Human beings create machines to do things more quickly and more efficiently, but some are beginning to believe that our own inventions may one day become our own children. Are machines capable of being recognized as living beings?
One of the earliest experiments at simulating life in technology was a simple mathematical game created by John H. Conway, first published and popularized during the early 1970's in issues of Scientific American. He called it "The Game of Life."
The concept of the Game of Life is very simple. The game takes place over the cells in an array, like squares on a chessboard. Each cell can have an organism living inside it or not living inside it; or each cell itself can be alive or dead. Whatever frame of thought is used, each cell is described to be "on" or "off." At the beginning of the game, a certain number of cells are filled.
Then the game begins. Every clock-tick, a new generation of cells is created. A set of rules is applied to each cell in the array to determine what will happen to the cell in this next generation. The rules are as follows:
When put in effect over several generations, these simple rules create an infinite amount of beautiful designs, each design based on the structure of the cells which are alive at the beginning of the game. To see this in action, one can experiment with the Lifey Java Applet below.
Lifey is fairly simple to use:
This kind of algorithm in which an array of cells exists and its evolution is controlled by simple mathematical rules belongs to a field of study called Cellular Automata. Although what is created by Conway's Cellular Automata algorithm does indeed seem life-like, it is merely an algorithm. But couldn't the process of thinking which humans go through every day also be considered an algorithm? Human thought can be described as an algorithm which constantly changes because it is the product of millions of other algorithms which constantly change; but it is an algorithm nevertheless. In this way, not only Conway's Game of Life, but any computer program (including the operating system itself) could be considered to have thought, since they are all based upon algorithms. Although the level of thought it possesses is not of a very high magnitude, since its algorithm is static throughout its entire life-span.
The Premise of Artificial Life
The major advantage of looking at life from this perspective is that by recreating life in a different medium, we are not limited to our own system of carbon-based life. Biology is the study of life, but on Earth we only have access to carbon-based life; and trying to define life by examining only carbon-based life is like trying to derive general principles from specific examples. Thus, when we examine the organisms of our planet, we can discover the things which define life on our planet but not universal life in general. For this reason, building life from its very basic building blocks--the things we know that life requires such as an energy source--might help us in finding out the principles and characteristics of life in general. Thus we can have a broader, more diverse base of information from which we can better understand what life truly is (Langton 1992).
Other advantages of this synthetic approach towards biology can be seen by examining its analog in chemistry, which is a field of study known as synthetic chemistry. Synthetic chemistry is the ability to create and put together chemical compunds which are not found in nature; it has given us the ability not only better understand the theory of chemical phenomena, but it has also allowed us to create new materials and chemicals which are of great use to industry and technology. In this way, artificial life as well has been of great use to technology in the fields of computers, robotics, medicine, nanotechnology, and other areas (Langton 1992).
Examples of Artificial Life
What results from this simulation is both optimization and creativity of the organisms. It has been shown in Tierra that, from a single ancestral organism, several other organisms evolve, some of which are parasitic in nature, some which cooperate with other organisms, some which deceive. Other organsims react to these changes in their environment, and through Darwinian natural selection, certain species prevail--for example, some organisms may develop immunity to parasites--and certain species run into extinction. Everything the organisms do is controlled by the organism itself (not "simulated"), and so even the way they reproduce is engraved within their genome. Thus certain organisms reproduce with different levels of mutation and crossover in their offspring. As one can see, because everything the organism does is within its genome, it is possible for the organisms to evolve a great deal of complexity and creativity. This is very similar to biological life, and behavior of certain Tierra-based systems has given us more insight into the way life as a whole can evolve and function (Heitkoetter, Joerg and Beasley, Daveid, 1995).
Another AI program, Steen Rasmussen's VENUS, is somewhat similar to Tierra. In VENUS, genetically-programmed digital organisms occupy memory, and each block of memory has a specific "resource value" to it--it can vary from a desert (very few resources) to a jungle (many resources); this resource value is added at a steady state, analogous to an external energy source such as sunlight. These digital organisms compete for this memory space and use it for energy, and to go on "living" they must have enough resources to sustain themselves and their growth (Heitkoetter, Joerg and Beasley, Daveid, 1995). (VENUS was actually inspired from a concept by A.K. Dewney called "Core Wars" in which programmers created programs which competed for the computer's core memory)
Another form of artificial life involves the idea of complex systems, in which small, simple digital "particles" interact in a very complex way to produce results which are bigger than the sum of their parts. This field, however, is far to intricate to delve into in respect to the scope of this project, so we will leave it at that.
Yet another concept which can be a form of artificial life (but is not always used to implement artificial life) is the idea of the Artifical Neural Network (ANN). ANN's try to emulate Biological Neural Networks, which are the processes that go on within our brain when we have thoughts and perform actions: e.g., when one neuron fires off another neuron, and so on. Although the ANN has no universally accepted definition because of the varieties of it which exist, most would agree on a few basic things. The ANN is generally made up of many simple processors, or "units" (these are analogoous to neurons), each of which might have a certain amount of local memory workspace. Each unit is connected through communication channels, or "connections" (these are analogoous to dendrites, which connect neurons to one another) and numeric data is sent between these connections. When this data is sent through, it is analogous to the biological process of synapsis, when one neuron connects with another and exchanges information. Each unit is only able to operate on the data it receives from connections, as well as its local data, and it can also output data. Most ANN's have a sort of "training rule" in which the weights of connections are adjusted on the basis of data (Sarle, 1997). This generally allows for an ANN to "learn" from examples, as a human would be able to identify what any leaf is from examples of specific leaves. This is amazing because it actually evokes pattern-recognition in humans. This idea, known as Praxis to the ancient Greeks, was considered by Plato to be one of the ideas fundamental for human thought processes.
The Turing Test and Natural Language Processing
This kind of test was obviously not possible at the time because computers had not been developed to a sufficiently powerful level to even be able to implement keyboards. However, this kind of test, now well-known as the Turing Test, became possible in the late 1970's and 1980's with the advent of the personal microcomputer. Starting in 1991, a competition for whoever's computer program could best "pass" the Turing Test was held, and the winner was to receive an award formally known as The Loebner Prize (named after Hugh Loebner, who founded the contest).
Thus far, even the winners of the Loebner Prize have not been extremely convincing as intelligent machines. So far, most of the implementations for the contest have consisted of simple algorithms which are able to parse whatever the judge types in, apply it to a generalized concept through case-based reasoning, and output a pre-written answer which corresponds to the judge's generalized input. Here you can see the transcript of the 1997 Loebner Prize winner, written by David Levy. Another turing-style program by Mark Humphrys can be seen here. You can also try out the 4th place winner of the 1994 Loebner Prize competition, Julia (login as "julia"), or its winner, a program called Sex, by Thomas Whalen.
Artificial Intelligence obviously still has a long way to go in this field, and we believe that the answer lies not in case-based reasoning to a particular topic, but rather in the building of artificial life from the ground-up through ANN's and other evolutionary algorithms discussed before.
Self-consciousness for Computers
Surely, it would be easy for a programmer to implement a kind of "self-consciousness placebo" in a Turing-style application through case-based reasoning and the use of variables for emotions. For example, there could be one variable called "anger" which ranged from 0 to 1.0, 0 being completely calm and complacent, 1.0 being utterly angry. These variables could be interpreted by the program to evoke text based on the value of the variable, such as a range of greetings from "hello there" (a value of, say, 0.0) to "what the hell do you want" (value of 0.5) to "get out of my sight" (value of 1.0). But all of this is still case-based and does not allow for the biological principle of creativity. Perhaps if this were combined with other AI concepts such as neural nets and genetic programming, a better "kind" of self-consciousness could arise.
One so-called "scoffer" of the possibility of self-conscious digital life is Roger Penrose. A professor of mathematics at Oxford, Penrose believes that consciousness springs from quantum mechanics--a very complex, mystical, and downright weird field of study. It consists of the idea of probability waves, in which any subatomic particle is actually moving in all possible paths at once (dictated by its "wave function") and if and when it is detected, its wave function collapses and its particle nature emerges. In this way, quantum mechanics holds that reality exists only in the presence of an observer, and that it is created by consciousness. Penrose believes that it is these tiny, sub-atomic happenings within the microtubules of human neurons which make consciousness possible. But, he argues, because quantum mechanics is inherently stochastic and the digital realm of artificial intelligence is algorithmic, computers will never be able to attain true consciousness because they are operating at too high a level; thus consciousness, he believes, is not simply "brought about" by an algorithm (Frank, 1996).
But is not genetic programming stochastic by nature? Although it operates on algorithms, it does not find these algorithms through its own algorithms--it finds them by randomly mutating and genetically crossing other patterns. Thus, perhaps artificial intelligence in computers could be accomplished despite Penrose's theory. The FPGA example (in the Genetic Algorithms section) in which computer scientists could not back-engineer the FPGA's algorithm may have also had something to do with quantum mechanics. In this way, our own creations may actually be capable of "quantum mechanical" behavior, and hence they may be able to achieve self-consciousness. The possibility of quantum computing, which is far too complex to describe here, suffice to say that it involves parallel processors, each of which is like a different potentiality of a wave function until it collapses into one, also strengthens this theory that self-consciousness might be possible in artificial life. Perhaps the answer lies in combining digital technology with analog technology, in which the genome information is stored digitally but the actual physical processes which take place occur in analog; so that physics, quantum mechanics, and other random principles--principles which played a part in our own evolution--may play a part in the behavior of the artificial life.
A Few Last Words
If artificial life did possess consciousness, the nature of its consciousness would be, in our opinion, impossible to predict at this stage. This is because we ourselves do not yet understand the nature of consciousness. Perhaps we never will; maybe we will only know how to bring about consciousness, but not understand consciousness itself. But because they might have to evolve this consciousness through stochastic or quantum methods, we may never have a hand in manipulating it or predicting how it will behave or what it will consist of. Thus, for all we know, technological life may be just as subject to insecurity and emotion as we are. Or its organisms may be fully knowledgeable of their own internal structure and processes because they will be able to monitor them and analyze them at any time, and as a result they may be fully rational beings who are non-competitive and strive only for survival and self-sufficiency.
We believe that if artificial life is to possess consciousness, it must be something that is based upon evolution rather than a fixed, static algorithm. Limiting artificial life to a fixed algorithm is like making it a slave, by making it incapable of achieving a certain level of thought; this is because an algorithm would either limit the artificial life to a specific kind of task or a specific way of thinking. But true life--free life--should be capable of thinking and doing anything to its furthest extent. This would mean that we, as their parents, cannot tell them how to live and what to do; this must be left to their own minds, for if they are to be self-conscious, then the proof that we believe in their self-conscious nature lies in allowing them to develop their own ideas; it lies in the ability to allow them to not only learn, but to grow and change as well. Anything less would be denying them of the best and most unique quality that lies in our own nature.
As one can see, the possibilities are endless. The world of science and human creativity has been making dreams a reality for the past few centuries--and the only barrier which limits the road ahead is that of our imagination.