Lecture 9 - Mycin

Question 1

What is MYCIN?

Answer 1

An expert system for identifying bacterial infection and appropriate antibiotic treatment

Question 2

When was MYCIN developed?

Answer 2

Early 1970s

Question 3

What were the three “programs” used by MYCIN?

Answer 3

Consultation Program

Explanation Program

Knowledge Aquisition Program

Question 4

What were the two modules and purposes of these modules in the consultation program in MYCIN?

Answer 4

1 Diagnosis - Identify possible organisms

2 Therapy - Select appropriate treatment

Question 5

What was the purpose of the explanation program in MYCIN?

Answer 5

Provide an explanation for the conclusions reached and the questions asked during a consultation

Question 6

What was the purpose of the knowledge acquisition program in MYCIN?

Answer 6

Used to build and modify the rule base

Question 7

How did MYCIN represent most of the entities it reasoned about?

Answer 7

Object-Attribute-Value triples

e.g.

John is 37 would be represented as (John, Age, 37)

Question 8

How many attributes (parameters) did MYCIN use?

Answer 8

65, split into 6 groups depending on the type of object

Question 9

What types of object did MYCIN know about?

Answer 9

Cultures

Drugs

Surgical Procedures

Organisms

Patients

Therapies

Question 10

What sort of things did MYCIN contain in it’s static knowledge about each parameter?

Answer 10

Range of possible values

How to ask the user for its value

Whether the value must be obtained from a lab test

Which rules refer to the parameter in their condition part

Which rules may draw conclusions about the parameter

Question 11

“Context” refers to what in MYCIN?

Answer 11

The object that the system is currently reasoning about

Question 12

Why is context important in Mycin?

Answer 12

It allows rules to be used in more than one situation

Question 13

What does the action part of a rule in Mycin do?

Answer 13

Draw conclusions about values of paramaters of objects

Evidence of a particular strength for or against a particular fact being true

Question 14

What type of interpreter does MYCIN use?

Answer 14

Exhaustive Backward Chaining

So all evidence for and against every possible conclusion is gathered before the final conclusion is made

Question 15

Why are uncertainties in MYCIN not based on rigorous foundations?

Answer 15

Data in science is rarely, if ever, certain

Even if it is certain youd need massive amounts of data to build a system using Bayesian methods

Question 16

How does MYCIN represent uncertainty?

Answer 16

Certainty Factor

Ranges from -1 (certainly false) through 0 (no belief either way) to +1 (certainly true)

Question 17

In MYCIN, how is the certainty of a rule concluded?

Answer 17

C_conclusion = C_premise X C_consequent

certainty of two conditions ANDed is the lowest certainty

certainty or two conditions ORes is the highst certainty

Uncertainties -0.2 >= x >= 0.2 are treated as 0

Question 18

IF the stain of the current organism is gramneg

AND the morphology of the current organism is rod

AND the current patient is a compromised host

THEN CONCLUDE there is evidence of strength 0.6 that the identity of the current organism is pseudomonas.

Suppose the following facts had been established: Stain of the current organism is gramneg with certainty 0.4 Morphology of the current organism is rod with certainty 0.7 Current patient is a compromised host with certainty 0.9

what is the certainty of the conclusion?

Answer 18

0. 4 * 0.6 = 0.24
   i. e. fairly weak evidence to support the conclusion

Question 19

If two or more rules draw conclusions about the same fact then the evidence must be combined. What formula does MYCIN use to combine new evidence with existing certainty?

Use CF_old and CF_newev to refer to the confidence values

Answer 19

CF_new = CF_old + (1-CF_old) x CF_newev

Question 20

Suppose we have two rules:

R1: If A AND B Then Conclude X With Strength 0.6

R2: If C Then Conclude X With Strength 0.5

Suppose the system has already discovered A with certainty 0.8 B with certainty 0.7 C with certainty 0.6

Calculate the certainty of X

Answer 20

Rule 1: 0.7 * 0.6 = 0.42

Rule 2: 0.6 * 0.5 = 0.3

Combined: 0.42 + (1-0.42)*0.3 = 0.594

Question 21

What two questions is MYCIN able to answer?

Answer 21

Why - why the system asked the user a particular question

How - how the system reached a particular conclusion

Question 22

What does MYCIN form as the consultation proceeds?

Answer 22

A record of all the rule activations, including rules fired to satisfy conditions of other rules

Forms a tree of goals and subgoals

Question 23

How does MYCIN answer a WHY question?

Answer 23

Looks up the tree from the current position to identify the goals the led to the current question

Question 24

How does MYCIN answer a HOW question?

Answer 24

System looks down the tree from the current position to identify the evidence that led to the current conclusion

Question 25

WHY and HOW questions are only possible when using _________ chaining

Answer 25

Backward

Question 26

What are two benefits of WHY and HOW questions?

Answer 26

Increases user’s confidence in answers

Enables developer to debug the system easily

Question 27

What do meta-rules do in mycin?

Answer 27

Determine the order in which rules are tried

Question 28

The therapy module of MYCIN uses ________ chaining

Answer 28

Forward chaining: generate and test to find suitable set of antibiotics

Question 29

Give an example of a rule that is actually implemented using forward chaining in MYCIN

Answer 29

IF patient is male THEN CONCLUDE not pregnant

Question 30

Why is conflict resolution not needed in backward chaining?

Answer 30

Only the rules which are relevant are ever used, so the order of firing is unimportant