Intro to AI Ch. 1 - Foundations in AI

Noonoouri

Created 2018. An 18-year-old digital avatar, the first virtual pop star signed to a record deal.

Generative AI

A type of AI that can create a wide variety of data, such as images, videos, audio, text, and 3D models.

It works using neural networks to identify patterns and structures within existing data to generate new and original content.

Examples: Chat-GPT, Noonouri, DALL-E

Artificial Intelligence

The science/engineering of making “intelligent” computer programs/machines.

Exhibit characteristics such as problem understanding, learning, reasoning, predicting, reacting, and planning

Weak AI

AI that seeks to simulate human behavior. The most limited and most common.

AKA “narrow AI”, “artificial narrow intelligence”

Ex. Siri, Alexa

Strong AI

AI that exhibits human-level intelligence.

AKA “general AI”, “artificiel general intelligence”

Ex. Google’s DeepMind

Super AI

AI that far surpasses the abilities of humans minds. Purely speculative at this point.

AKA “artificial superintelligence'“, “super intelligence”

Alan Turing

Lived 1912 - 1954

• Contributed to code breaking in WWII

• Invented the Turing Machine

• Invented the Turing Test

Turing Test

1950

“The Imitation Game”

If a human interrogator cannot reliably tell the machine from the human, the machine is said to have passed the test.

Alternatives:

• Kurzweil-Kapor Test: A computer must hold a two-hour conversation, and 2/3 judges must believe it to be a human being.

• Coffee Test: A robot must be able to go to a stranger’s home, locate the kitchen, and brew a cup of coffee.

When was the term “AI” coined?

1955/1956

The Dartmouth, New Hampshire Summer Workshop

ELIZA

1966

A chatbot that served as a psychoanalyst. A user would type in questions and ELIZA would provide counsel.

Designed by MIT professor Joseph Weizenbaum

Shakey

1966

A general-purpose mobile robot that reasons about its own actions. Limited ability to perceive and model its environment. Could perform tasks that required planning, route-finding, and rearranging simple objects.

Stanford Research Institute

Life Magazine (1970) referred to Shakey as “the first electronic person”

The AI Winter

1970s - 1980s

Cutbacks in AI research funding, due to it not living up to high expectations.

AI advances were mostly only in controlled academic environments

How useful is AI, really?

Advancements limited by computer hardware of the time

Damaging report (1973) by Professor Sir James Lighthill

Deep Blue

1997

The first chess-playing program to win both a game and a match against a reigning world champion under regular time controls

Developed by IBM and a group of CMU graduates

AIBO

1999

A robotic dog designed to be an intelligence and trainable robot companion. Exhibits behavior typical of a dog, and can adapt its behavior based on interactions with humans

Created by Sony

Roomba

2002

First mass produced robotic vacuum cleaner with capability such as identifying walls and avoiding stairs.

Launched by iRobot

Data Science

2008

A field of study that focuses on understanding data. Spans programming, statistics, data mining, machine learning, analytics, business intelligence, data visualization, etc.

Coined 2008 by data team leads from LinkedIn and Facebook

SIRI

2011

Apple’s intelligent virtual assistant, first integrated in the iPhone 4S. Uses voice query interface. Delegates requests to a set of internet services. Can also perform actions.

Watson

2011

Supercomputer capable of answering questions posed in natural language. Won Jeopardy. Created by IBM.

AlexNet

2012

A convolutional neural network that won the ImageNet Large Scale Visual Recognition Challenge. Correctly detect and classify objects and scenes.

Alexa

2014

Amazon’s virtual voice assistant. Can control smart devices within a home.

Eugene

2014

A chatbot that poses as a 13-year-old boy, rumored to have passed the Turing Test with 10/30 judges believing Eugene is human.

TAY - Thinking About You

2016

Released by Microsoft via Twitter

An AI twitter bot designed to mimic the language patterns of a 19-year-old American girl and learn from interacting with human users.

Shut down after 16 hours and 96,000 tweets as Tay began to post inflammatory, racist, and offensive tweets.

An example of weak AI as trolls exploited Tay’s “repeat after me” feature.

AlphaGO

2016 and 2017

A computer program that plays the board game GO, a game a googol times more complex than chess. Beat professional player in 2016 and world champion in 2017.

Developed by DeepMind Technologies, later acquired by Google.

Duplex

2018

A virtual assistant that can make calls for you to make appointments, book reservations, etc.

Passed an informal Turing test.

Big Data Leak

2018

Facebook-Cambridge Analytica data breach. Millions of Facebook users personal data was harvested without consent by Cambridge Analytica. Data used for political advertising. CEO Zuckerberg testified in front of US Congress.

AI Ethics Guidelines

2018

Prepared by the High-Level Expert Group on AI

Set up by European Commission in June 2018

7 key requirements for AI to be deemed trustworthy:

• Human agency and oversight

• Technical robustness and safety

• Privacy and data governance

• Transparency

• Diversity, non-discrimination, and fairness

• Societal and environmental well-being

• Accountability

Human-Centered AI

AI is to enhance humans rather than to replace them.

Steps to Problem Solving

1. Define the problem precisely

2. Isolate and represent the task knowledge that is necessary solve the problem

3. Choose and apply the best problem-solving techniques to the particular problem

Good problem representation…

• Make important objects and relations explicit

• Suppress unimportant irrelevant detail

• Expose natural constraints

• Concise - efficiently describe a given scenario

  • Complete - everything necessary can be described

State

A description of a system at some given point in time.

Ex. start state, goal state

Operator set

Each operator can transform one state into another

8-tile puzzle example: Move the blank spot up, down, left, right

Transition model

Returns the resulting state following the application of an operator to a particular state

Goal test

Checks whether a given state is a goal state

Path cost function

Assigns a numeric cost to a path.

Assume: the cost of a path can be described as the sum of the costs of the individual operator actions along the path

8-tile puzzle example: Each step costs 1 move, so the path cost is the number of steps in the path

Safe/unsafe state

Unsafe states are not allowed to be part of the solution due to the specifications of the problem.

Example: in Farmer Jones River Problem, a state in which the wolf and sheep are left alone together.

Search Problem

Find a path from some start state to some goal state.

State Space

The set of all valid states that can exist in the “problem universe”. Each node corresponds to an individual state. Nodes are connected by transforming operators.

Search Tree

The sequence of paths that can be reached from a given start state. Rooted at a particular start state. Each node corresponds to a sequence of states from the start state to a given state.