Nowadays the vast majority utilise the term "Artificial Intelligence" very casually as a buzzword. Additionally, the generally comprehended definition of the term Artificial Intelligence has evolved over time. Today what we imply when using the term A.I held a different meaning when it was used in the 1960s. Artificial Intelligence has progressively evolved since the time of its inception. Its evolution can be classified into 3 distinct stages:
- Rule Based Intelligence
- Machine Learning
- Machine Intelligence
Rule Based Intelligence
The Rule Based Intelligence, or as I like to call it, the “Old School A.I” was initially designed to be able to perform tasks such as perceiving objects in a picture or understanding content, basically, problems that could effortlessly be performed by solved by the human brain. The advances were gradual and thus the outcomes were discouraging. It was deduced by researchers that in order for a computer to be "smart", it would require acquiring a reservoir of information. Thus, the "expert systems" were born.
By asking an array of questions, the expert system would be able to draw conclusions to a problem as it has been equipped with a database of rules provided by domain experts. A common example is WebMD, a web service that performs medical diagnoses. To solve a particular issue, the Rule Based Intelligence system can be greatly customised, however, this technology lacks the ability to learn independently and therefore does not quite resemble the general human intelligence.
Machine Learning
Once the brink of the Rule Based Intelligence had been identified, researchers began examining the concept of neurone models for artificial intelligence. As a result of combining this concept with back propagation, the emergence of Artificial Neural Networks (ANNs) was formed 50 years ago. Today we can see that the core design of the ANNs has hardly changed even though advancements in physiology and neural anatomy have been made. Therefore, today, the relation between real neurones and the architecture of ANNs lack similarities.
Machine Learning comprises of statistical and mathematical techniques including Artificial Neural Networks. Machine Learning is able to analyse results by deriving data from a reservoir of information thus identifying patterns. This technology has been able to solve many issues where the Rule Based Intelligence was unsuccessful. Although its accuracy and effectiveness can fall if it is not provided with an adequate amount of data for training the program. In the case of dynamic patterns in which data is continuously changing, machine learning is unable to perform here and this is its major drawback.
Machine Intelligence
This is where the next step of evolution comes in, that how can we transform towards systems that exist without human supervision as they start to learn from data themselves. This concept is modelled after the human brain. It is believed that machine intelligence is based on a technique used by the brain to illustrate information called sparse distributed representations (SDRs), which are fundamental for creativity linguistic reasoning. SDRs are the core foundation of Machine Intelligence and so cannot simply be added onto to Machine Learning. The other fundamental characteristics comprise of the following:
- Learning must be continuous
- A fundamental attribute of all learning is Behaviour
- An arrangement of patterns is Memory
Machine Intelligence would ultimately be able to identify aberrations and form predictions through its ability to understand the structure of streaming data.
Summary
The term A.I is often used by people to refer to all 3 of the above mentioned evolutionary stages. This causes great confusion about this fascinating field of Computer Science. The term "Machine Learning" comprises of the simple neural models like ANNs and Deep Learning which are basically machines that learn from data. The term "Machine Intelligence" refers to machines that continuously learn and predict anomalies. By modelling after the human brain we now have an idea where to focus our efforts in the future and slowly but surely we will one day achieve absolute intelligent machines.