Original article was published on Deep Learning on Medium
The turning test given by Alan Turing gives us a satisfactory definition of intelligence. It is being said that a computer will be considered intelligent if an interrogator couldn’t distinguish the response from a Human and a Computer. But physical presence is not necessary because physical properties don’t measure intelligence. On the other hand, In the advanced Turing test, a video signal should be used for perceptual testing of the computer program. In simple language, one can consider an intelligent computer program that could be able to do the majority of tasks rather than just doing one task(Like game playing agents that are too good at playing a specific game only). At the time being one can represent a computer program doing all the tasks by using different methods and strategies as:-
a.) Natural Language Processing:- to be able to communicate with others in English(or another language).
b.) Knowledge Representation:- to store observable data of an environment as using some knowledge representation.
c.) Automated Reasoning:- to be able to generate answers or reasons by using the stored information or knowledge.
d.) Machine Learning:- to detect patterns in the data using the data taken from the environment.
We can say that the computer program is acting like a human but is that a computer program is intelligent enough to think like a human. It is obvious that if someone wants to determine whether a computer program is thinking like a human or not, one needs to have a full understanding of how humans think? This is being done by using three strategies:-
- To understand human thinking by observing human thoughts by using psychology experiments.
- Observing a person in action.
- Observing the brain in action(Brain imaging).
Once we have a precise theory of mind then it could be converted into a computer program. By checking Input-Output of the computer program and then I/O of human thinking it can be assured that some similar computer computation is happening in the human brain too.
The Greek philosopher Aristotle was one of the first who attempted the “right-thinking”. He gave us a method of drawing a conclusion from a set of premises. Forex. All humans are dumb, Ram is human thus Ram is dumb. Here, conclusion:- Ram is dumb to have been drawn using two premises that are 1. All humans are dumb 2. Ram is human.
This methodology provided us the Logic and later on, the relations between premises and objects are drawn using quantifiers and predicates that gave us the FOL(First Order Logic). FOL is used today in designing intelligent agents in an environment. But there are limitations too, one of the limitations is that execution of FOL in an intelligent agent requires huge computations.
An agent is something that acts. And from a computer agent, there are more expectations than just as to act.
A rational computer agent is a computer agent that acts in such a way that it always achieves the optimized or best result and in uncertainty, it is expected to achieve the best result.
Turing test also helps agent to achieve rationality. Because knowledge representation and logically thinking or reasoning will get an agent to its best place and actions can be explained by NLP.
Hence, turning test seems more reliable to measure the intelligence of a computer program.
At the end, you can see that along with learning State of the art technology we must have knowledge that what AI researchers want to achieve.
In the next article, we will talk about “Various inter-disciplinary fields that are important for the AI”.