During a lot of years, philosophers have been trying to define the process of intelligence without a noticeable success. Philosophers and scientists have done a lot of advances on matters related to intelligence as logic or memory, but we do not know much about the origin of the intelligence.
Initially, intelligence was considered a process related to the soul. As soul is immaterial, it was impossible to consider that intelligence could be replicated in an artificial way. However, ancient philosophers were as interested in the analysis of the process of knowledge as in the origin of the universe. Today we know a lot of things about the universe but intelligence is one of the biggest mysteries of the humankind.
The advances in biology have provided a new way to focus the search for an artificial thinker. If the intelligence have a material support based on an interconnected network of neurons, it is possible, in theory, to replicate the process of intelligence.
An old friend of mine always remembered that a perfect description of the problem has got the solution in itself. Intelligence is one of those things that cannot be precisely defined, and there is not a global definition accepted by all the scientific community.
Science always begins with a measurable description, psychologists use the IQ as a measure of intelligence, but this index is more useful to compare individuals than to analyze the process itself because there is nothing that can link the index with its physical support in the brain. A more scientific approach can be the use of modern devices in order to analyze the zones of the brain that are activated under certain stimuli, but this measure is very gross in order to become useful.
In brief, intelligence is a so complex process that today we cannot understand it. Science is making advances analyzing partial aspects of the problem as logic reasoning, language, memory, pattern identification but there is not a global model of intelligence that can provide us with software to be implemented in a computer that can substitute a human operator for any task.
Artificial intelligence can be analyzed from the complexity science for several reasons:
– Complexity is related to functionality and intelligence is the most advanced natural mechanism to provide functionality (adaptability to multiple tasks).
– The complexity of a neural system can be measured in a quantitative way with modern quantitative complexity techniques.
– Quantitative complexity science provides a scheme to analyze the required balance between functionality and robustness.
Now, I am going to put an example of this. Some years ago I was working in a research group that worked producing a 3D perception through virtual spatialized sounds. These sounds are sounds sent by headphones that are perceived as if their sources were on points of surfaces in the outer world.
We found that the first signal integrated as a sonorous stimulus following a periodic pattern produced confusion for the testing individuals as they perceived an irritating noise, however, if the stimuli were sent randomly people could perceive surfaces though this kind of signal.
This experiment gave us a lot of information about the perceptual mechanisms of the brain and how information can be integrated by the brain, but now, I am only interested in the stimulus signal and how can be analyzed in terms of complexity.
The signals tried to transfer the structure of the outer world that has associated a certain amount of complexity; a wall with windows is more complex than a flat wall. In order to be perceived we need to introduce additional complexity in the audio signal because different positions of the virtual sources were sent at different times. The former signal with a periodic pattern is a signal with additional structure and the latter one is a signal with additional uncertainty.
The experiment shows that the human brain is oriented naturally to act as a filter of the uncertainty but it does not work well with the unexpected (and unnatural) structure.
The 3D perception mechanism is global, not point to point although the sensors of the retina are discrete; this is one of the reasons why it was possible to get a 3D perception of surfaces from senses different to vision.
Back to an AI design, many AI systems introduce a lot of processing rules trying to avoid uncertainty. These rules can be providing unnatural structure to the initial stimuli. Looking at the nature in terms of complexity processing, perhaps, this is not the best way to focus the problem.