Notes on Intelligent Machines

by Keith Vander Linden, Spring, 2001

The pages of science fiction are filled with intelligent machines. Some are robots with human-like appearances and others are disembodied computer programs. Some are friendly; some are hostile. Many of these machines have worked their way into popular culture. See if you can identify the following examples along with the movie or television series from which they come⁰:

A.	B.	C.
D.	E.	F.

We'll have occasion to discuss each of these machines somewhere in this article, but for now note that while their appearances may differ, they invariably exhibit what we would call "intelligent" behavior. For example, robot F, whom most will recognize as Lt. Commander Data from the Star Trek Next Generation series, can do it all. He can move around in a human-like manner, formulate plans of action, and, like most other self-respecting robots, speak English.¹ We find nothing truly remarkable in this. Though such robots don't exist today, it's not hard to imagine building them one day soon. After all, what could be easier than walking, planning and speaking, particularly given the capabilities of modern technology? Most any child can do these things, surely computers could as well. It's no surprise, therefore,that science fiction writers dreamed them up in such numbers. Neither is it any surprise that producers included them in movie and television scripts. They don't even require special effects, all that is required is to wrap some guy up in aluminum foil and have him go around beating people at chess.

It is interesting to note that early computer researchers held the same optimistic view of the potential for intelligent machines. By the late 1940's, existing computers were being used to break secret codes and to perform arithmetic calculations, all in a fraction of the time required by humans. By the late 1950's, researchers had begun to turn serious attention to the problem of implementing "intelligent" behaviors. This new field of research came to be known as Artificial Intelligence,or AI.

Early work in AI produced some tantalizing successes. Arthur Samuel built a checkers playing program that was able to improve its own behavior through practice.² It eventually beat Samuel himself and went on to defeat a nationally known player. Newell, Shaw, and Simon built a program that proved theorems in mathematical logic, called the Logic Theorist. One of the proofs that it produced was somewhat more elegant than the version included by Russell and Whitehead in their classic Principia Mathematica.³ Emboldened by their success, Newell, Shaw, and Simon tried, albeit unsuccessfully,to publish a paper describing their work with the Logic Theorist listed as a co-author. These where heady days of enthusiasm, as evidenced by the following, now famous, 1958 quote from Herbert Simon:⁴

It is not my aim to surprise or shock you--but the simplest way I can summarize is to say that there are now in the world machines that can think, that can learn and that can create. Moreover, their ability to do these things is going to increase rapidly until - in a visible future- the range of problems they can handle will be coextensive with the range to which the human mind has been applied.

Unfortunately, most of this hype was grossly premature. While these early programs were able to do arguably interesting things in narrow domains such as chess or logic, none was able to exhibit any generally intelligent behavior like human-like motion, realistic problem solving, or natural language processing. None came, nor indeed has yet come anywhere close to the level of sophistication of Lt. Commander Data.

Basic intelligent tasks, which are so easy for most humans to do and so easy for writers to include in a script, have proven to be profoundly difficult for computers. Simply put, being human is a lot harder than it looks. Alan Perlis once quipped "A year spent in Artificial Intelligence is enough to make one believe in God."⁵ If there is one single lesson to learn from 60 years of research in AI,it is that humans are profoundly complex creations, beautiful in their ability to be flexible, insightful, and innovative. The Psalmist expresses this as follows:

For you created my inmost being;
you knit me together in my mother's womb.
I praise you because I am fearfully and wonderfully made;
your works are wonderful, I know that full well.
- Psalm 139:13,14

AI can be seen as the attempt to plumb the depths of this beauty. As is commonly the case in science, we cannot begin to appreciate the complexity of God's creation without studying it carefully and attempting to build realistic models of it. As an example of this, consider the following problem in a simple world of blocks. You are initially given a table with 3 blocks on it, as shown below on the left. Your goal is to rearrange those blocks so that they appear as shown in the goal state on the right. You may only use one hand, and may only pick up one block at a time:

Initial State:

Goal State:

The best solution is to perform three actions in sequence: (1) Put block C on the table, (2) Put block B on block C, and (3) Put block A on block B. Pretty easy, right? Perhaps it is for a human, but for a computer it's harder than you might think. There are subtle complexities in this problem that prevented the AI programs developed in the 1960's and early 1970's from solving it at all. These programs characterized the solution as a conjunction of two separate sub-goals: (1) block B must be on block C,and (2) block A must be on block B. The program would then try to solve each sub-goal separately. Try to do this yourself. First try to get block B onto block C. This turns out to be very simple,just move B on top of C. That achieves the first sub-goal. Now try to achieve the second sub-goal of having A on B. This isn't too hard either, but first you need to "uncover" block A so that you can move it. You can do this first and then put A on B. Unfortunately, in uncovering A, you messed up the solution to the first sub-goal, B is no longer on C. This problem occurs regardless of the order in which you try to solve the sub-goals. A good solution to this problem was not found until the mid-1970's. As you can see, seemingly"simple" tasks are not so simple as you might have thought. You may be able to do them "without thinking", but that doesn't mean that they are easy. Myriad problems such as these render the robots we've discussed rather more fiction than science.

The goal of AI is to understand and implement these sorts of intelligent behavior. Doing so would help us to appreciate the nature of intelligence more fully, and might produce some useful tools as well. This is a rather difficult goal, however, given that no one really knows what intelligence is. We all have a general notion of what it is. We presume to know who has it and who doesn't. We'd probably recognize it if we saw it. Unfortunately,building an intelligent technology requires that we specify the behaviors in detail. AI researchers have attempted to do this in a number of key areas, including:

Reasoning and Problem solving Humans are able to reason about their world and to plan their own actions accordingly. This encompasses a variety of behaviors, including game playing, mathematical theorem proving,and the planning of actions. Robot B from above, the nameless but endearing robot from Lost in Space, was able to identify trouble and cry "Run,Will Robinson, Run!". This required reasoning capabilities. Deep Blue,a modern chess playing program developed by IBM, exemplifies reasoning as well. It defeated Garry Kasparov, the world champion, in a 1997 chess match.
Motion and manipulation Humans are able to move around in their world. The area of AI that deals with this is called robotics.Robot E from above, Arnold Schwarzenegger as the "Terminator" clearly exemplified this skill. He was able to walk, run, drive a variety of vehicles and manipulate destructive machinery of various sorts. Implementing this behavior has proven to be very difficult, but a number of practical assembly-line robots have been developed. Huskvarna currently markets an autonomous lawn mower, which, though not so destructive as the Terminator,is considerably more useful.
Perception Humans are able to see and to hear. The areas of AI concerned with these behaviors are called computer vision and speech recognition. Choice A, the electronic eye of HAL from Stanley Kubrick's 2001:A Space Odyssey, exemplified this behavior. It was able to take the crew off guard by reading lips at a critical point in the plot, and was also able to declare in an implacably calm voice "I'm sorry Dave, I'm afraid I can't do that." AI has been successful in fielding real systems that support optical character recognition, friend-or-foe recognition in military applications, and speech-to-text dictation systems (but none that read lips).
Language Processing Humans are able to process natural, human languages. In AI, the behaviors studied include the ability to understand,to generate and to translate human languages. While all the robots shown above speak English, Choice D, C3PO from the Star Wars series, most clearly exemplifies this skill; he can speak 6 million languages. One modern system that exemplifies this behavior is Systran, a machine translation system heavily used by the European Union to translate documents between the various languages spoken in union.
Learning Humans are also able to learn from past experiences.Study of this phenomenon is called machine learning. We are told that when Commander Data was first activated, he had to learn social skills like a child. By the time of the Star Trek: TNG series he had done quite well, except that he never learned how to use contractions. One practical application of this behavior involves data mining, the task of identifying consistent patterns in large amounts of data. Clementine, from SPSS, for example, has been trained to predict the audience-share for potential BBC television shows based on recorded data from previous seasons.

The paradox in all of this work is that many of the behaviors we humans consider the most difficult, have actually turned out to be the most tractable.Chess, for example, and medical diagnosis, have been implemented at levels of sophistication rivaling or even surpassing the capabilities of human experts. Conversely, "common sense" behaviors like walking or engaging in conversation have proven considerably more difficult. It is interesting to note that while Deep Blue, the chess playing program, can formulate effective moves in chess, it can't actually move its own pieces. Indeed, moving the pieces is probably harder than formulating the moves themselves.

Though the term "Artificial Intelligence" is less commonly used today than it once was, research on all of these areas continues, and the profound questions that this field attempts to address remain. Will we ever succeed in building an machine that behaves as a human? If we did, would it actually be human? Is intelligence all there is to being human? What are we to make of futurists such as Ray Kurzweil, who predict that intelligent machines will exceed the capabilities of humans within our life times and are destined to be the next step in the evolution of intelligent life? The Bible teaches that humans have a spiritual dimension,but the nature of the relationship between it and our intelligence is far from clear. If you are interested in pursing these questions, and others, consider going to one of the additional sources listed below.

You may have noticed that we have yet to mention Robot C. This mystery machine is Marvin the Paranoid Android from Douglas Adams's Hitchhikers Guide to the Galaxy. Marvin,a super-intelligent robot that is reported to have "solved all the universe's problems, except his own", lived a pitiful existence and probably wouldn't be surprised to be left until last. He once lamented"Here I am, brain the size of a planet and they ask me to take you down to the bridge. Call that job satisfaction? 'Cos I don't." If his experience is any guide, we need to look beyond mere intelligence to find meaning and purpose in life.

Resources

• Russell and Norvig's text Artificial Intelligence, A Modern Approach, Prentice Hall, 1995 --- This is a good, comprehensive introduction to the field.
• Some of the early papers in AI are still as incisive today as they were when they first came out---Turing wrote a landmark paper on the nature of AI, "Computing Machinery and Intelligence", Mind, 59:433-460,1950. Searle wrote an oft-cited critique of Turing's vision, "Minds, Brains and Programs", Behavioral and Brain Sciences, 3:417-424, 1980.
• There are also extensive materials available on the internet: Carnegie Mellon's AI repository at has an extensive collection of documents and systems. The Kasparov vs. Deep Blue chess match is discussed at length at IBM's chess site. You can tryout a version of the Systran machine translation system at Altavista's Babelfish site. Try translating a simple English phrase into another language and then back to English. The results are usually interesting and frequently comical.

Notes

0. The images were found at the following sites: Robot from http://www.lostinspacetv.com/,Marvin from http://www.wikipedia.org/,Terminator from http://www.stud.ifi.uio.no/,HAL from http://www.tbid.com/hal/,C3PO from http://www.starwars.com,and Data from http://www.startrek.com/.

1. R2-D2 from Star Wars is one notable exception to this rule.

2. Samuel, A.L. (1959) "Some studies in machine learning using the game of checkers" IBM Journal of Research and Development, 3(3):210-229.

3. Whitehead, A.N. and Russell, B.(1910) Principia Mathematica. Cambridge University Press, Cambridge.

4. Simon, H.A., Newell, A. (1958)"Heuristic problem solving: the next advance in operations research". Operations Research, 6.

5. Perlis, A.J. (1982) "Epigrams in Programming", ACM SIGPLAN.