introduction to causal network analysis and betty’s brain

Post on 23-Mar-2016

56 Views

Category:

Documents

2 Downloads

Preview:

Click to see full reader

DESCRIPTION

Introduction to causal network analysis and Betty’s Brain. Kurt Vanlehn CPI 494, March 17, 2009. Causal networks: Structure. Node’s represent factors, variables E.g., CO2 in a stream; amount of algae in a stream Links represent causation or correlation - PowerPoint PPT Presentation

TRANSCRIPT

Introduction to causal network analysis and Betty’s Brain

Kurt VanlehnCPI 494, March 17, 2009

Causal networks: Structure

• Node’s represent factors, variables– E.g., CO2 in a stream; amount of algae in a stream

• Links represent causation or correlation– E.g., The more C02 in the stream, the more algae– + or – as labels

CO2 in stream

Algae in stream

+

Causal networks: Behavior

• Specify a change in one (or more) variable– Qualitative: it increases or it decreases

• Specify (or assume) that other variables do not change

• Infer which other variables change

example

• If fertilizer in the stream increases, what happens to the CO2 in the stream?

CO2 in stream

Algae in stream

Fertilizer in stream

+-

Fish population

-

Causal networks: Function in science curriculum

• Students are given texts and other resources on a science topic, e.g., stream ecology

• Students draw a causal network that models the physical system

• Repeat this for every topic in the curriculum – Carbon cycle– Photosynthesis– Plate tectonics– Etc.

Try it yourself on scratch paper• Normally, sunlight is absorbed by the surface of the earth,

creating heat. However, when the surface of the earth is covered in snow, most of the sunlight reflects back and is not absorbed.

• The surface heat radiates back in to space. However, greenhouse gases (water vapor, CO2 and methane), block the radiation of heat, so the earth doesn’t cool off as fast. If there is too little greenhouse gas, the heat gets too cold. If there is too much, the earth gets too hot.

• As the earth gets colder, water vapor precipitates as snow. On the other hand, as the earth gets hotter, snow melts, exposing the earth and releasing water vapor.

Learner’s problems

• There’s a lot of text: which phrases are nodes?• What a cause?• Direct vs. indirect causation?• What can be ignored?• What generalization? Level of detail.• Types of arrows: flow of material, flow of entery,

causation, type/subtype, math proportionality…• Positive or negative arrows

Teachers’ problems• If it doesn’t count in the grade, it doesn’t get done.• Different models may all be correct

– Level of detail– Names in the nodes

• Supplying feedback. Varies per model. May need to talk to kid to figure out mind bug

• Cheating• Engaging the students somehow• Prior knowledge; skills; e.g. Reading• User interface training• DECOMPOSITION!!

Design an answer-based tutoring system (on scratch paper)

• What do students receive?• What do students enter as an answer?• What feedback/hints does the tutor give?– On the answer– When students ask for help

• Is your design technically feasible?– How does it work?

What do students receive?

• Intro • Motivation; goals for exercise• Lesson on content e.g., greenhouse effect• Lesson on user interface; editing; meaning of

notation• Worked example: simple problem + causal

network + “not the only solution…”• Can go back to the above

Alternative intro training

• Give students a problem statement & a graph• Have them explain it• Have them debug or extend it• Self-explanation!

What do students enter?Maria’s matrix• List nodes & links• Labels on links• Perhaps a matrix with + or – in some cells• Allows somewhat more variablesAndre’s• Draw nodes & links • Force labeling• Then perhaps fade• M akes loop somewhat easier to seeDarren’s• Drag boxes onto workspace• Green arrows vs. red arros for + vs. –

What feedback/hints does tutor give on answer?

• System says what is wrong & how to fix• Put the feedback as “does sunlight really…?”• Wrongness– Wrong +/- label– Missing nodes & missing links– Extra nodes (or vague) & extra links?– Vague names, differentiation needed – Repeated nodes

What feedback/hints does tutor give when students ask for a hint?

• Analyze what they have already; missing…• Can ask them what they are trying to do– Indicate text that they are trying to represetn– Drag the text into the help box

How does it work?

• Compare student’s model to the ideal model• Count model elements…• See if the student’s model produces the same

implications as the ideal model• How to match nodes without forcing them to

use the ideal model’s node names– Kermit’s draging of phrases from text– Lots of possible correct phrases

What are the pedagogical issues with such an answer-based tutor?

Design a step-based tutoring system (on scratch paper)

• What are the steps?• When does the tutor give feedback/hints?• How does it work, technically?

What are the steps?

When does the tutor give feedback/hints?

How does it work, technically?

What are the pedagogical issues of such a step-based tutor?

Betty’s Brain: A teachable agent

• Student (user) can do– Define node or link; delete; modify– Ask question; ask for explanation of answer– Have agent take a test

• Agent (computer) does in response– Edits to the graph: Does nothing– When asked question/explanation: Prints answer/chain– When asked to take test: Does so, and test results (per

question) are shown.

How does Betty’s Brain differ from an answer based-tutoring system?

Game show feature

• Students have agent enter game show instead of test1. the game host poses questions to the agents2. the students choose a wager that their agent will

answer correctly3. the agents answer based on what they have

been taught4. the host reveals the correct answer5. wager points are awarded

top related