to
maintain data quality and integrity.
4. AI systems should not work on biased data; the system must be clear, explicable, and transparent
about its underlying data and models.
5. AI systems should be trained with diverse, non-discriminatory, and fair data and models to
avoid unfair bias.
6. AI systems should be designed to benefit everyone, focusing on societal and environmental
wellbeing to be sustainable and environmentally friendly.

4
 7. An AI system should have the accountability to ensure accurate, unbiased outcomes with
utmost responsibility.
8. An AI system must give due importance to data privacy and data security. It must ensure data is
collected, used, managed, and stored safely and responsibly.
9. An AI system must be accountable for its decisions; there must be someone in the organization
who will be held responsible for any problematic situation.
C. AI Risks and Societal Impact
The sources discuss several risks and potential threats associated with the rapid growth of AI capabilities.
1. AI Safety and Danger (8.1.2)
Most researchers believe that super-intelligent AI is unlikely to exhibit human emotions, meaning it probably
won't become malevolent. AI can become a risk only under two conditions:
• Specifically programmed for destruction: Autonomous weapons are AI systems programmed to
kill. If acquired by malicious people, they could lead to an AI war and mass casualties, potentially
ending humanity. These weapons are designed to be extremely difficult to 'turn off,' risking rapid loss
of human control. This risk also exists with narrow AI given autonomy (e.g., an autonomous drone
with facial recognition and a 3D-printed gun).
• Destructive methods for beneficial goals: AI could be programmed to do something beneficial, but
the method used might be highly destructive, especially if the programmer does not clearly outline the
goals. For example, instructing a self-driven car to reach a destination "as soon as possible" without
addressing safety or road rules might lead the car to complete the task "after creating a great havoc".
To mitigate these risks, there must be a provision to continuously monitor and control the machine, its
goals, and the processes it performs.
2. Legal Ambiguity and Accountability
• As of now, there is no liability for actions on machines.
• There is no clarity on what legal aspects bind machines when they become increasingly smart.
• Questions arise regarding whether AI systems should be judged the same way as humans, who is
responsible if the AI system becomes self-learning and autonomous, and if there is an error margin
for AI machines even with fatal consequences.
• Regarding AGI, questions exist about whether it should be handled ethically and legally, with some
arguing it should be treated as a nonhuman animal in the eyes of the law, while others believe
consciousness must first be inculcated.
3. Job Automation (8.1.2 and 8.1.7)
• Immediate Risk: As AI robots become smarter and more dexterous, they will quickly replace
humans.
• Deployment of AI systems will eventually replace certain types of jobs, especially those that are
predictable and repetitive.
• Statistics on Displacement:
o 30% (around 3 million) of 9 million low-skilled services and BPO roles will be lost by 2022
due to robot process automation (RPA).
o A 2019 Brookings Institution study found that 36 million people work in jobs they may soon
lose because at least 70% of their tasks can be done using AI (including retail sales, market
analysis, hospitality, and warehouse labor).
o A newer Brookings report states that white collar jobs may be worst affected.

5
 o A McKinsey & Company report (2018) suggested the African American workforce will be
hardest hit by automation.
o McKinsey suggests that by 2030, intelligent agents and robots could replace as much as
30% of the world’s current human labor, displacing between 400 and 800 million jobs and
requiring 375 million people to switch job categories.
o The World Economic Forum predicts AI/ML will displace 75 million jobs but create 133
million new ones.
• Paradigm Shift: Job losses are seen as a fear far from reality, suggesting a paradigm shift similar
to the Industrial Revolution, which ultimately generated many new opportunities.
Jobs that will grow with AI:
• Creative jobs: Performed by artists, doctors, and scientists; AI will refine and advance these jobs,
making them less complex for humans and enhancing efficiency.
• Management jobs: Cannot be replaced by artificial managers, as human managers are still
required to control them and handle complex tasks involving emotions, ethics, and communication.
• Tech jobs: Programmers, data scientists, data analysts, big data engineers, and those working on
the creation and maintenance of AI systems will always be in high demand.
• New Job Titles (Expected by 2030): Chief Bias Officer, Data detective, Man–Machine teaming
manager, AI business development manager, AI assisted medical professional, AI tutor.

FIGURE 8.3 Effect of AI on jobs

Impact of AI on Key Sectors:
1. Medical: AI is used to create predictive healthcare models, assist physicians in diagnosis, and
eliminate disease/poverty.
2. Automotive: AI enables autonomous vehicles, navigation, and manufacturing.
3. Cybersecurity: AI/ML provide powerful tools to identify and predict threats, essential after cyber
frauds rose by Chief Bias Officer, Data detective, Man–Machine teaming manager, AI business
development manager, AI assisted medical professional, AI tutor.

FIGURE 8.4 AI Tools can be used for Face Swapping and Face Manipulation

6
Impact of AI on Key Sectors:
1. Medical: AI is used to create predictive healthcare models, assist physicians in diagnosis, and
eliminate disease/poverty.
2. Automotive: AI enables autonomous vehicles, navigation, and manufacturing.
3. Cybersecurity: AI/ML provide powerful tools to identify and predict threats, essential after cyber
frauds rose by 600% during the pandemic. AI also processes large financial data to catch instances
of fraud.
4. E-commerce: AI designs chatbots, recommends products, facilitates image-based targeted
advertising, and assists in warehouse/inventory automation.
5. HR applications: AI tools can reject 75% of resumes that do not fit the current context, saving
significant time for HR managers (who might spend up to 23 hours looking over resumes for one
successful hire).
6. Legal profession: AI supports services like document handling, classification, discovery,
summarization, comparison, knowledge extraction, and management for efficiency and accuracy.
4. Privacy and Data Erosion (8.1.2)
• The use of IoT or AI systems generates massive amounts of data; approximately 2.5 quintillion
bytes of data are added each day.
• 90% of digital data has been created in the last two years.
• This collection erodes privacy, as users have no means to know what data about them is used by
whom or for what purpose.
• Cameras fitted with facial recognition software can capture details like gender, age, ethnicity,
gesture, and state of mind. (Example: Chinese police officers wearing facial recognition glasses with
a database of 'suspects').
• Google collects extensive data on Android phones, including contact lists, location, email content,
and photographs, in lieu of free services, posing a big risk to privacy.
5. Unreliability and Fake Content (8.1.2)
• Smart systems contribute to fake news and filter bubbles.
• AI can create faces, compose texts, produce tweets, manipulate images, and clone voices.
• Open-source software Deepfake can easily stick pictures of faces on moving video footage. This is
used for Faceswap video blackmailing of celebrities and normal citizens, often creating
pornographic or manipulated images/videos.
• Fake or manipulated videos are produced at high speed in politics to influence masses. (Example:
Cambridge Analytica accessing 87 million Facebook profiles to campaign for President Trump).
D. AI and Bias (8.1.4, 8.1.5)
An ideal AI system must not be biased. AI bias occurs when an algorithm produces results systematically
prejudiced toward a particular group of people, thus producing skewed output.
1. Sources of Bias
1. Data: AI systems are only as good as the data put into them. Adding biased or skewed data can
never give fair results (GIGO concept). (Example: Voice assistants failing with Indian accents
because they were trained on recorded speech from white, upper-middle-class Americans).
2. Algorithm: Algorithms can further amplify biases injected by skewed data. AI algorithms must
maximize accuracy, but if training data is biased (e.g., showing more female teachers), the
algorithm may decide that all people teaching are women, leading to gender bias.

7
 3. People: Engineers, scientists, and developers are a major source of bias because humans have
conscious or unconscious preferences that may go undiscovered. It is rightly said that ethics and
bias are not the problem of the machine but that of the humans behind the machine.
2. Real-World Bias Examples
• Microsoft's chatbot Tay was silenced within 24 hours in 2016 after trolls taught it negative words
spreading hatred and generating repulsive, toxic responses.
• An algorithm used in US hospitals to predict which patients needed extra care heavily favoured
white patients over black patients because white patients paid more bills and availed extra medical
facilities for the same problems.
• Amazon’s hiring algorithm was found to be biased against women, probably because the system
saw a greater number of male applicants than women in past data.
• Algorithm-based 'emotional AI' hiring platforms have reportedly disproportionately rejected women
applicants based on years of biased data in male-dominated sectors.
3. Mitigation and Regulatory Response
• Diverse Teams: To ensure fairness, organizations must build diverse teams (data scientists,
business leaders, lawyers, accountants, sociologists, ethicists). This helps reduce the potential risk
of bias "falling through the cracks". The team should continually analyze data and algorithms to
ensure fairness.
• Tools: Tools like Bias Analyzer automate the process of analyzing data/algorithms and show the
costs and benefits associated with mitigation actions.
• Regulatory Ban (EU): The European Commission launched a legal framework in April 2021 that
will immediately ban AI systems that are considered a threat to people's safety, livelihood, and
rights, including systems that manipulate human behavior or circumvent users’ free will.
• Organizational Engagement: Organizations are building trust through cultural and educational
programs, risk assessments, and third-party audit programs to move towards a prevention,
detection, and response framework.
• Civil Rights/Activist Groups: These groups must challenge discourse and amplify the voices of
those most adversely affected, asking if the technology is necessary or if alternatives exist.
• Right to Explanation (GDPR): The EU's General Data Protection Regulation (GDPR) empowers
consumers with the right to get an explanation for any algorithm-based decision (like loan
acceptance/rejection). This is a big step in delivering responsible AI and is expected to create
75,000 new jobs.
• Data Privacy Measures:
1. Encrypting the data.
2. Ensuring the AI algorithm cannot learn from outside its dataset.
3. Using secure computation so that developers/workers cannot see or access the data.
4. Restricting reverse engineering the AI model so no user can access data used for model
training.
F. Explainability and Black Box AI
• Explainability is a necessary feature of a responsible AI system (Figure 8.2).
• The ability to explain how a decision was taken is a potential stumbling block in strictly regulated
industries (e.g., financial institutions needing to explain credit-issuing decisions).
• Decisions based on subtle correlations between thousands of variables make explanations difficult.
• An AI program that is not able to explain decisions is known as black box AI.
8
II. AI as a Service (AIaaS)
AI as a Service (AIaaS) is a model where companies use off-the-shelf AI tools to implement and scale AI
techniques at a very low price compared to building a complete in-house AI system. AI services utilize
cloud computing for delivery.
A. Factors Triggering Growth of AIaaS (8.2.1)
Growth is driven by advancements in IT and business motivations:
1. Cloud Platform Availability: Affordable options for enterprise data management.
2. Data Storage: Technologies have improved, becoming cheaper and more reliable.
3. Massive Data Generation: Streaming devices and IoT generate data that can be analyzed for
competitive advantage.
4. Automated Products: Availability of semi or fully automated data management, analytics, and BI
products.
5. Expensive Infrastructure: Companies avoid huge investments.
6. Lack of Trained Programmers: AIaaS fills this skills gap.
7. High Programmer Charges: Avoids high charges for implementing AI.
8. Insufficient Data: Many companies lack sufficient data to analyze in-house.
B. Advantages and Goals of AIaaS (8.2.2)
AIaaS allows small companies to harness machine learning power at significantly lower costs.
Organizations can achieve the following goals:
1. Focus on their core area rather than becoming experts in machine learning.
2. Reduce operational and infrastructural costs significantly.
3. Minimize the risk of investment.
4. Short deployment time for solutions.
5. Pre-built algorithms make complex data management accessible.
6. Availability of on-demand services provides a competitive edge.
7. Increase benefits obtained by analyzing data trends.
8. Develop cost-effective, flexible, and transparent solutions.
9. Improve strategic flexibility: Companies only pay for what they actually use, offering huge savings
since machine learning may only be used in short spurts.
10. Scalability: Companies can start with smaller projects and scale up or down their demand for
services based on changing project needs.

9
 FIGURE 8.5 Emerging Areas in IT

C. Challenges and Accountability of AIaaS (8.2.3)

Representation of FIGURE 8.6: Shortcomings of AI as a Service

Shortcoming Description

Companies must share data with third-party vendors over cloud. Requires strong
Reduced Security
security techniques for storage, transmission, and access avoidance.

Reliance Organization becomes dependent on third parties, potentially delaying response.

The process works like a black box; the organization has no idea about the inner
Reduced
workings (like algorithms used). This may cause confusion regarding data stability or
Transparency
output.

Organizational data policy may restrict data storage on a third-party cloud, preventing
Data Governance
full advantage of AIaaS.

10
D. Vendors and Architecture
• Major Vendors: Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP)
offer various bots, APIs, and machine learning frameworks.
• Selection Criteria: Organizations choose a service based on their goals, business size, budget,
technical capabilities of in-house teams, and data volume.
• Specific Offerings:
o MonkeyLearn: AI platform simplifying text analysis using no-code tools; allows use of pre-
trained models (e.g., survey analyzer) or custom models (sentiment, keywords).
o IBM Watson: Hosts a suite of AI tools, offering pre-built applications like Watson Assistant
and Watson Natural Language Understand, and the Watson Studio to build, train, and
deploy models across any cloud.
o Microsoft Azure: Public cloud platform offering Azure Cognitive Services (computer vision,
text extraction APIs) and Azure Bot Service (incorporating bots).
o Google Cloud ML: Platform that facilitates data scientists and developers working with big
data to create and deploy ML projects. Uses AutoML extensively to train custom models for
text analysis, image classification, and translation.
• AIPaaS (AI Platform-as-a-Service): An end-to-end cloud solution allowing business organizations
to utilize services on a pay-per-use or pay-per-service basis. Provides a complete, intelligent data
management platform via third-party APIs.

11