Download Artificial Intelligence

AI Report
Artificial Intelligence
By: Angelo Rizzolo and Alexandre Ferland
April 30th 2008
1
AI Report
Introduction
In the world around us, there are many bright people that devote their lives to research. These people
work extremely hard to try and discover something new. Robots are one of the inventions that
researchers have put a lot of time and effort to develop. Robots are capable of replicating things that
human are able to replicate. The piece of information that is missing to create intelligent robots is to
actually give them intelligence. A program called Artificial Intelligence gives these robots their
knowledge. Robots are extremely capable of doing things exactly as humans can, but will robots be able
one day to act as a human would automatically? In the past, Artificial Intelligence was a dream and
today robotic technology is advancing rapidly. Presently, scientists are trying to predict how robots will
be in the future. But for robots to evolve, we must understand the human brain and answer the
following questions: What is Intelligence, What is Artificial Intelligence, What are the types of Artificial
Intelligence, History of Artificial Intelligence, Examples of Intelligent Robots, Senses: How robots
navigate in the world around them, Learning, and the Future of Artificial Intelligence.
What Is Intelligence
All things are intelligent in one way or another. Intelligence is used regularly when designing and
programming robots. Intelligence is a big word for scientist because there are many different definitions
and you can put the word in many different contexts. This is why scientists are not able to agree on an
exact definition of intelligence (Davis, D., Dwight, D., & Harris, L, 1986). Since scientists are not able to
put an exact definition, they keep trying to produce more and more intelligent robots (Davis, D., Dwight,
D., & Harris, L, 1986). But with mass production of machines, there are advantages and disadvantages.
The disadvantage is that in the future, “humans might be able to duplicate a machine that will be able to
think” (Davis, D., Dwight, D., & Harris, L, 1986) or even does everything that humans are able to do.
Another disadvantage would be that there are increasing demands for these intelligent machines, so
researchers are not able to do the full work because they are too concentrated on one thing and are not
2
AI Report
able to continue their studies or their dreams. But crating “intelligence” isn’t bad for researchers since
we have people constantly working on these intelligent robots. One day, man might be able to create an
artificial human because every time we try to improve our intelligent robots, humans start to
understand a new part of how the brain works (Davis, D., Dwight, D., & Harris, L (1986). But behind
these intelligent robots, there are intelligent programs that tell these robots what to do a nd what gives
the knowledge. These programs are called Artificial Intelligent programs (AI). They are the masterminds
behind all intelligent robots and with them; it is possible for scientists to discover new ways to improve
on their robots. Every time that discoveries are done like this one, humans get closer to finding a way to
recreate the human brain artificially.
What is Artificial Intelligence
Artificial Intelligence is a subject that is very vast and extremely complex because if you want to create
an intelligent machine, you have to make it work as if it was a normal human interacting with another
human. This subject is so complex that even the smartest scientists are not able to crack the exact
definition of intelligence or artificial intelligence (Perry, R, 2000). One of the reasons for this
phenomenon is because scientists have not yet fully understood how the human brain works (Perry, R,
2000). But scientists have made a temporary definition on artificial intelligence and it is: "Artificial"
means that it is made by humans by biological, chemical or physical process (Perry, R, 2000). The
definition of intelligence includes: "To learn or understand from experiences"," To gain and retain new
knowledge", "to change the behavior to achieve goals", "To use reason and logic to make decision and
solve problems", "To use mental processes to carry out tasks" (Perry, R, 2000). This definition is know all
over the science industry, but if you use all of the definitions above and the robot is able to do those
task, you have invented one of the closest replicas of an artificial human (Perry, R, 2000), and AI is what
makes robot look alive in our society (D'Ignazio & Wold (1984)). In houses, almost everyone has a
machine that uses AI technology and they have a limited knowledge and that is the plain old computer
3
AI Report
(D'Ignazio & Wold (1984)). The computer uses technology to follow our command but they have so
much limited knowledge that they don’t have any emotion or think by themselves (D'Ignazio & Wold
(1984)). But in the future, scientists want to make all known robots on earth created in the future react
like human and make computer “learn” and have “emotions” (D'Ignazio & Wold (1984)). But before we
can do all of this, we must be able to understand the human mind and be able to develop this
technology of the future.
Types of Artificial Intelligence
Like in any other subject, Artificial Intelligence is divided into two schools of thought. One school of
thought is that human thinking is a series of mathematical rules and it would be possible to duplicate
the human brain in computers and the computer would be able to think, feel and react like a human.
The other AI school of though believes that the mind is unique and computers could only simulate some
aspects of the way people think, for people are unique.
AI technology and studies are divided into 3 major categories: "heuristics or rule-based AI, pattern
recognition and natural learning" (Perry, R, 2000). Heuristics is a method used to help to solve a
problem. The solution to the problem is usually reasonably close. Heuristics are basically "rules of
thumb", educated guesses, or simply common sense (Perry, R, 2000). Pattern recognition works when
electronic waves are sent and a computer detects them, the computer then translates it and it repeats
the signal out loud, or AI sends a message back to the sender. Natural learning is something that is still
in question because we want computers to be able to learn its environment, but this is not possible
because computers don’t have the technology that is needed for it to learn. One example of a natural
learner robot, or computer, is IBM'st Deep Blue computer, a computer that is developed for chess. This
computer is very smart because it defeated Kasparov, one of the most famous chess players and best
amateur chess player in the world. Deep blue defeated Kasparov because it had been programmed to
4
AI Report
know all of the moves that Kasparov did (Perry, R, 2000). But Deep blue also defeated Kasparov because
he did millions of calculations, he would choose the best move and play that move (Perry, R, 2000).
Machines of this type are starting to help researchers understand how to make a robot think by itself,
but the main key to success, is to understand how the human brain works so that researchers can create
a robot that is similar to a human, and reacts like a human. With this type of success, man might be able,
one day, to recreate a human artificialy.
History of Artificial Intelligence
The idea of artificial intelligence goes back very far in the history of humans. The ancient Greeks
themselves thought about how humans might be able to recreate themselves. In Jewish folklore, people
also thought about something called a golem which is “an animated being created entirely from
inanimate matter” But this dream did not become a science/true field of study until 1950, when Alan M.
Turning, a British mathematician, invented a test for Artificial Intelligence (Perry, R , 2000). The test
required that a human and a robot be asked a series of questions, and if the judge was unable to
determine whether the human or the robot is answering the questions, the test would be a pass and
man would have invented the first Artificial Human. In the past, this test was used many times, and not
a single robot has been close to getting a perfect score (Perry, R, 2000). Since 1956, AI researchers have
pushed themselves in two different directions: "1. An AI machine should carry out definite tasks in a way
that would be considered "Intelligent" if humans did those tasks", and “2. An AI machine should show
all-purpose, human-level intelligence." (Perry, R, 2000). These directions have taken AI professionals into
different fields: “robotics, handwriting and voice recognition, speech synthesis, natural language
machines, machine vision, expert systems, fuzzy logic systems, neutral networks, and more." (Perry, R,
2000). In other words, AI is cutting edge computer science (Artificial Intelligence, 1986). Computer
scientists conveyed the first AI conference (Davis, D., Dwight, D., & Harris, L, 1986). With, this, scientist
start to create a path to go to success, and on their way to success, scientist will discover something new
5
AI Report
or another subject related with the topic. In, all, AI has a very long history period of what humans
wanted to happen, but this dream only started at the beginning of the computer age. With the time
since the beginning of the project, researchers have been making a lot of progress on the topic, and this
progress will continue to evolve until someone discovers how to make a robot act like a human being.
Examples of Intelligent Robots
In the world around us, robots are at work in factories, and even in our houses. There are many different
types of robots, but only some robots are going to be more interesting than others. There are many
types of robot in the world, and every single one of them has a different utility in the community. NASA,
is one of the organizations that make some of the most advanced robots and some of the most robust
because most of the time, these robots are used on other planets in the galaxy. Right now, NASA is using
a rover to move around the planet of Mars and is used for research and discovery. These robots are
inserted with AI, but they stay under constant supervision at NASA headquarters and workers might be
required to intervene with the robot once in a while (Bluck, J). These rovers are being tested with new
types of AI and NASA would like to be able to send more rovers on Mars with a more advanced AI
system. Anther very intelligent robot was created by Honda, and he is currently being used at Honda’s
reception. This robots is named is ASIMO. ASIMO is the most advanced robot that is existence right now,
and he is very human like because he can do things that human do (Honda Worldwide). Not only can a
human be impersonated by robots, so can any pet. AIBO, a small little dog developed by Sony Corp., is
using AI to react like a dog and to adapt to its environment (Steenhuysen J). Another very intelligent
machine is the IBM’s computer Deep Blue that is developed to play chess. All of these robots were made
out of what we have learned for mistakes and from results shown in the past. These robots are good
examples of what our technology can create if we use it wisely. Robot are things that have been used in
factories for many years and scientists believe that if these robots continue to evolve the way they are
doing right now, they might be sold to the public and they would be used in many different ways.
6
AI Report
Learning
Now that you have a good understanding of what AI is and examples of intelligent robots, we
are now ready to explore the science behind it all…how AIs/Intelligent Robots work. AIs must be able to
learn and perform activities that involve advanced programming. According to Perry, R (2000), AIs can
do various things, for example, recognize patterns, write, recognize voices, recognize moving objects,
recognize environment, distinguish sounds, recognize faces and learn.
AIs can be considered expert systems and can contain neural networks (copy of human brain) in
some cases. First, there’s pattern recognition. Perry, R. (2000) states that computers are programmed to
understand patterns of information, then repeat it, or use it, or respond to it. Handwriting and voice
recognition are the most known AI programs (Perry, R. 2000). These programs are used on palm pilots or
handspring's visors. On these machines, there is a remote area where you can write down information
or speak into a microphone where the program will "learn" and recognize the voice. This is because the
sound waves from your voice are transmitted through the air to the machine where the individual
electrical impulses/vibrations enter the CPU of the machine and store the waves exactly as they were
pronounced (Perry, R. 2000). For the machine to execute the program properly, it must be able to
understand quickly and correctly the first time.
According to Honda Worldwide (n.d), the following AI programs are much more complicated
and require a very advanced programming technique: recognition of moving objects (takes picture to try
to replicate movement), recognition of posture and gesture(calculates points of significance and acts
upon them accordingly), environment recognition (determines objects according to shape, size, color
and other characteristics and acts upon them to either move around them, pick them up etc.) and face
recognition (calculates characteristics of significant parts of the face to identify someone). The ability to
learn is probably the most complicated AI program of them all. To make AIs learn, scientists must first
7
AI Report
identify how humans learn. So through trying to make a robot learn/think a greater understanding of
the human brain is needed. What scientists then decided to do was to mimic the human brain
mechanically. The results are neural networks.
According to Jefferis, D. (1999), Mikhail Korkin, a Russian scientist, has developed a system that
allows brain-like functions to evolve in seconds a robot’s neural networks have simple "nodes", brain
cells in humans. The nodes make decisions based on trial and error which learn as they go along. The
neural networks in AIs use "fuzzy" reasoning. This is a fairly powerful system which can recognize a
dollar bill by being shown enough other bills to learn the common features. Some cars are even made
with "fuzzy-logic", which is a software system that can learn a person's driving style. But this
reconstruction of the brain is not near perfect but experts believe that is will go far.
Jefferis, D. (1999) states that, Australian roboticist Rodney Brooks created the learning
humanoid robot known as "Cog", deriving from the Latin word “cogito” meaning “I think”. Cog learns
through layers. For example, if an object got in the way, Cog would go down one instruction layer to
obey a rule saying "step over the obstacle". If the object was too big to step over, the robot would go
down a further layer to "back off and walk around the object". In this way, Cog avoided obstacles
without having to "think" at all. Cog is known by his team of inventors as a "baby" born/programmed
with basic desires but no knowledge. The hope is that Cog will "learn" like a child. Whether Cog will
become intelligent enough to be aware, like humans, is a question that will be answered as a result of
further research (Jefferis, D. 1999).
8
AI Report
Senses
Like humans, robots and AIs need to be aware of their surroundings. For machines, sensing is very
different if not better than human senses. Sensors are the things which handle this complex task. The
following should give a good description of the senses in AIs/Intelligent Robots (Jefferis, D. 1999).
1. Seeing:
Little video cameras containing about 1 million light sensors all on a chip of about 1/2 an inch
across (color or black and white). Video cameras are placed in place most convenient area.
2. Hearing:
Microphones can be placed also like video cameras at most convenient place. Machines that use
these microphones can act on simple spoken commands and can also "speak" themselves
through voice synthesizers.
3. Touch:
Used for robots who move, pack or arrange things. This sense is achieved through contact
switches. When a switch comes into contact with something, it closes and sends a signal to the
computer. Special sensors called strain gauges can record the amount of pressure needed to
grasp an object. With these sensors a robotic hand can as easily lift a heavy steel beam as it can
pick up a fragile egg without breaking it.
4. Taste:
Powerful "computer" needed. One tea company claims to have built a robot more sensitive to
differences in blends than human tea tasters with years of experience
5. Smell:
Powerful “computer” needed. "Chemical Sniffers" have been made to detect the faint fumes
given off by explosives. Used in big airports to check baggage for terrorist bombs.
9
AI Report
According to Jefferis, D. (1999), senses connect with the brain using nerves and neural pathways, they
do the same in machines. Electrical signals travel 1000 times faster in machines than in humans send
electrical signals to their brains. Contact switches and strain gauges register every movement and send it
to the CPU (Central Processing Unit)/Brain (Jefferis, D. 1999).
Issues
When it comes to issues about AI, there are many and they are very controversial. For example,
should robots be given the same/similar rights as humans if they can think? One of the issues
that AI technology will encounter in the future is: will these intelligent machines have rights?
This will be a big question to answer because they don’t have the same physical appearance as
humans today, but they do have the same functioning mind. Since they have the same
emotions and intelligent characteristics as humans, they should be treated the same was as
humans. But what if they get these machines earn these rights? What would happen? The
result would be that they would go into different jobs and have the same success as humans,
but if this is the case, will they overthrow and become more powerful than humans?
The Future of AI
It is always interesting to know what the future has in store for us. When it comes to AI, the
advancements are predicted by experts, to be incredible. From cyborgs to thinking homes, to biorobots
10
AI Report
and even virtual humans, AIs future has big projects ready to come to our homes. According to Perry, R.
(2000), robots will become a lot more useful in the future and will be friendly, and inexpensive. For
example, an advanced softbot called a verbot with virtual personality would be used and programmed
to understand and recognize your voice. Virtual humans may sound scary, but they are actually designed
to seem friendly. The Firby toy was one of the first robots to be classified as a virtual human. Virtual
humans are programmed to reflect human personalities. These virtual humans may be able to express
their feelings, or at least respond to our feelings with emotional intelligence. Dr. Marvin Minsky thinks
that emotions can be simulated with AI programs. A virtual personality called Weleby already performs
services like scheduling an appointment, phone calls, etc. This uses AI to understand the spoken voice,
get information, and reply as a normal human and in a normal human voice (Perry, R. 2000). So by 2010,
scientists think that computers might be able to know when you are happy and when you are sad (Perry,
R. 2000). Ever imagine living in a home that “thinks”? Well in the future, it is predicted that our houses
will be able to control almost everything that goes on in them. This would all be connected to a satellite
dish on the roof of the home and would run on microchips that will "learn" what lights to light up in the
night, when to wake up etc (Perry, R. 2000). Another example of a great invention in the future would
be Cobourg’s. According to Jefferis, D. (1999) a cyborg is a mixture of human and machine, half way
between a living thing and a robot. Cyborgs do not exist yet but the idea is not new. People have been
trying to improve nature's original designs for years. Jefferis, D. (1999) also states that with brain
implants, permanent connection to worldwide electronic communications may just be around the
corner. Super senses could be provided for better touch, sight, or sound using add-on sensors. Super
strength may even occur using mechanical bodies. AI could also bring exoskeletons (wearable outer
body for people who have lost limbs), closer to reality. Exoskeletons would be useful for missions such
as firefighting. For example a firefighter would sit behind a window and walk straight into a raging
inferno fire without getting burnt. Biochip implants would link the firefighter with AI at the fire station
11
AI Report
(Jefferis, D. 1999). Ichiro Kato, Japanese robot researcher, plans to develop robots that help humans and
are kind as well as efficient. This would be called "Project Humanoid". The robots would assist humans
in a friendly manner. The idea is that if a machine's behavior can be programmed, designers could make
sure AI reflects the good side of people (Perry, R. 2000). As AIs develop, awareness might come
gradually, as is seems to with humans. For scientists to know for sure, an AI might have to communicate
on its own. What has this world come to?
Conclusion
In conclusion, technology has gone a long way, from black and white television, to the possibility of bio
implants. AI might as well be the best form of technology known to man to this day. Technology changes
all the time and very rapidly. The possibilities are endless. AI will aid people in every aspect of life, from
medical advancements to playing with children; AI will always be there to make the world an easier
place to live in.
.
12
AI Report
Bibliography
Bluck, J. (n.d.). NASA - . Retrieved March 5, 2008, from
http://www.nasa.gov/vision/universe/roboticexplorers/robust_artificial_intelligence_jb.h
tml.
Crevier, D. (1994). Ai: The Tumultuous History of the Search for Artificial Intelligence. New
York: Basic Books.
D'Ignazio, F., & Wold, A. (1984). The Science of Artificial Intelligence (Computer-Awareness
First Book). United States: Franklin Watts.
Davis, D., Dwight, D., & Harris, L. (1986). Artificial Intelligence Enters the Marketplace. New
York: Bantam Books.
Honda Worldwide | ASIMO | Technology. (n.d.). Retrieved March 31, 2008, from
http://world.honda.com/ASIMO/technology/intelligence.html.
Perry, R. (2000). Artificial Intelligence. Canada: Franklin Watts, a Division of Grolier Publisher.
Steenhuysen, J. (n.d.). Robot as good as real dog at easing lonely hours - Yahoo! News.
Retrieved March 3, 2008, from
http://news.yahoo.com/s/nm/20080227/od_uk_nm/oukoe_uk_robot_dog.
Tabuchi, H. (n.d.). Japan looks to a robot future - Yahoo! News. Retrieved March 3, 2008, from
http://news.yahoo.com/s/ap/japan_robot_nation.
(1986). Artificial Intelligence (Understanding Computers). New York: Time-Life Books.
Created at www.bibme.org
13