Introduction

Artificial Intelligence is categorized into four main approaches based on whether the goal is to mimic humans or to be rational, and whether the focus is on thought processes or behavior

• Thinking Like Humans: This falls under Cognitive Science. The goal is to mimic human thought, but it faces counter-arguments: human intelligence is not necessarily the pinnacle of intelligence, and mimicking it can introduce human biases into agents

• Thinking Rationally: This approach uses logical rules and a systematic framework to store facts and arrive at inferences. A limitation is that it often discounts reflexes; for example, a rational system might calculate the danger of a hot pan before moving, whereas a reflex is immediate

• Acting Like Humans: This is often associated with the Turing Test, where a machine might deliberately make spelling mistakes to appear more human, making it difficult to distinguish from human work

• Acting Rationally: This focuses on "doing the right thing" through a mix of logical and reflexive thinking

◦ Weak AI: The definition that a machine should merely act intelligently

◦ Strong AI: The definition that a machine must both act and think intelligently

2. Historical Milestones in AI

• 1950: Alan Turing poses the foundational question, "Can machines think?"

• 1956 (Dartmouth Conference): A pivotal event where the "Logic Theorist" (a theorem prover) was presented by Herbert Simon and Alan Newell

◦ Optimism: Simon and Newell predicted an AI would be a chess champion within 10 years

◦ Skepticism: Chess master David Levy challenged that no machine could beat him; while he initially succeeded, he eventually lost to the AI "Deep Thought"

• 1959-1965: Early successes included Gelernter's Geometry theorem prover and the General Problem Solver (GPS), which aimed to mimic human thought processes using goal-directed search

• 1997: IBM’s Deep Blue defeated Grand Master Gary Kasparov in chess

3. The "AI Winter" Periods

The field experienced two major periods of decline due to unmet expectations:

• First AI Winter (1974–1980): Caused by failures in Machine Translation (e.g., issues with idioms and sentence parsing) and the discovery of limitations in the Perceptron model regarding non-linear problems

. In 1966, the ALPAC report suggested reducing funding for translation research

• Second AI Winter (1987–1993): Marked by the decline of the LISP programming language and a decrease in specialized hardware for expert systems

4. Expert Systems & Knowledge Representation

• Languages & Tools: LISP was developed for arbitrary data structures, followed by the first planner (STRIPS) in 1971, PROLOG in 1972, and representational schemes like Frames and Semantic Nets

• Key Expert Systems:

◦ Dendral: Helped chemists identify molecular structures from mass spectrograms

◦ MYCIN: Used for medical diagnosis but faced ethical concerns that prevented its implementation

• Bottleneck: A major issue for these systems was the difficulty in acquiring knowledge from domain experts who were sometimes unable or unwilling to share their expertise

5. Modern AI (Post-2000)

Current AI success is driven by a "trijunction" of Data, Computation Power, and Algorithms

• Key Developments: The rise of Big Data, Machine Learning, and Deep Learning using GPUs

• Generative AI: The current "3rd wave" includes models like AlphaFold, GPT-3, GPT-4, and ChatGPT

• Applications: AI is now applied in healthcare, finance, social media, entertainment (recommendation engines), and creative fields like Style Transfer

• Ethical Issues: Modern applications like ChatGPT face limitations and potential ethical concerns regarding the information they generate