Homework1x

stemswedishΤεχνίτη Νοημοσύνη και Ρομποτική

15 Οκτ 2013 (πριν από 3 χρόνια και 10 μήνες)

92 εμφανίσεις

Sean LaPlante









2/6/2012

Dr. Mullick










CS495

Homework 1

1.

Artificial Intelligence:
The ability for a non
-
human entity to make decisions and act on
those decisions based on learning from its environment or based on a set of pre
-
defined
rules. Another definition of AI is the ability for a computer to act like a human.


Agent:

The agent is

the thing doing the actions (the computer, or other non
-
human AI
apparatus). The agent can interact with its environment, change it, and possibly learn
from it.


Rationality:

An artificial intelligence system is rational if it is able to do the right thin
g
based on what it knows or has learned. Another definition of rationality is the ability for a
computer to make
informed decisions (resulting in the preferred outcome for the given
situation) based on what it has learned from the environment, or what it w
as programmed
to already know.


2.

Yes,
I believe
reflex actions (such as flinching from a hot stove or putting out your arms
to catch yourself) are rational

because they are the right thing to do
.

However
,

they are
rational without involving inference, meani
ng, we don’t think before we pull our hands
away from the hot stove because taking time to think about flinching would eliminate the
purpose of flinching (it is supposed to be a quick thing). Reflex actions ar
e

handled by
the spinal cord, flinching from a
hot stove happens before the hot signal reaches are
brains, so we don’t have time to think about it. Therefore I believe that reflex actions are
not

intelligent

because

no thought goes into them, we do them automatically
(instinctively).


3.

Supermarket bar c
ode scanners:
I believe supermarket bar code scanners are not
intelligent in the sense that they are concerned with thinking. I think bar code scanners
are concerned with behavior (accomplishing a task). Bar code scanners read a bar code
and find (in a dat
abase) what item that bar code belongs to. There is no decision making
(beyond picking the matching item)
or learning involved.

If the proper item is not found a
bar code scanner will not try to pick the closest item, it will simply report that the item
wa
s not found.

However, the bar code scanner can be seen as an agent

with sensors
(lasers) that interact with the environment and then the agent does work on the
information gathered from the environment. I believe that if bar code scanners have to be
consi
dered intelligent, that they are at
the lowest end of intelligence. According to the six
disciplines that make up most of AI, the bar code scanner demonstrates computer vision
at a very low level (scanning a bar code with lasers).


Web search engines:
I be
lieve web search engines are a significant step up from bar
code scanners with respect to artificial intelligence. A web search engine takes
information from the environment (input by a human) and looks around the internet
for
that exact phrase or word. I
f

it cannot find it exactly, instead of returning no results, it
looks for things that relate to what you searched for. Web search engines can also search
for the words appearing separately instead of all together the way the user typed them.
Also, search e
ngines like Google can attempt to figure out if the user meant to type what
he or she typed. If the user made a spelling error the search engine can suggest a fix to
that.

According to the six disciplines that make up most of AI, the web search engine
demo
nstrates automated reasoning and machine learning. It uses automated reasoning to
figure out what you could have meant based on other search queries by other users. It
used machine learning to bring search results that relate to what you usually search or
have searched in the past (i.e. Google).


Voice activated telephone menus:

I believe that voice activated telephone menus are
intelligent because they demonstrate an ability to interpret human voice and respond
accordingly. These types of systems are not t
he best; however, they demonstrate the
ability to understand different accents (to an extent) and different speeds and pitches of
human voice.

According to the six disciplines of AI, the voice activated telephone menus
demonstrate natural language processi
ng to communicate successfully in English.


4.

I believe that it is true that a computer can only do what its programmer tells it. But I also
believe that a programmer can program a computer in such a way that it is able to “think”
for itself. For example, a programmer could write a program that has mu
ltiple outcomes
based on the situation the computer is presented with. There is no way for the
programmer to know how the computer will react unless the programmer knows the
situation that it will be reacting to. The computer is still only doing what the p
rogrammer
tells it, the programmer merely gave the computer options and “taught” the computer
how to choose the right option based on the situation. So I believe that computers can be
intelligent, just not in the same way that humans are intelligent. I bel
ieve that it is
impossible for a human to build a machine as intelligent as he or she is simply because
we don’t fully understand our own intelligence. However, I do believe that we can build
machines that are slightly less intelligent than we are.


5.

True/
False

a.

False. In the vacuum cleaner example, the vacuum cleaner only has partial information
because it does not know the condition of the square next to it; however, it is still rational
because it is able to choose the proper action based on its curren
t square.

b.

True.
This is because one of the task environments could be constant. For example, if
you roll a die in one environment and the die is weighted to always be 1, and you roll a
die in the other environment which has a random outcome. Then if th
e computer always
guesses the number 1, it is being perfectly rational in two distinct environments. Since
there is no way to know what the random die will be, it is rational to always the same
number, however the computer knows what one of the die will be
, therefore it should
always pick the number 1.

c.

False. If we assume that the statement is true (a perfectly rational poker
-
playing agent
never loses) then if two perfectly rational poker playing agents were to face each other
they would both win. However, this contradicts the rules of poker because the
re can only
ever be one winner. Therefore a perfectly rational poker
-
playing agent does not always
win, instead a perfectly rational poker
-
playing agent should know when it does not stand
a good chance, and fold.



6.

PEAS Descriptions

Agent Type

Performance

Measure

Environment

Actuators

Sensors

Playing Soccer

Where is the
goal, how far
away is the
goal, how hard
should the ball
be kicked

Grass field,
indoor turf,
indoor gym
floor

Mechanical
foot

Sight for
sensing where
the goal is

and
where the ball
is

Pla
ying a
tennis match

Knowledge of
tennis court
size, and tennis
rules. Ability to
predict good
shots based on
opponent
position.

Tennis court

Side to side
movement,
mechanical arm
with racket to
hit the ball

Sight to see the
ball, aim, and
detect location
of opponent.

Performing a
high jump

Knowledge of
distance to pole,
ability to pick
jumping spot
based on pol
height.


Straight strip of
flat ground
leading to a pole
to jump over

Mechanical
legs for
running,
balance, and
jumping

Sight to know
when to jump

and how high
to jump.

Bidding on an
item in an
auction

Usefulness of
item, some
desirability
level assigned
to items (how
valuable is it)

Internet auction
sites, or an
actual auction
room

Ability to post
a bid over the
internet, or
speech
actuators to
say
a bid out loud
in an auction
room

item sensors
(what is the
item), either
sight (auction
room), or
image
processing
(internet)








Playing Soccer

Playing a
Tennis Match

Performing a
high jump

Bidding on an
item in an
auction

Observable

No***

Yes**

Yes**

Maybe*

Deterministic

No

No

Yes

Partly
****

Episodic

Yes*****

Yes*****

Yes*****

Yes******

Static

No

No

Semi

Yes

Discrete

No

No

Yes

Yes

Single Agent

No

No

Yes

No

*If the auction is online the AI cannot see the others bidding, if the auction is in

an
auction room then the sight sensors should be able to see the others.

**Because the computer can see the whole playing field at all times with its sight sensors.

***Because there are parts of the field and players out of view of the sensors.

****There
are others bidding so it’s not entirely determined by the computers actions.

*****If the AI has the ball then it shoots, otherwise it doesn’t, it doesn’t matter what
happened before

******If the AI is winning then it bids, otherwise it does not. No previou
s knowledge is
required.


7.

Define:

a. State Space:
The state space
is the set of all states that can be reached from the initial
state by a sequence of actions.

b.

Search Tree:

The initial state is at the root and the root branches off where each
branch is an action and each node is a state in the state space of the problem.

c.

Goal:

The goal state is the solution to the problem. When the goal state is reached the
search is done
.

d.

Heuristic:

Heuristics provide knowledge for directing or constraining searches.
Heuristics help guide the search to the result in a faster more efficient way than just going
through the search procedurally.


8.

Apply search algorithms to tree on homework

sheet.


a. Depth
-
first:


1. Initialize open = {S
}, closed = {}


2. Retrieve next state left of open = {A}, closed = {S}


3. A is not the goal, generate children, open = {D}, closed = {S,A}


4. D is not goal, generate children open = {G}, closed = {S,A,D}


5. G is goal, put it on closed list. Done. Open = {} closed = {S,A,D,G}

Final Path

Dept
h
-
first:
S,A,D,G



b.
Breadth
-
First:



1. Initialize open = {S} closed = {}


2. Generate children open = {A,B,C}, closed = {S}


3. Check A,B,C, no goal, generate childr
en open = {D,E,F} closed = {S,A,B,C}


4. Check D,E,F, no goal, generate children open = {G1,G2,G3}


closed = {S,A,B,C,D,E,F}


5. Check G1,G2,G3, G1 is goal, put it on closed. Done.



Open = {} closed = {S,A,B,C,D,E,F,G1}


Final Path Brea
dth
-
First:
S,A,B,C,D,E,F,G1


c. Best
-
First

(I will only be using ‘h’ values not arc values)
:


1. Initialize open = {S} closed = {}


2.

Generate children, move S to closed, open = {A
-
13,B
-
12,C
-
11} closed = {S}


3. Order the open list, open = {C
-
11,B
-
12,A
-
1
3} closed = {S}


4. Examine C
-
11’s children and order. Open = {B
-
12,A
-
13,F
-
13} closed = {S,C}


5. Examine B
-
12’s children, order. Open = {A
-
13,F
-
13,E
-
13} closed = {S,C,B}


6. Examine A
-
13’s children, order. Open = {F
-
13,E
-
13,D
-
13} closed = {S,C,B,A}


7. Ex
amine F
-
13’s children, goal found, generate path, closed = {S,C,B,A,F,G3}


Final Path Best
-
First:

S,C,B,A,F,G3