Understanding AI: Anthropomorphism, History, and ML

• Anthropomorphism is when human characteristics, emotions, or intentions are attributed to non-human entities

Artificial Intelligence (AI)

• AI enables machines to learn, adapt to new data, and mimic human behavior and abilities
• Early AI systems were rule-based or expert systems that relied on programmed rules created by experts
• Advancements in AI are due to data availability, computing power, storage, algorithm improvements, accessibility, and abstraction

Milestones in AI

• Alan Turing introduced the Turing Test to assess if machines can think
• John McCarthy coined the term "AI"
• Neural networks gained widespread use
• IBM's Deep Blue defeated Garry Kasparov in chess
• IBM's Watson beat Jeopardy champions
• DeepMind's AlphaGo beat world champion Go player Lee Sodol
• ChatGPT was publicly released

Machine Learning

• Machine Learning uses algorithms and statistical models enabling computers to learn and make predictions without explicit programming
• It involves feeding machines data to identify patterns, correlations, and relationships
• Three types of Machine Learning exists: Supervised, Unsupervised, and Reinforcement Learning

Supervised Machine Learning

• Relies on labeled examples (training data)
• Humans are needed to supply the labels
• Used for prediction tasks

Unsupervised Machine Learning

• Discovers patterns and relationships in unlabeled data
• It is used when no human guidance is needed
• Used to uncover clusters or insights without preconceived notions
• Used for tasks like Netflix movie recommendations

Reinforcement Learning

• Learns from trial and error, with zero initial knowledge
• Goals are predefined
• Decision-making improves with feedback over time
• Strategy is derived from experience
• No fixed training dataset required
• Commonly used in games, robotics, and autonomous vehicles

Two Approaches to Conversational Agent Design

• Deterministic approach
• Generative approach

Deterministic Response

• These are consistent and predictable
• They are accurate if the user's intent is correctly identified
• Insensible responses are given if the user's intent is misidentified
• Uses structured flow
• Domain-specific
• Fewer computational resources used

Generative Response

• LLMs produce probabilistic and non-repeatable responses
• The risk of hallucination is present
• More robust understanding of user intent leads to more sensible responses
• Dynamic and less structured flow
• Higher versatility across domains
• Significant computational resources used

Retrieval Augmented Generation (RAG)

• RAG involves providing extra context/info to the LLM, to provide more accurate and up-to-date information to the LLM instead of directly asking it to answer the question
• The process consists of:
• loading documents
• splitting documents into chunks
• converting chunks into embeddings
• retrieving relevant chunks
• defining a prompt template

Multi-agent LLMs

• Involves integrating multiple LLMs with specific specializations and purposes to provide a better agent
• Design of multiple agents and its integration increases complexity
• Multiple agents allows for higher level of specialization across domains or tasks
• Coordination between multiple agents may lead to higher variability in responses

Three Communication Models

Transmission

• Linear, a one-way process
• No feedback
• Focuses on the sender
• Barriers such as noise affect it (environmental, semantic)

Interaction

• Dynamic, two-way
• Alternate positions between sender and receiver
• Incorporates feedback
• Takes physical environment and psychological context into account

Transaction

• Humans are communicators who are simultaneously senders and receivers
• Co-creation of meaning
• More robust context (physical/psychological + social, cultural, relational) that can affect its transmission

Language and Meaning

• There is an infinite number of utterances
• This opens up the possibility for miscommunication

Triangle of Meaning

• Thought: The idea a person references

• Symbol: The word representing the thought

• Referent: The object or idea the symbol refers to

• Language is learned; symbols have no meaning until we assign referents to them

Nonverbal Communication Principles

• Conveys interpersonal and emotional messages more
• It forms 65-90% of our meaning from nonverbal signals
• It is more primary and instinctive
• Bypasses cognitive processing
• Provides nuances not expressed by words alone
• More involuntary than verbal
• Not completely involuntary, but more difficult to control or "fake"
• Exposes our underlying thoughts or feelings
• Non-action is still nonverbal communication
• More ambiguous
• A wider spectrum of interpretations
• More credible
• Actions are more trustable than words
• Since it is less easy to fake, it is more honest and credible
• We rely on nonverbal signals when there is a conflict

Machine-to-Machine Communications

• Data is represented using bits, which is the most basic unit of information is used to store, process and communicate
• A group of 8 bits is called a byte and is used to represent characters
• ASCII is a standard 'dictionary' used to identify the symbols of each alphabet

Application Programming Interface (API)

• Set of rules and protocols to allow computer programs to communicate
• Provides a common standard across different languages and frameworks

API Keys

• Used to authenticate a client making requests to an API
• Tracks Usage of API