Approaches to AI
Lays out the four definitions of AI (thinking or acting, humanly or rationally), poses the Winograd schema questions, and sorts AI applications into the four approaches on a shareable board.
- Lecture reference: Introduction to AI · slides 2–18
What is intelligence?
Let’s brainstorm characteristics and components of intelligence. The usual list includes:
- Decision making
- Learning
- Recall/memory
- Logic or rationality
- Communication/language
- Humor
- Creativity
This starting point skews our definition of artificial intelligence…
Dimensions of AI definitions
Lecture reference: Introduction to AI · slide 3| Human | Rational (ideal) | |
|---|---|---|
| Behavior | ||
| Thought |
Rows: behavior or thought. Columns: measured against humans or against an ideal of rationality. Select a cell to show that approach.
Acting humanly
Behavior · Human The Turing test approach
Behave so that an interrogator cannot tell machine from human
The Turing test
Lecture reference: Introduction to AI · slide 5Proposed by Alan Turing in 1950. A machine “passes” the test if a human interrogator cannot tell machine from human.
Would require (at least):
- Natural language processing
- Knowledge representation
- Automated reasoning
- Machine learning
Adds interaction with people and objects. The robot would also need:
- Computer vision and speech recognition/generation
- Robotic manipulations
A better Turing test?
Lecture reference: Introduction to AI · slides 7–9On Our Best Behavior (Levesque, IJCAI 2013). Winograd schema: Multiple choice questions that can be easily answered by people but cannot be answered by computers using “cheap tricks”.
Pick an answer to each question, then reveal it.
The trophy would not fit in the brown suitcase because it was so large.
What was so large?
The sack of potatoes had been placed below the bag of flour, so it had to be moved first.
What had to be moved first?
The large ball crashed right through the table because it was made of steel.
What was made of steel?
Winograd vs Turing
Lecture reference: Introduction to AI · slide 10Advantages over the standard Turing test
- Test can be administered and graded by machine
- Does not depend on human subjectivity
- Does not require ability to generate English sentences
- Questions cannot be evaded using verbal dodges
- Questions can be made “Google-proof” (at least for now…)
Winograd schema challenge
Strong vs. weak AI
Lecture reference: Introduction to AI · slide 11- “Weak” AI
- Computer is limited to being a tool for studying intelligence and developing useful technology.
- “Strong” AI
- Computer could (in principle) be programmed to actually BE a mind, to be intelligent, to understand, perceive, have beliefs, and exhibit other cognitive states normally ascribed to human beings.
John Searle’s Chinese Room thought experiment (1980)
Not on the slide A person who does not understand Chinese follows written rules to answer questions written in Chinese; the answers can look fluent with no understanding of Chinese behind them.
The problem dictates the approach
Brainstorm various AI applications and decide which approach is best for the application… Lecture reference: Introduction to AI · slide 18
Each application has a small 2 × 2 grid laid out like the table: pick a square to place the application in that approach, or drag the application onto a quadrant. Copy link shares the board.
Unplaced
10- Autonomous vehicles
- Legged locomotion
- Autonomous planning and scheduling
- Machine translation
- Speech recognition
- Recommendations
- Game playing
- Image understanding
- Medicine
- Climate science
Acting humanly
Behavior · Human 0No applications yet.
Acting rationally
Behavior · Rational (ideal) 0No applications yet.
Thinking humanly
Thought · Human 0No applications yet.
Thinking rationally
Thought · Rational (ideal) 0No applications yet.