53LOGICAGENT

advertisement
Agents that use logic
Knowledge, inference and action
in the
environment
Logic – based agents
 reason by
 inference on KB (search)
 evaluating logical expressions
 knowledge of environment
where do goals, plans, actions fit in?
in the KB or external?
D Goforth - COSC 4117, fall 2003
2
Example – reasoning only
 Smith, Jones and Clark
each has a distinct job
 programmer, knowledge engineer, manager
 Jones owes programmer $1000
 manager’s spouse doesn’t allow borrowing
 Smith is not married
 Who has which job?
D Goforth - COSC 4117, fall 2003
3
Knowledge
 9 propositions:
 Cp, Ck, Cm
 Jp, Jk, Jm
 Sp, Sk, Sm
 29 = 512 rows in truth table
S: Smith; J: Jones; C: Clark
p: programmer; k: knowledge engineer; m: manager
D Goforth - COSC 4117, fall 2003
4
Knowledge
 general knowledge




Jp \/ Sp \/ Cp // there’s a programmer
Jp => ~Sp ^ ~Cp //just one programmer
Sp => ~Jp ^ ~Cp
Cp => ~Sp ^ ~Jp
D Goforth - COSC 4117, fall 2003
5
Knowledge
 general knowledge




Jp \/ Jm \/ Jk // Jones has a job
Jp => ~Jm ^ ~Jk //just one job
Jm => ~Jp ^ ~Jk
Jk => ~Jm ^ ~Jp
D Goforth - COSC 4117, fall 2003
6
Knowledge
 facts
 ~Jp // Jones owes programmer $1000
 ~Jm // manager doesn’t borrow
 ~Sm // Smith not married, not mgr
Construct the truth table headings
D Goforth - COSC 4117, fall 2003
7
Knowledge base
9 propositions
Cp Ck Cm …
24 general knowledge
sentences
CpJpSp …
facts
~Jp ~Jm ~Sm
KB=ΛRi
Reasoning
Common Sense
Smith is not
married
Jones owes
programmer $1000
manager’s
spouse doesn’t
allow borrowing
D Goforth - COSC 4117, fall 2003
8
Dynamic environments
 Logic is based on static knowledge
 If agents can act to change their
environment, logical representation
has problems because “facts” are
changing
 What happens to conclusions based on
original facts?
D Goforth - COSC 4117, fall 2003
9
Wumpus world
D Goforth - COSC 4117, fall 2003
10
Representing the world
 Wumpus world environment facts






16
16
16
16
16
16
propositions
propositions
propositions
propositions
propositions
propositions
for
for
for
for
for
for
pits
breeze
wumpus
stench
gold
glitter
D Goforth - COSC 4117, fall 2003
11
Representing the world
 Wumpus world
general knowledge
 relate pits and
breezes
 relate wumpus and
stench
 relate gold to glitter
D Goforth - COSC 4117, fall 2003
12
Adding changing facts, plans and
actions
 Wumpus world agent location and
orientation
 proposition for location: L11
 proposition for orientation: R, U, D, L
(facing Right, etc)
 propositions for moving: Fwd, TL, TR
 ... plus shooting, grabbing gold.
 problem – information is not static
 at L23 – after move: no longer true, at L22
D Goforth - COSC 4117, fall 2003
13
Possible agent models
1. implement timed logic
 Facts are true at specific time
2. keep other changing information
outside KB (e.g., position, orientation
of agent)
D Goforth - COSC 4117, fall 2003
14
Timed logic
 solution: location and orientation have
time also





L111, at location (1,1) at time 1
L239
R7, facing right at time 7
R7  TR7  D8
R7  L117  Fwd7  L218
D Goforth - COSC 4117, fall 2003
15
Dynamic problem:
too many propositions
t=0
1
2
3
4
5
6
KB keeps a ‘history’ of the
activity as a series of
snapshots
D Goforth - COSC 4117, fall 2003
16
Dynamic problem:
too many propositions
t1
= 234
0 56
Partial solutions:
1. More powerful logic (next chapter)
2. Keep dynamic knowledge separate
static reasoning in KB
dynamic in:
algorithm in agent (p.226)
reflex wired agent (p.229)
D Goforth - COSC 4117, fall 2003
17
2. Static reasoning only
plus dynamic algorithm
outside
KB
2. Static reasoning only
in reflex ‘circuit’ agent
Vector of
perceptions
State
variable
Actions
2. Static reasoning only
in reflex ‘circuit’ agent
Vector of
perceptions
State
variables
Actions
Download