Artificial Intelligence Heidi Newton Peter Andreae 2 Overview Talk about three of the topics from the online book AI with language: Chatterbots AI for selling stuff: Association rules AI for games: Basic game search Peter Andreae
3 AI with language Using language is one of the distinguishing features of human intelligence. Making computers use language well is hard! We have not solved this problem yet Would it be useful to have computers that could use human language? Where? Peter Andreae
Chatterbots 4 What are they? general fixed vs domain specific vs learning Lets try some: Eliza: http://nlp-addiction.com/eliza/
ALICE: http://www.pandorabots.com/pandora/talk? botid=f5d922d97e345aa1 Cleverbot: http://www.cleverbot.com/ Peter Andreae Chatterbots 5 Which was best?
Were they any good? How could you tell it wasnt a person? What were its limitations? Peter Andreae 6 Turing Test Turing test: the most famous test for successful artificial intelligence Based on computers using language A Human and a Computer, each try to convince a
judge that they are the human, communicating only over a typed chat session If the judge is fooled, then the computer must be intelligent. Book describes an activity for students to run a Turing test using one of the chatterbots Peter Andreae 7 2. Association Rules What does Amazon (and many online retailers) do
when you click on an item? How do supermarkets decide where to place items? Note: Recommender Systems are closely related. Peter Andreae 8 Association Rule Learning Machine learning algorithms look for patterns in purchasing data. If 30% of the people who buy an LED torch also buy rechargable batteries, then Whenever a person selects an LED torch, suggest that
they might want to buy rechargable batteries. Should you recommend torches if they buy batteries? If 20% of people who buy corn chips also buy salsa dip, then put salsa dip on a shelf near the corn chips to increase sales If 80% of the people who buy milk also buy bread, then put them on opposite sides of the store to make people pass by as many other shelves as possible Peter Andreae 9 Association Rule Learning
How do you find the associations? How do you tell if they are strong enough We are planning an activity for the book involving searching for associations in supermarket reciepts Peter Andreae 10 3. AI for games Lots of different applications of AI in games Games were part of AI research from the beginning Modern computer games need even more AI. Min-Max search is a basic AI game playing strategy
Applicable to 2-person board games Fundamental to the AI chess playing systems (eg Deep Blue) Paper and pencil activities to explore this. Peter Andreae 11 Min Max search for Os & Xs Suppose it is Os turn and the board looks like this What should O do? Peter Andreae
Its Os turn: what should O do? Os choices Xs Choices lose lose lose Os choices
win lose lose lose win Xs Choices loselose lose
lose lose Search forwards to explore all the options Peter Andreae Its Os turn: what should O do? Os choice Xs Choice Os choice
Xs Choice Work backwards to determine status of earlier positions X will try to make O lose Peter Andreae Its Os turn: what should O do? Os choices Xs Choice Os choice
Work backwards to determine status of earlier positions O will try to win Peter Andreae Its Os turn: what should O do? Os choice Xs Choices Work backwards to determine status of earlier positions X will try to make O lose Peter Andreae
Its Os turn: what should O do? Os choice Work backwards to determine status of earlier positions O will try to win Peter Andreae More Min Max Search What happens if you cant search all the way to the end? Have to stop and measure how good the board looks. Eg, a piece count Work backwards with the minimum score (for opponents turn)
maximum score (for players turn) Peter Andreae Blacks turn 0 advantage +1 +1 +2
Peter Andreae Blacks turn 0 advantage +1 +1 +2 0 Peter Andreae
