Turing test
Checks whether a machine’s behavior is indistinguishable from a human in conversation. It tests human-like intelligence not consciousness
Searle’s Chinese Room Argument
A man follows rules to manipulate Chinese symbols and output correct answers but doesn’t understand the language. Therefore, a computer can simulate understanding by following rules, but does not truly understand. So machines today are not conscious or intelligent in the human sense
Eliza(1966) and why was it important
ELIZA mimicked a therapist using pattern matching and canned responses. It showed that computers can appear to understand language without actually understanding.
SHRDLU
A program that manipulated blocks based on natural language commands. Showed early symbolic AI and reasoning in a limited world.
Deep Blue (1997)
Beat world chess champion Gary Kasparov
Used search trees and evaluation functions, not learning or intuition
Search tree in simple terms
A branching structure of every possible future move. You evaluate win/tie/loss at the bottom(leaves) and pick the best move using minimax
minimax
Algorithm that chooses the move with the best guaranteed outcome assuming the opponent plays perfectly
Supervised learning
Learning from labeled data to categorize new data. Goal: find a boundary/line/plane that separates classes
Unsupervised learning
Learning without labels.
Groups data into natural clusters based similarity.
Reinforcement learning
Learning by rewards and penalties. Requires planning ahead for long-term
Feature vector
A list of numbers representing the features of a data point (like size, shape, color)
How do LLMs learn
They trained on massive text dataset and learn statistical patterns between words. They predict the next token, not by understanding, but pattern matching at scale
Where does AI training data come from
Wikipedia, stackoverflow, open textbooks,images, audios
AI bias
When AI outputs reflect human biases found training data. Examples: facia; recognition failures, sports stat bias, gender bias in explanations.
why does AI bias
Because humans produce biased data -> algorithms learn it and amplify
Give one real examples of AI bias
Facial recognition failing on darker-skinned women
Environmental impact of AI
Massive electricity for training
Huger water usage
Mining for GPU hardware
E-waste from discarded chip
Digital divide
Unequal access to internet due to geography, income etc. 2.6 billion people lack broadband
AI vs ML
AI = broad goal of simulating intelligence
ML = Methods that let machines learn from data
