A Literature Review On AI fairness Best Practices

Artificial intelligence (AI) refers to computer systems capable of performing complex tasks that historically only a human could do.They are train on vast amountof data which helps machines recognize patterns, make decisions, and improve over time with experience. For example, AI is what powers things like voice assistants (e.g., Siri, Alexa), chatbots (e.g. ChatGPT), and recommendation systems (like Netflix or YouTube suggesting what to watch).

Machine learning (ML) is a branch of AI that teaches computers to learn from data instead of being explicitly programmed. It’s like training a computer with examples so it can figure out patterns, make predictions or decisions on its own , classify information, cluster data points, reduce dimensionality and even generate new content . For example, teaching an app to recognize photos of cats by showing it lots of pictures of cats and other animals.

Fairness in AI and ML systems, means designing systems that make decisions or predictions without discrimination. It ensures that the outcomes are equitable for all individuals or groups, regardless of factors like race, gender, age, or socioeconomic status. For example, a fair AI system for hiring should evaluate candidates based on skills and qualifications, not on irrelevant or biased criteria.

Bias in AI and ML refers to systematic errors or unfairness in a model’s predictions caused by imbalanced data, flawed algorithms, or human assumptions. Bias can lead to unfair outcomes, such as favoring or disadvantaging certain groups.For example:If a model is trained mostly on data about men, it might perform poorly for women.A hiring algorithm might favor candidates from certain schools if the training data reflects past biased hiring practices. Addressing bias is critical to ensure AI systems are fair and reliable.

Natural Language Processing (NLP) is a field of AI that focuses on enabling computers to understand, interpret, generate, and interact with human language. It bridges the gap between human communication and machine understanding by processing text or speech in natural (human) languages, such as English, Spanish, French or Mandarin. Some applications of NLP are Text Analysis, Language Translation, Speech Recognition and Chatbots and Virtual Assistants.

False Positive Rate (FPR) is a performance metric used in statistical models, machine learning, and classification systems to measure the proportion of negative instances that are incorrectly classified as positive.

Formula for False Positive Rate $FPR = False Positives (FP) / False Positives (FP) + True Negatives (TN)$

where , False Positives (FP): These are cases where the model incorrectly predicts a positive outcome for an instance that is actually negative.

True Negatives (TN): These are cases where the model correctly predicts a negative outcome for an instance that is actually negative.

An LLM (Large Language Models) is a type of advanced artificial intelligence (AI) system designed to understand and generate human-like text. These models are typically trained on vast amounts of text data and use deep learning techniques, especially a class of neural networks called transformers, to process and produce language. Some examples of LLMs are

• OpenAI GPT (e.g., GPT-4, GPT-3.5): A series of language models known for natural language understanding and generation.
• Google’s PaLM and Bard: Models focused on reasoning and dialogue.
• Meta’s LLaMA (Large Language Model Meta AI): An open-access LLM.
• Anthropic’s Claude: Another conversational AI model.

The False Negative Rate (FNR) measures the proportion of positive instances that are incorrectly classified as negative. It answers the question: How often does the model fail to detect a positive case?

Formula for False Negative Rate $FNR = False Negatives (FN) / False Negatives (FN) + True Positives (TP)$

The True Positive Rate (TPR), also known as sensitivity or recall, measures the proportion of positive instances that are correctly identified as positive. It answers the question: How good is the model at identifying positive cases?

Formula for True Positive Rate $TPR = True Positives (TP) / True Positives (TP) + False Negatives (FN)$

The True Negative Rate (TNR), also known as specificity, measures the proportion of negative instances that are correctly classified as negative. It answers the question: How good is the model at identifying negative cases?

Formula for True Negative Rate $TNR = True Negatives (TN) / True Negatives (TN) + False Positives (FP)$

Data Science is a multidisciplinary field that focuses on extracting meaningful insights and knowledge from structured and unstructured data using scientific methods, algorithms, processes, and systems. It combines elements of mathematics, statistics, computer science, and domain expertise to analyze data, make predictions, and drive decisions. It is important because it enables organizations to harness data for competitive advantage, it helps in uncovering patterns and trends that would otherwise remain hidden and it provides the basis for automation, decision-making, and innovation. Data science is applied in fields like healthcare (Predicting diseases, personalizing treatments, and optimizing healthcare resources) , business (Enhancing customer experience, demand forecasting, and fraud detection.), Finance (Risk assessment, algorithmic trading and credit scoring.), Technology (Recommender systems, natural language processing (NLP), and computer vision.) and Social Sciences (Analyzing societal trends, predicting election outcomes, and studying human behavior.)