1.1 Revisiting the AI Project Cycle
Similarly, building an AI project needs a framework — and that framework is called the AI Project Cycle.
🔹 The 6 Stages of the AI Project Cycle
Set the goal for your AI project by clearly stating the problem you wish to solve. Look at various parameters that affect the problem so the picture becomes clearer. Use the 4Ws Problem Canvas — Who, What, Where, Why.
Acquire data from reliable and authentic sources — this data is the base of the project. Sources: surveys, sensors, cameras, CSV files, APIs, online datasets. The data helps you understand the parameters that relate to the problem.
Convert the data into visual representations — graphs, databases, flow charts, maps — so patterns become easy to interpret. Exploring patterns helps you decide which kind of model to build.
Decide and research several models that can achieve the goal. Select the most suitable one. Develop your algorithm around the chosen model. Modeling answers: how should the machine learn to solve this?
Test the model on some newly fetched data (data it has never seen). The results help you judge how accurate the model is, and where it needs improvement.
After evaluation, integrate the model into a real-world application so users and stakeholders can actually use it — a mobile app, a website, a hospital system, a self-driving car, etc.
1.2 Introduction to AI Domains
1. Statistical Data
Works on numerical / tabular data. Extracts meaning from large datasets to make decisions.2. Computer Vision (CV)
Works on images & videos. Teaches machines to "see" and make sense of visual information.3. Natural Language Processing (NLP)
Works on text & speech. Helps computers read, understand and generate human language.📊 1. Statistical Data Domain
🔹 Real-Life Applications of Statistical Data
- Price Comparison Websites — PriceGrabber, PriceRunner, Junglee, Shopzilla, DealTime. They compare product prices across many vendors in one place. Common in technology, hospitality, automobiles, apparel.
- Weather Forecasting — IMD and private weather apps use historical data + real-time readings to predict temperature, rain and storms.
- Stock Market Analysis — AI analyses price, volume, news and sentiment data to predict market trends.
- Recommendation Systems — Amazon, Flipkart, Netflix and YouTube recommendations based on user statistics.
- Credit Scoring — Banks use statistical data to decide loan eligibility.
- Sports Analytics — Predicting match outcomes, player performance, fantasy cricket.
👁️ 2. Computer Vision (CV) Domain
In CV, the input can be photographs, videos and pictures from thermal or infrared sensors. CV projects translate digital visual data into descriptions, then into computer-readable language for decision-making. The main objective is to teach machines to collect information from pixels.
🔹 Real-Life Applications of Computer Vision
💬 3. Natural Language Processing (NLP) Domain
🔹 Real-Life Applications of NLP
1.3 Frameworks & Ethical Frameworks
🧭 1. What is a Framework?
⚖️ 2. What is an Ethical Framework?
Ethical Framework = a framework that ensures the choices we make do not cause unintended harm.
Ethical frameworks provide a systematic approach to navigating complex moral dilemmas by considering different ethical principles and perspectives. Using ethical frameworks, individuals and organisations make well-informed decisions that align with their values and promote positive outcomes for all stakeholders.
🤔 3. Why Do We Need Ethical Frameworks for AI?
🔹 Reasons Ethical Frameworks for AI are Needed
- Avoid Bias — example: a hiring algorithm that was biased against women applicants.
- Prevent Discrimination — AI must not treat people unfairly based on race, gender, religion or region.
- Ensure Transparency — users should understand how the AI reached its decision.
- Protect Privacy — personal data must not be misused.
- Ensure Accountability — when AI goes wrong, someone must be answerable.
- Build Trust — people only accept AI in healthcare, law and finance if they trust it.
- Catch Problems Before They Happen — ethical frameworks help us avoid unintended outcomes before they take place.
1.4 Activity — "My Goodness"
What it is: An interactive game where players make 10 decisions about how they would like to donate to charity. In most cases, players see details about the recipient and the intended use of the money. In a few cases, info is hidden but can be revealed.
Goal: To understand how our decisions get influenced by our personal morals, values and ethics — and to uncover potential biases within us!
🔹 Hidden Factors That Influence Decisions
- Identity of the charity recipient.
- Location of the recipient.
- Bias towards relatives or known groups.
- Uncovered information (available but we ignore).
1.5 Factors That Influence Our Decision-Making
Knowingly or unknowingly, many factors shape our decisions. Any ethical framework must take these into account:
1.6 Types of Ethical Frameworks
Ethical frameworks for AI can be categorised into two main types:
🏭 1. Sector-Based Frameworks
Examples of sector-based frameworks:
- Bioethics — Healthcare and life sciences (patient privacy, data security, AI in medical decisions).
- Finance Ethics — Banking, insurance, credit decisions.
- Education Ethics — AI in classrooms, student data, automated grading.
- Transportation Ethics — Self-driving cars, traffic AI.
- Agriculture Ethics — AI for crop/livestock decisions.
- Governance & Law-Enforcement Ethics — Face recognition, predictive policing.
💎 2. Value-Based Frameworks
Value-based frameworks are further classified into three categories:
Prioritises protection of human rights and dignity. Values human life over other considerations. Emphasises autonomy, dignity and freedoms. In AI: ensure systems do not violate human rights or discriminate.
Evaluates actions by maximising overall utility / good. Seeks outcomes that offer the greatest benefit for the greatest number of people and minimise harm. In AI: weigh benefits of AI against risks like job displacement or privacy.
Focuses on the character and intentions of the people involved in decision-making. Asks whether actions align with virtuous principles — honesty, compassion, integrity. In AI: do developers, users and regulators uphold ethical values throughout the AI lifecycle?
1.7 Bioethics — A Sector-Based Framework in Healthcare
🔹 The 4 Principles of Bioethics
Enable users to be fully aware of decision-making. Users of an AI algorithm should know how it functions. Data on which models are trained — and used to make decisions — should be reproducible and accessible to patients. If there are performance concerns, model predictions and data labels should be released.
Harm to anyone (human or non-human) must be avoided at all costs. If no choice is available, the path of least harm must always be chosen. AI algorithms must be trained on data that equitably reduces harm for all, not just for some groups. Promote well-being, minimise harm, distribute benefits and harms justly.
Not only should we avoid harm — our actions must provide the maximum benefit possible. The solution should be held to clinical practice standards, not merely technological ethics. It should go beyond nonmaleficence and strive for beneficence. Use training data that reflects healthcare needs and outcomes of patients of all races; the data must be unbiased.
All benefits and burdens of a choice must be distributed in a justified manner across people, irrespective of background. Solution development requires deep knowledge of social structures that result in racism, sexism and other societal biases. The solution must be aware of social determinants of healthcare and actively work against those structures.
🔹 Key Bioethical Terms
| Term | Meaning |
|---|---|
| Autonomy | Freedom to make one's own decisions with full information. |
| Non-Maleficence | Ethical principle of avoiding causing harm or negative consequences. Emphasises minimising harm as much as possible. |
| Maleficence | The concept of intentionally causing harm or wrongdoing. The opposite of non-maleficence. |
| Beneficence | Ethical principle of promoting and maximising well-being. Emphasises actions that produce positive outcomes for all stakeholders. |
| Justice | Fair distribution of benefits and burdens across all people. |
1.8 Case Study — Bioethics Applied to Healthcare AI
🔹 Why the Problem Happened
- The algorithm was trained on healthcare expense data as a measure of health — instead of actual physical illness.
- It was created in the United States where less money is spent on Western-region patient healthcare than on other ethnic patient healthcare.
- So lower expenses wrongly signalled lower need — when actually those patients were simply being under-served financially.
🔹 How the 4 Bioethical Principles Could Have Prevented It
Patients would have known how the algorithm works and what data trained it — so they could flag issues early.
If followed, the algorithm would never have been allowed to send more intensive care to less-ill patients while denying it to Western-region patients who were actually sicker.
A better dataset — one that reflects healthcare outcomes of patients of all races — would have ensured everyone received maximum possible benefit.
The developers would have consciously worked against social structures (racism, unequal healthcare spending) and considered social determinants while designing the algorithm.
1.9 Applying Ethical Frameworks — Quick Checklist
For every AI project you build, ask yourself these questions at each stage:
| Stage | Ethical Question |
|---|---|
| 1. Problem Scoping | Does the problem definition include all affected groups? Or are we scoping only for the majority? |
| 2. Data Acquisition | Is the data representative of all users? Are any communities missing from the data? |
| 3. Data Exploration | Do we see any bias in the patterns — or gaps that favour one group? |
| 4. Modeling | Is the model simple and explainable enough for users to trust? |
| 5. Evaluation | Have we tested for fairness across groups — not just overall accuracy? |
| 6. Deployment | Who is accountable if the model causes harm? Is there a feedback & takedown mechanism? |
Quick Revision — Key Points to Remember
- AI Project Cycle = 6-stage cyclical process to build any AI project.
- 6 Stages: Problem Scoping → Data Acquisition → Data Exploration → Modeling → Evaluation → Deployment.
- AI is iterative — go back if evaluation shows problems.
- Three Domains of AI: Statistical Data · Computer Vision · Natural Language Processing — based on type of data fed in.
- Statistical Data — numerical/tabular data. Examples: Price Comparison (Junglee, PriceGrabber), weather, stock market, recommendations, credit scoring, sports analytics.
- Computer Vision — images & videos. Steps: acquire → screen → analyse → identify → extract. Examples: agricultural monitoring (drones), surveillance, medical imaging, self-driving cars, face filters.
- NLP — text & speech. Examples: email filters (spam), machine translation (Google Translate), voice assistants, chatbots, sentiment analysis, auto-complete.
- Framework = set of steps for solving problems in an organised way.
- Ethical Framework = framework ensuring choices do not cause unintended harm.
- Why needed for AI: avoid bias, prevent discrimination, ensure transparency, protect privacy, ensure accountability, build trust, catch problems early.
- "My Goodness" activity — reveals personal biases in decision-making (my-goodness.net).
- Factors influencing decisions: Culture · Religion · Intuition & Values · Laws · Family/Society · Self-Interest.
- Types of Ethical Frameworks: Sector-based (Bioethics, Finance, Education, Transport, etc.) & Value-based.
- Value-based subtypes: Rights-Based (human rights) · Utility-Based (greatest good) · Virtue-Based (character & honesty).
- Bioethics — for healthcare & life sciences.
- 4 Principles of Bioethics: Respect for Autonomy · Do Not Harm (Non-maleficence) · Maximum Benefit (Beneficence) · Justice.
- Case study: US hospital AI was biased against Western-region patients — trained on healthcare expense data instead of actual illness. Bioethics could have prevented it.
- Key terms: Autonomy · Non-maleficence (avoid harm) · Maleficence (cause harm) · Beneficence (maximise well-being) · Justice (fair distribution).