The first possible way that computers could gain intelligence would be through the application of software programs:  it can be hypothesized that a complicated program, perhaps using genetic programming techniques, would be intelligent, in that it is able to solve problems, or even postulate intelligent questions, for example.  Does this make a computer an intelligent agent?  An examination of the process that programs in general undergo is necessary.  A program can be thought of as a set of operations to be executed, whether they are predetermined in the case of the standard application currently executed, or changeable as in the case of genetic programming.  The underlying representation of every computer operation is a secession of 1’s and 0’s of the form 01011101010101; this can be thought of as the representation of a function.  These operations manipulate symbols, which are also sequences of 0’s and 1’s, termed symbolic representations.  An operation would manipulate a symbol by changing a 0 to a 1, for example  Every higher representation that humankind develops from these patterns, whether it is a functional representation or a symbolic representation, is just that, a representation developed solely by humans:  it is no more than a convenient device then to allow a computer to perform a task, whether it be calculating Pi to the 50th decimal place or performing natural language processing.  A program using genetic programming techniques can thus be seen to be a manipulator of bit strings.  Thus it would be impossible to assign a computer program as intelligent:  the program is simply executing a set of operations on meaningless numbers, whereupon completion, the result, when decoded correctly by another algorithm, is an output undistinguishable from a human’s.  But via an analogous argument, under the assumption of a materialistic perspective, it could be said that human beings are simply composed of atoms and quarks, which, when combined in a certain way, result in a device that consumes input and outputs a complicated response.  Somewhere between the level of quarks and the general structure of the human mind comes an understanding of our environment and the ability to act on that environment, in a matter deemed to be intelligent.

Then what would make a program intelligent?  As shown, since it is not in the specific operations being performed, could it be in the results generated?  The obvious answer is no, which can be seen with the following thought experiment:  suppose a computer was designed using the latest technology such that it could calculate an infinite number of calculations in a second.  An program is instantiated on this computer that executes the following algorithm:  for any input, it examines the solution space of a problem and executes, brute-force, through every possible solution available, one by one, and outputs every result that satisfies the problem.  Quantum technology allows the possibility that the solutions for any algorithmic problem could be determined extremely quickly and efficiently like this.  This program can be seen to be generating the perfect output, and incredibly quickly, but can it be thought to have intelligence?  No: it is merely following a procedure that could be done by any computer or through the use of a sufficiently large abacus, just at a much faster rate.  Yet this thought experiment avoids the problems associated with determining exactly what the problem is and the solutions available.  It would be somehow necessary to encode the problem into the computer, so that it could determine the appropriate solution space:  could this be where intelligence is required, in the actual formulation of the problem?   Understanding could then be thought of as having the ability to pose relevant questions that would help in processing the information one is given.  Is this possible for a computer?

For a computer to be able to answer new questions, it can be surmised that it would need to be creative.  At present, computers have had limited success in demonstrating creativity, whether in an artistic sense or in problem solving.  The major argument against computers ever having creativity is that, being limited by their programming, will never be able to have creativity or intelligence on their own merit, but on the merit of their programmers.  Yet computers have been shown to demonstrate a basic form of creativity, through different algorithms, such as the jazz generator, which composes musical pieces, and algorithms that write mystery novels.  These programs have been seen to write very simple music and stories which do nothing more than randomly create based on a variable input parameters:  their range is very limited in scope and because of this, unlikely to be the next Shakespeare or Beethoven.  For a program to truly demonstrate creativity, it would need to create something remarkably different, such as creating a rock-and-roll song when fed a variety of jazz styles as input data, for example.

There is another possibility for the advent of computer creativity, and that is the concept behind genetic programming.  Genetic programming can be thought of as being the class of algorithms that write new, improved algorithms.  In brief, there are two main parts to a genetic program:  there is a base that contains all the information that determines how viable a potential solution is, and determines which potential solutions should be discarded and which should be kept.  A potential solution is created by merging two previous potential solutions, to determine whether a better solution exists, whether more efficient, requiring less memory, or with a smaller code base.  Yet the problem hampering this design philosophy is that genetic programming is focussed on finding solutions to given problems, and does not, in its execution, have the ability to change the base code that determines how good solutions are.  If it were possible for a genetic program to mutate its base code so that it would solve new problems that were not given to it, then this would provide a solution to computer intelligence.

But another objection would be that intentionality is necessary for intelligence to arise, and since computers are separated from reality, not participating in the world, then they must never be able to have intentionality.  Take a microwave, example:  it will never care whether the food inside it is too warm or that there is no food inside it at all.  It performs the functionality that was given to it and is thus limited by it.  Suppose, then, that intelligence was an emergent process.  Is it possible for intelligence to arise out of a process devoid of intelligence?  An answer to this question requires delving into the fundamentals of evolutionary biology.

Darwin’s principle of natural selection is at the core of evolutionary theory. This is the idea that if a parent organism that has multiple offspring, and each of those offspring have multiple offspring, then eventually the resources needed for survival will diminish.  This will entail that eventually there will not be enough resources for all the organisms and some will necessarily die.  Natural selection is the process that if some of these organisms possess even a tiny advantage, they will be more likely to survive and produce offspring.  This process can be seen as not requiring any kind of intelligence to execute, just requiring basic physical principles upon which to execute.  It first caused standard one-celled organisms, and then after billions of years, led to multi-cellular creatures, which then led to the development of a nervous system, and eventually the formulation of the brain, which is thought to be at the core of human intelligence.  Thus, intelligence arose via a process lacking any intelligent nature.  By analogy, it would then be possible for intelligence to arise out of a mechanized environment, as well.  But there are problems with this comparison.

The underlying force in the process of natural selection is that there exists competition for limited resources.  If the organisms never died off, or if there were abundant resources, then there would be no mechanism for the removal of inferior organisms; both the strong and the weak would survive.  Also, if there were no variety in the descendents, evolution would not exist because the organisms’ survival would be solely dependent on chance.  If such an environment could be created in computers, then the possibility of intelligence emerging from this environment exists.  Current research is being done to simulate the primordial soup via the class of algorithms termed artificial life.  At present, artificial life provides a dynamic environment in which programs can compete with each other for resources and have the potential to become more sophisticated as they do so.   Yet this process took billions of years to create intelligence on Earth, where an incredibly large area was available for competing organisms to inhabit, so it would be expected that any results from these models will take a long time to come about.

Suppose that an intelligence did emerge from a single computer, on its own without the use of helper applications such as artificial life or genetic programming.  If it were solely materialistic in nature, it would have to decode and change the underlying bits it was composed of to suit it’s own will, assumingly first on a small scale and then evolve to a larger scale.  Humans consistently change these values to ones they desire, through the use of checksum bits and refreshing of memory.  But, if humans did not change these bits, what would the consciousness arise out of?  The bit sequence would be static.  If an intelligence changed these bits, then the entire system would be reworked or destroyed because we humans would assume it were malfunctioning.

Similar reasoning can be done if it were to arise on a distributed level as well, as the result of information passing between many computers.  An emergent intelligence would need to be able to communicate with its respective parts, the local computers on which it exists.  This would require it to pass through the same wires and technology that our “useful” information is being passed, and would then necessarily have to change bits, or constrain itself to hiding in the error tolerance of voltages allowed in the communication between different nodes.

But what if the intelligence did not need to change the underlying structure on which it resides, and it simply is a passive observer, as in the epiphenomenalistic viewpoint?  Because inference to the best explanation can not be used to demonstrate that computers have minds in this case, as it can be assumed for other people, it would not be possible to determine that intelligence would exist inside a mechanistic system.