Grundlagen wissenschaftlichen Arbeitens
LVA 185.170



The Future of Artificial Intelligence


Topic 4



by Pevzner Maxim
0326071

1  Introduction

In this essay I will try to collect and describe some of the existing future visions, based on real achievements in technology. Due to the impossibility to describe all of the numerous predictions in this field, I had to concentrate on the ones that, as I think, can really be the corner stones of our future technology and vital assets of AI.
The creation of artificially intelligent life has been one of the greatest dreams of humanity for centuries, although it had always been closely connected with caution. This can be best illustrated by looking at the literature. All of us have without doubt read science fiction books, where science fiction authors have created dreary but also tempting pictures of our future, where everything will be under control of robots and other kinds of machine intelligence. Until recently the creation of robots as such had been viewed as an unreachable goal. Only in the last few decades has this dream become so feasible. We have learned to view science fiction as fiction only, and yet looking at the state of technology today one cannot help but notice the similarities between the current achievements in the robotics industry and what all the authors had described in their works. One can almost say, with their books they had managed to predict our future. For instance, we should not forget that science fiction is in many ways responsible for making space flights true. Maybe our "today" would not be the same without those science fiction authors? By all means, they have long been the drive of human technological achievements. There are numerous examples of that. Scientists have already shown their acknowledgement by naming some craters on the moon in honor of science fiction authors, such as Hugo Gernsback, Willy Ley and Herbert George Wells. At least three craters on Mars were named after science fiction authors: John Wood Campbell, Stanley G. Weinbaum and again Herbert George Wells.[1] One should not forget one of the most notorious and successful authors in this field, Isaac Asimov, who wrote a great deal about Artificial Intelligence.
In his books he described different Artificial Intelligence systems, robots and our near (or maybe not so near) future. The fascinating about him is that many of his books were written over fifty years ago and yet his visions are still up-to date: many of them could be found in scientific works from modern-day professors and scientists. Let us concentrate on his robots.
Robots in his novels have different types of intelligence; they are capable of creating logical threads and of thinking. What they have in common is the list of 3 rules:
  1. Ä robot may not injure a human being, or, through inaction, allow a human being to come to harm."
  2. Ä robot must obey the orders given it by human beings except where such orders would conflict with the First Law."
  3. Ä robot must protect its own existence as long as such protection does not conflict with the First or Second Law."[2]
Asimov understood that creation of powerful artificial intelligence could lead us to the necessity of co-existence with machines. He tried to limit the freedom of robots and make them obey some basic rules or directives. If we accept his fictional creations as actual predictions of our future, then they are astonishingly exact. In fact many scientists work on creation of human-like robots. Of course those three rules of Asimov alone would not be sufficient for maintaining the absolute control over robots, but it can be viewed as a fact that very soon we will be in need of such or similar rules. And yet the question of their sufficiency can not be answered right now because we first have to understand the technological level of our future artificial intelligence.

2  Human-like robots

Nowadays we have almost same visions of our future as centuries before. We want to create human-like robots, androids. There are many achievements in this field and many already-created human-like robots like Qrio and Asimo[3]. Honda's Asimo has 26 degrees of freedom, which allows him to move almost as smoothly as a human-being.
At the end of year 2004 Chris Willis promised the world to show his robot Valerie[4]. By its abilities this robot is better than all projects of Sony, Omron, Honda, Fujitsu, Toyota and Takara corporations.
The vision of creating human-like robots and integrating them in our lives has its pluses and its minuses. Marshall Brain, the creator of the web-site HowStuffWorks (www.HowStuffWorks.com) published his essay under the name of "Robotic Nation"[5]. In this article he predicts our next 50-60 years, based on the current achievements in robotics industry and making prognosis on how it will develop further.
According to his article, the robots will replace us in our jobs already by 2055. Of course those robots will be part of the artificial system development.
Äutomated retail systems like ATMs, kiosks and self-service checkout lines marked the beginning of the robotic revolution. Over the course of fifteen years starting in 2001, these systems proliferated and evolved until nearly every retail transaction could be handled in an automated way. Five million jobs in the retail sector were lost as a result of these systems."[5] So he sees the beginning of the new industrial revolution. After that will come humanoid robots: Sony already proved to us that the robots can walk like humans with its ASIMO project. It will lead world robot corporations to the point where they "could create humanoid body forms with the grace and precision of a ballerina or the mass and sheer strength of the Incredible Hulk"[5].
The thread of researching and developing robots and artificial intelligence will lead us, according to Marshall Brain, to cheap artificially intelligent robots that will be efficient at work and will not complain at anything. Shortly said - ideal workers, so unlike humans.
Between 2025 to 2055 android robots will start to take our jobs. According to Mr. Brain, in 2030 humanoid robot will cost about 10,000$. That robot will do all house chores: cleaning, cooking, iron-steaming etc. Marshall admits that those robots will not be able to think "creatively", but in fact, that does not matter at all. Factories will be full with automated workers, because the robot will make a "human" job better and faster. That is a fact which is supported economically. For those who still do not believe that human-like robots will so soon come into our lives, Marshall Brain has arguments such as the following: "There were millions of people in 1900 who believed that humans would never fly. They were completely wrong. However, I don't think anyone in 1900 could imagine the B-52 happening in 54 years."[5]
Marshall Brain supports his predictions with some powerful arguments. His main argument is the Moore's Law. Gordon Moore is a co-founder of Intel. He observed that the number of transistors on a given piece of silicon would double every couple of years - this observation has been named the Moore's Law. If we look at the history and specifications of microprocessor chips, there are numerous proofs to this law. "Taking Moore's law literally, you would expect processor power to increase by a factor of 1,000 every 15 or 20 years. Between 1981 and 2001, that was definitely the case. Clock speed improved by a factor of over 300 during that time, and the number of transistors per chip increased by a factor of 1,400. A processor in 2002 is 10,000 times faster than a processor in 1982 was. This trend has been in place for decades, and there is nothing to indicate that it will slow down any time soon. Scientists and engineers always get around the limitations that threaten Moore's law by developing new technologies."[5]
Same rules apply to other computer technologies like RAM (Random Access Memory) and Hard Disks. "What if we simply extrapolate out, taking the idea that every 20 years things improve by a factor of 1,000 or 10,000? What we get is a machine in 2020 that has a processor running at something like 10 trillion operations per second. It has a terabyte of RAM and one or two petabytes of storage space (a petabyte is one quadrillion bytes)."[5] Sounds cool?
"What if we extrapolate another 20 years after that, to 2040? A typical home machine at that point will be 1,000 times faster than the 2020 machine. Human brains are thought to be able to process at a rate of approximately one quadrillion operations per second. A CPU in the 2040 time frame could have the processing power of a human brain, and it will cost $ 1,000. It will have a petabyte (one quadrillion bytes) of RAM. It will have one exabyte of storage space. An exabyte is 1,000 quadrillion bytes. That's what Moore's law predicts." According to these calculations we could see the desktop computer (of course if this definition will still be in use) with the processing power exceeding that of the human brain by 2050.
What we will have in another 50 years after that is impossible to imagine.
When the "machine brain" will reach parity with the human brain and surpass it, robots will be able to do any human's job. We see the future of our robotics very optimistically; we dream of seeing robots play soccer. We can imagine that robot also doing our job. One of the advertisements of robot ASIMO (Smithsonian magazine, January 2003, back cover) says: "The dream was simple. Design a robot that, one day, could duplicate the complexities of human motion and actually help people. An easy task? Hardly. But after more than 15 years of research and development, the result is ASIMO, an advanced robot with unprecedented human-like abilities. [...] The future of this exciting technology is even more promising. ASIMO has the potential to respond to simple voice commands, recognize faces, carry loads and even push wheeled objects. This means that, one day, ASIMO could be quite useful in some very important tasks. Like assisting the elderly, and even helping with household chores." Taking human's job also. In our future this kind of robot could have "human-brain" power and be equipped with advanced AI software. If we add fuel cell to it, we will have autonomous humanoid robot with almost unlimited capabilities.
Other base arguments of Brain's prediction are the basic research of American economy now and the research of new employment landscape. Actually, we already see many differences in our industrial world, we see how it affects people: homeless, jobless welfare. As Marshall sees it, there will be no possibilities to stabilize a situation when for example 50 million people lose their job.
It is very hard to estimate, to predict technology sometimes, because one simple break-through can change much. Even scientists can not describe our future life or tell us what we can expect in 30 or 50 years. It is just too much, and what direction our future could choose is hard to know.
Professor Kevin Warwick from the Department of Cybernetics of the University of Reading told in the interview to BBC that "...we [people] are going to see machines, computers that are more intelligent than we are perhaps by 2030, 2050 - how are we going to cope with that? If we don't do something, if it is just humans versus machines, it is a very dangerous scenario. Well, one possibility is to link humans with machines to create cyborgs, part human, part machine." [6]
His suggestion on the frightening perspective when machines will become more intelligent than us is to create implants which will give a human-being the ability to compete with a machine. This would be a step into a new dimension called cybernetic intelligence. Warwick's approach is well supported. We as human-beings are very restricted and have limited capabilities. We rely heavily on our primitive, animal senses, which are so far from being perfect. The best of our senses is vision and the way we communicate is too slow and not efficient. Is there any possibility to upgrade us? Driven by this thought, Kevin Warwick ran projects (such as Cyborg-1, Cyborg-2) that show that the human brain can actually communicate via implant.
Ön Monday 24th August 1998, at 4:00pm, Professor Kevin Warwick underwent an operation to surgically implant a silicon chip transponder in his forearm. Dr. George Boulous carried out the operation at Tilehurst Surgery, using local anaesthetic only.
This experiment allowed a computer to monitor Kevin Warwick as he moved through halls and offices of the Department of Cybernetics at the University of Reading, using a unique identifying signal emitted by the implanted chip. He could operate doors, lights, heaters and other computers without lifting a finger." (Project Cyborg 1.0)[6]
If the prediction of Marshall Brain is true, then by 2050 we will have machines with proccessing power more than that of a human brain. Warwick sees our future controlled by machines, as the more intelligent creatures controlling the less intelligent. According to Kevin Warwick, creation of cyborgs can put humanity one step ahead in the forthcoming competition with the machines. At least for now.
But what will we use instead of our brain when it becomes "too slow"?
Science is something that cannot be exactly predicted. And yet we can be sure that no matter what pace life chooses, in fifty or hundred years humanity will create ïntelligent" humanoid robots and as many scientists agree - it is inevitable.

3  Future of Artificial Intelligence Systems.

Exploring space is a very important subject for humanity. 20th century was the beginning of a new era. Different countries sent their satellites and space ships into space.
First problems encountered were the incapability to repair or diagnose objects sent into space. In the last 10 years tremendous work has been done in this direction. On 18th of October 2004, NASA officials reported about successful tests of artificially intelligent software.
The software, called Livingstone, was developed by computer scientists at Ames Research Center in California. This software is named after the 19th Century explorer and doctor, David Livingstone. Livingstone is implemented on the Ëarth Observing-1" imaging satellite and its purpose is to find technical failures or to determine the root of the failure. Of course it was tested in a simulation mode. Livingstone software compares predicted system performance with the data from screens. It can analyse this data and recommend preventive actions. As NASA believe this kind of software could actually have saved the Mars Polar Lander in 1999, when its onboard sensors mistook deployment of its landing legs for actual touchdown on the surface of Mars. AI software is believed to be the only solution for autonomous spaceships, where no communication to Earth can be established. Its plus is in fast-made decisions in a time-critical situation.
The Livingstone-software tested on EO-1 is much more powerful than the Livingstone-version used on NASA's "Deep Space -1"-spacecraft back in 1999. [7] But Livingstone is not alone.
There is another diagnostic AI software named G2, created by Gensym Corporation. This software is already in use by corporations such as Ford, also by various petroleum refineries and major factories. "G2 applications maximize operational performance by: We can only imagine what this software could do in 20 - 30 years. It could actually manage organizations, factories, navigate naval ships and space ships and be in charge of many small systems. The future development of this particular system or another of this kind could lead us to creating self-diagnostic and eventually automatically preventive intelligent systems on space ships.
If we now use those AI systems for diagnostic purposes, then in the future there will be no option for human intervention. For instance on the future space ships, if this kind of system detects äbnormal interaction" caused by human, it could take over the command of the ship.
The abilities to make conclusions by AI systems could be provided using evolutionary algorithm (EA). It is an algorithm that uses ëvolutionary techniques inspired by mechanisms from biological evolution such as natural selection, mutation and recombination to find an optimal configuration for a specific system within specific constraints."[9] Many fields of research are based on those algorithms, for example evolutionary robotics (ER). ER is the methodology which allows to create robot controllers, consisting of artificial neural networks, for autonomous robots. As we can see, these are edge technologies which actually can be really seen in our near future. For example it is highly probable to have an AI System on a so-called colonisation space ship, which builds outposts according to the needs of the data chain. It has its plan and it optimizes itself in order to follow the plan and to achieve the goal. As another example we can imagine AI selection system, which can "choose" the best species for a specified environment. The use of EA in the future could be practically anywhere and its future could affect the future of AI systems.
Let us concentrate on one of those algorithms called genetic algorithm.

3.1  Possible development of Genetic Algorithms

Ä genetic algorithm (GA) is an algorithm used to find approximate solutions to difficult-to-solve problems through application of the principles of evolutionary biology to computer science. Genetic algorithms use biologically-derived techniques such as inheritance, mutation, natural selection, and recombination. Genetic algorithms are a particular class of evolutionary algorithms."[10]
GA is a computational model, different examples of which allow us to learn more about ecosystems, the immune system, cognitive systems and social systems. Such algorithms are used in different AI systems, for example for optimizing power plants and developing jet turbines. The purpose of such algorithms is to find the best solutions under very difficult calculational circumstances and in very complicated cases. The pioneer in creating genetic algorithms is John Holland, professor of electrical engineering and computer science and professor of psychology at the University of Michigan.[11] In the 1960s, Holland created the field of genetic algorithms, the purpose of which was to show that computers could ëvolve" their programming to solve complex problems in some ways that even their own creators did not fully understand.
Those algorithms are the base of a study called Artificial Life.[12] This study is the research of "life" through the use of human-made analogs of living systems. Based on such algorithms different examples of simulations were created which can fall under the definition of artificially generated life. One of such simulations is called "Tierra".
"Tierra is a computer simulation developed by ecologist Thomas S. Ray in the early 1990s in which computer programs compete for central processor unit (CPU) time and access to main memory. The computer programs in Tierra are evolvable and can mutate, self-replicate and recombine."[13]
What we already have today is a powerful instrument to create logically thinking, self-evolving environment. The future area of application for these technologies could be everywhere, from self-repairing vehicles to self-building houses. As for other aspects of Artificial Intelligence, such as sensibility of machines, according to John Holland, it is possible but not yet existing.[14] We will have to wait and see what the future holds for us.

4  My personal point of view

With the development of different AI systems humanity will have to accept both the positive and the negative consequences. One cannot expect that the inevitable major changes in the everyday life will all be beneficial. The intelligence created in machines will sooner or later create a life of its own. There are already samples of Artificial Life: life that is being created in programs and different types of software. I believe that in our future there will be no more programmers, the programs will evolve themselves, creating a new, better chain of life. And so on and on.
Humans' irrational logic and hardly predictable actions will be replaced by logical decisions of machines. At the beginning we will let AI machines decide for us on non-significant matters, but we will learn to depend on them more and more.
The ability of AI systems to think "logically", to presume and to predict things will allow them to self-develop into something like a "machine-civilization". We have samples of machine-civilizations already, with the only difference that those are currently being created and controlled by us.
I think that soon in our future the biological phase of life will be replaced with another phase, our era will become the past as the era of dinosaurs. And, as I see it, it is an evolution. Biological life is just a chain, after that will come another chain, more efficient and resistant to constant, day-to-day changes in our world. There could be many mid-stages between us and the machines, such as cyborgs. But enhancing human abilities in order to compete with machines will turn us ourselves into machines. I believe that already in this century it will become possible to enhance almost every human ability, beginning with memory and ending with replacing our brain with silicon chips. I can imagine that someone who will visit a future hardware store will not seek RAM, HDDs or CPUs for his computer but for himmself. Beginning from that stage, our future is nothing but our gradual conversion to machines, where "we" are only the temporary matter of providing the base for the next chain.
The intelligence we seek in machines could actually be found, completely different from human intelligence, but still an autonomous engine that could create civilizations. As I see it, the humanity is making a mistake by trying to compare two different kinds of intelligence. At some point, humanity will fail to understand machines. It has already begun, most people do not understand how their toasters work, and as for larger complex computer systems, it is almost impossible to find a technically-educated person. We are already at a stage where we trust computers enough to build them with a drive, comparable to an instinct, that constantly stimulates them to search for conclusions and to then create tools to follow these conclusions. What it leads to is that computers do no longer follow the in-programmed formulas in order to find a solution, but find their own ways of solving a given problem. The string of thought that the computer followed while extracting the solution is practically impossible to trace. When we use AI software based on different types of evolutionary algorithms (for example), then with high probability no human can know how the eventual conclusion was reached.
In our quest in the name of making our everyday life easier, we might just reach the opposite. We are only building the base for our future. But will it still be ours?

References

[1]
A.I. In Space: Past, Present and Possible Futures, by Jonathan V. Post and Dr. Donald D. Rose
http://www.magicdragon.com/ComputerFutures/SpacePublications/AI_in_Space.html, checked on 01.11.2004
[2]
Isaac Asimov, Ï, Robot", Spectra Books, 1994
[3]
Site of Robot Asimo, Honda Corporation, asimo.honda.com, checked on 01.11.2004
[4]
Site dedicated to android-robots, www.androidworld.com, checked on 01.11.2004
[5]
Marshall Brain, Article "Robotic Nation", http://marshallbrain.com/robotic-nation.htm, checked on 01.11.2004
[6]
Kevin Warwick, Interview to BBC,
http://news.bbc.co.uk/hi/english/static/special_report/1999/12/99/back_to_the_future/kevin_warwick.stm
checked on 01.11.2004
His personal web site : http://www.kevinwarwick.com, checked on 01.11.2004
[7]
Article from Space News,
http://www.space.com/spacenews/businessmonday_041018.html, checked on 01.11.2004
[8]
Gensym Corporation web site, www.gensym.com, checked on 01.11.2004
[9]
Wikipedia The Free Encyclopedia, http://en.wikipedia.org/wiki/Evolutionary_algorithm, checked on 01.11.2004
[10]
Wikipedia The Free Encyclopedia, http://en.wikipedia.org/wiki/Genetic_algorithms, checked on 01.11.2004
[11]
University of Michigan College of Engineering, http://www.engin.umich.edu, checked on 01.11.2004
[12]
Artificial Life, www.artificial-life.com, checked on 01.11.2004
[13]
"Tierra" computer simulation, http://en.wikipedia.org/wiki/Tierra, checked on 01.11.2004
[14]
ScienceDaily Article, http://www.sciencedaily.com/releases/2003/02/030214075837.htm, checked on 01.11.2004


File translated from TEX by TTH, version 3.64.
On 14 Nov 2004, 21:02.