Artificial
Intelligence
(From my science essay
collection How Do we know? )
by
Kenny A. Chaffin
All Rights Reserved © 2013 Kenny A. Chaffin
The first
Alien Intelligence we meet may not be from another planet, but from our own computer
labs. Many of us walk around with a computer in our pocket capable of
listening, parsing, and responding – sometimes even correctly – to our voice. We
Google for information by typing in phrases, sentences or disconnected words
and the artificial intelligence in Google’s search engine almost always comes
back with what we are looking for. These dedicated applications are on the
verge of intelligent behavior and could certainly in their domain be called
intelligent. Other systems are even more so and in some cases demonstrate more
generic intelligence such as the Watson system from IBM that recently defeated
the all-time Jeopardy! champions. So how soon until we get to meet these Alien
Intelligences of our own creation? We could see them perhaps within the century
and almost certainly (provided we don’t kill ourselves off or get whacked with
an asteroid) by next century. A Watson-like system is being rolled out by IBM
to assist in medical diagnosis. Google, Google Voice, and Siri will continue to
improve. New research into machine learning, user interfaces and the human
brain are being brought from the lab into practice. It’s been a long and bumpy
road, at least by technological progress measurement since that first Dartmouth
conference on machine learning in 1956 when Marvin Minsky boldly predicted that
"within a generation ... the problem of creating 'artificial intelligence'
will substantially be solved." Other bold predictions followed every few
years, then every decade until with little actual success the predictions
stopped. In some ways that was when the work really began. And that work as
often happens emerged not from the expected avenues, but from the back allies
and offshoots of other research.
Artificial
intelligence is defined as a branch of computer science dealing with the
simulation of intelligent behavior in computers. John McCarthy, one of the
Dartmouth conference organizers who coined the term defines it as "the
science and engineering of making intelligent machines." The founding idea
was that the central feature of humanity – intelligence – could be analyzed,
described and simulated by a machine. The core issues of accomplishing this
have to do with perception, communication, analysis of sensory input,
reasoning, learning, planning and responding to real-world events. The ability
to perform these functions in a general manner (like a human) is known as
Strong AI though much of the work and research is done as subsets of the larger
goal. There are a number of associated areas as well such as neuron simulation,
learning theory and knowledge representation.
The key of
course is understanding intelligence, what it is, what it does, and perhaps
even why it does what it does. But it is a bit like art or pornography – “I may
not be able to define it for you, but I can certainly tell you what it is when
I see it.” There are many often disparate definitions of intelligence, but what
seems to be the core is problem solving. It involves identifying a problem or
obstacle, seeking or creating a solution, applying that solution and then
evaluating the result. The evaluation provides a feedback process to inform
future decisions.
The
artificial intelligence field got its start at the Dartmouth conference with
four key figures – Alan Newell, John McCarthy, Marvin Minsky, and Herbert Simon
all of whom were computer scientists with exception of Simon who was more of a
psychologist/sociologist as well as being knowledgeable in other disciplines. All
of them were in fact somewhat cross-disciplinary. This idea of building a
machine capable of human intelligence came not long after the advent of the
first practical computers. Given the broad capabilities exhibited by computer
programming as a result of the Von-Neumann architecture based on Alan Turing’s
mathematical concepts it seemed quite possible to program a computer to emulate
human intelligence and decision making. But oh what a tangled web was to be weaved
from this.
Throughout
history there have been numerous attempts, desires, stories and examples of building
or bringing inanimate objects to life. There are clockwork robotic devices, statues
and puppets and trees that came to life (Pygmalion, Pinocchio), the mechanical
devices built to simulate/emulate/recreate human behavior sometimes even with
dwarves or children inside to huckster the crowd, and back beyond even that to the
oracle at Delphi. Given all the literature, myths, stories and actual devices
there must be something very deep in the human psyche that longs to re-create
itself. Perhaps it is even down to the genetic drive of reproducing,
recreating, perpetuating ourselves, perhaps a kind of genetic imperative drives
our attempts, our need to explore artificial intelligence.
Nevertheless
the work seriously got underway following the Dartmouth conference and there
was serious money behind it, primarily funded by the departments of defense in
the United States and in Britain as well as Russia and other world powers of
the mid-20th century.
The
perceived promise led Darpa to invest approximately $3 million a year from 1963
to the mid 1970’s. Similar investments took place in Britain. Darpa of course
was looking for potential military applications during this cold war time of
tension around the world. The promise and the culture of the time though led to
a devastating situation. The funds flowed with little oversight and the field
went in many directions that resulted in little applicable output. This was the
early days of computers and the algorithms and programs designed to emulate
things like human logic and reasoning were quite complex and resource
demanding. They did not work well on the hardware available at the time. Either
the programs had to be scaled back and limited in their scope or very long
time-frames had to be allowed in order to get results. Neither came close to
approaching the abilities of a human brain on any level. Sensing capabilities
such as vision and audio which were being worked on as a subset of the AI
problem required massive programming just to acquire and manipulate the data
into a form that could be used by the AI components.
By the mid 70’s the faltering field
was stripped of funding and mostly dropped. This in now known as the first AI
Winter and would last almost a decade until the early 80’s. Some work continued,
but without the freely flowing funds it was much more focused and more a labor
of love rather than more random experimentation. During this time as well much
criticism was leveled at the computer scientists by other academic departments.
Philosophy, psychology, biology and mathematics all took shots, but by the same
token they were all interested in the field that they had been shut out of in
this early phase and as a result many research institutes began bringing
together diverse cross-disciplinary groups to work on the research as well as
providing means for them to work better together. As a result we get learning
specialists helping to design computer learning applications. We get knowledge
management experts helping to devise search and storage hardware and software.
And we find neurological experts working with programmers to simulate neural
networks.
This ‘background’ research led to
the next step in AI - Expert systems,
which were the rage in the 80’s. An expert system was intended to be a subject
matter expert in a specific or limited domain. It incorporated a knowledge base
and a means of searching and retrieving (as well as updating) information. This
lead to a boom in database research and development. Computers were rushing
along following Moore’s Law of doubling capabilities every two years. This
allowed for more complex search algorithms which were ever faster as were the
database search and retrieval. The programming language of choice for these
systems was Lisp a symbolic manipulation language thought to better model
thought processes, symbolic manipulation and such. Certainly there was some
success for these expert systems but again the result failed to match the
expectations and once more the field floundered.
In the meantime Japan initiated the
Fifth Generation Computer Project which was intended to create computers and
programs that could communicate using natural language, do visual processing
and recognition as well as emulate human reasoning. They dropped Lisp and chose
a newer language Prolog as the core programming language perhaps to leave the
old ways behind and start anew. Other countries responded in kind to this ‘threat’
of computer dominance. During this time much work was beginning to focus on
neural networks and emulating the brain in hopes of breaking free of the
step-by-step von-Neumann style of programming. This took place (and continues
to this day) in both hardware and software. Emulating the workings of
individual brain neurons as well as connecting them in the manner of a
biological brain. But again the lack of substantial applicable results to
business or military uses brought on another ice-age. The second AI Winter
lasted from the late 80’s to the mid 90’s.
By this time the field of robotics was
rising particularly due to the use of robotics in assembly factories such as
car manufacturers and electronic assembly plants. There was money to be had for
robotics research and a new slant on the AI field emerged. By providing the
means to these factory robots to handle ambiguity, recognize defective parts,
to align and assemble them properly without supervision or through extremely
precise programming and logistics provided a new venue to AI. It wasn’t just
emulating human intelligence and reasoning, but performing the tasks a human
would do in a real-world assembly factory.
A separate but similar revolution
was taking place in space exploration. Our robotic probes to Mars, Venus,
Saturn and the outer planets were being designed with increasingly autonomous
and error-correcting capabilities. These robots – rovers and probes of various
styles had to operate autonomously in much more demanding and dangerous
situations than the factory floor. NASA and the military funded some of the
best minds, universities and corporations to build these mechanical emissaries
to the cosmos.
There
was a completely different revolution taking place during these years as well.
From the mid/late 80’s the business demands for data storage and retrieval have
exploded like a sun going nova. This fueled much database research and even
special purpose hardware research such as Teradata and Britton Lee database
machines. The advent of the internet brought search engines to the fore and the
star of course that emerged was Google now a household name/word and verb
equivalent to internet search. The massive data problem is far from solved, it
continues to grow. Everything has gone digital. Businesses store all their
corporate data digitally, our space telescopes produce massive amounts of data
as do research projects such as the Human Genome Project and other DNA and
biological analysis research as well as the recently announced Human Brain
Mapping initiative. This issue is known today as the Big Data problem and
significant amounts of cash from government and private industry are pouring
into managing the problem. The results of this are applicable as well to AI
research because one of the obstacles is providing and managing the amazing
amount of information storage required to emulate a human brain.
A human brain has about the same
number of neurons as there are stars in the Milky Way galaxy – about 100
million. And each of these neurons may be connected to thousands of other
neurons. This creates an amazingly complex multi-processing system that is not
only difficult to emulate but requires computing capabilities that are
currently beyond present-day systems. We may see it in the next half-century
though.
All of these areas of research and
application are at the fore-front of today’s computer, information and
cognitive science. Google is increasingly capable of parsing and analyzing
natural language inputs and providing (in extreme short time-frames) relevant
results. We have cell phones that are capable of processing speech input and
providing similar search results or actions based on the spoken words. Our
robots are exploring Mars, Voyager (launched 35 years ago) is still functional
and approaching the edge of interstellar space. It requires 20 minutes for
radio messages to travel to or from it. The autonomous land vehicle trials by
Darpa and Google continue. Google’s vehicle has already been given approval for
commercial operation of these vehicles in several states.
It seems we are now approaching
real AI – the capabilities of humans from several oblique angles following
failures of direct methods of programming rational decision making, expert
systems, and embedded logic. It seems that real AI is coming not from the
research labs, but from the factory floor, our autonomous space probes and
vehicles, and from our information management needs. We continue to attempt to
emulate the physical structure and workings of the human brain but some of our best
results are in our pockets -- our cell phones with voice-actuated access to the
world’s knowledge at the tip of our tongues.
References/Resources/Links
Artificial Intelligence:
History of Artificial Intelligence:
Watson:
Human Brain:
Human Brain Mapping:
Blue Brain Project:
Neural Network Software:
Expert Systems:
Strong AI:
Big Data:
About the Author
No comments:
Post a Comment