Session 26

In Search of Intelligent Systems

810:161

Artificial Intelligence

The Structured Matcher Paper

After reading the paper, do the following:

• Summarize the paper in two sentences or less.
• What are the three most important ideas in the paper?
• Which parts of the paper most need improvement? In what ways?

I would love to hear what you think...

One "big idea" follows from something we learned about when studying learning: Exploit regularities. Structured Matcher describes a regularity in the world:

• a problem that is common in many domains, and
• a solution that is common to a lot of problem solvers.

The Principle of Convergent Intelligence tells us that intelligent agents must exploit regularities in the world.

• They recognize the regularity.
• They use what they know about the regularity to perform better.

The world of problem solving is full of regularities of this sort.

1. Selecting an acquisition structuring plan.
2. Detemining what is wrong with a patient.

What regularities does Structured Matching recognize?

• We often know all possible answers to a question, or nearly all.

• We often know that some answers are more specific varieties of others.

• We use generality and specificity as a way to organize our knowledge about the domain.

One way to try to understand how intelligent agents solve problems is to try to identify what regularities they use--and how they use them.

Now that you know all about AI...

We say that AI has solved a problem if we can build a program to solve the problem or do the task as well as or better than humans.

Construct two lists for me:

• Which activities (or phenomena) that you consider "intelligent" do you think AI will be able to solve?

• Which activities (or phenomena) that you consider "intelligent" do you think AI will not be able to "solve"? Why not?

Because this is the last day of class, please humor me and spend all of the allotted time actually working on the exercise. :-)

A Few More Questions

Are there any problems that you think we should not use AI to study? Why?

Has your opinion of the Turing Test changed at all? Why (not)?

Will a computer ever be conscious in the way humans are?
Will a computer ever have emotions in the way humans do?

Who will be legally responsible when the first intelligent program commits a crime?

The AI Enterprise

Artificial intelligence is the computational study of how a system can perceive, reason, and act in complex environments.

• agents that solve problems

• agents that reason logically

• agents that plan their actions

• agents that can handle uncertainty

• agents that use knowledge of world and self

• agents that learn
  • from observation
  • from experience
  • using existing knowledge

• agents that communicate

• agents that use physical effectors

What Were the Big Ideas You Encountered?

Here is one:

The world manifests constraints and regularities. If an agent is to exhibit intelligence, it must exploit these constraints and regularities, no matter the nature of the agent's physical make-up.

How many more can you add to the list? What makes them big ideas?

Views of AI

There are at least three ways to view AI:

• as a set of programming techniques

• as a means for simulating intelligence

• as a means for studying intelligence

At the simplest level, each ultimately relies on the same notion: the concepts of AI constitute a language for describing the phenomena of intelligent behavior.

Progressing through the list, one makes increasingly stronger commitments to what "describe" means in the last sentence.

For the Exam

See the Sessions index page for an outline of the topics we have covered. All are fair game on the exam.

The final exam is Wednesday, December 19, at 1:00 PM. Good luck!