Artificial Intelligence

 

 

So, what is Artificial Intelligence aka AI?

AI is the with the ability to learn or solve problems or understand the structure and meaning of language. My interest in AI started in the early 1980’s at University. I worked on an application which asserted that learning was the ability to modify the definition of knowledge is a key characteristic of acquiring knowledge.

There have been waves of interest in AI over the past decades but never achieved its functional goals or sales objectives.

Think about driving a car. We have developed self driving cars which are essentially rule based application such as “Rule Infront101: If there is a car 100 metres in front slow down to the cars speed less 2 kilometers per hour”. Replicate these rules an apply them to a scenario such as “In front” as well as numerous other rules “Lane Change, Exit Sign, Exit” etc. and the car will give the allusion of a thinking machine.

This autopilot provides the allusion of intelligence through the implementation of a complex system of rules.

Just one change in the scenario which has not been programmed into the system will yield a condition where the car does not know how to drive.

In a Tesla, if a condition is uncounted that the system cannot handle, the command and control if the car handed back to the driver. To do so in a moving automobile means that driver has to be on alert to take over. There are numerous cases involving environmental conditions such as deep snow, landslides and mud that are unlikely pre-programmed into the system. Another example is where a new detour or change from a two way to one way street was temporarily implemented to handle a water main break. If the system has not captured the new rules for the street being driven on, the autopilot will fail.

My point about AI is that to make it semi intelligent requires a huge amount of engineering to be able to capture the environment it is working in and have a mechanism of handing the system back the user when the system is not working. Changes to the basic assumptions in the system will require reprogramming so the system behaves as it should in the changed world.

It is also unlikely the system will be able to implement the basic set of rules out of the box but rather require changes to implement the purpose of the system desired by the customer.

Suppose we developed a system to produce a summary of a large text document. The system would first need to understand the discourse. If it is a complex technical or medical document, the system would need to be programmed to understand each term such as “TNM Classification of Malignant Tumours” and its relevance to the article being produced.

My key point is the effort to keep the machine’s knowledge base system current and handle exceptions to the rule amounts to process of capturing how humans learn to understand what is being read.

Humans have spent a long time learning how to understand written text.

This is no simple task.