sikilmahfud

When we still sat on the High School desk, we were often crammed by a number of new vocabularies. We must memorize tenth of vocabs until night, we feel so sleepy but tomorrow we have English test. But now we have already known much of those vocabs very well. Even though we forget a word and we hear someone say that word, we will think, “Wait, look like I've heard this word before”. Our brain will search the word, match the word with our “database”. Just like my lecturer said, “Human brain is amazing.”

Yeah, I really agree with my lecturer statement. How can we be considered as smart? I ever read that it's not depend on how much data we know, but it is depend on how can we link every data that we have. For example I'll tell you about one word: mountain. If me, I'll think: mountain...hmmm, the mountain I do want to visit is Mount Fuji. Mount Fuji is located in Japan. Japan has so many anime. But watch too much anime is not good because it can make us addicted......Stop here! Mountain and addicted....it has already gone too far!

What if we apply such intelligence into machine?

In the future, we will need smart machine. A machine that's not just run the program sequentially. Do you ever write a program to process factorial number? Or the program that's used to encode and decode message? Those are sequential program, because the program processes data input by a process that has many formulas that are defined before, and then results output. Data input must be complete in order to run the program successfully.


Flowchart of sequential program

Learning machine doesn't need complete data to run, because learning machine has rules. Rules is used to handle and link incomplete data input. Unlike sequential program that must have exact input and output, learning program can handle uncertainty situation. If needed, program can add new rule from its own predefined rules, by the formula that is given before. More often the program used, more rules the program will have and hence the output will be more accurate. Just like human, more experience that we have, we will be wiser.

At the moment, learning machine is usually used in expert system, a system that act as an expert. For example a software that is aimed to ask for patient's disease. At the beginning operation, the software needs human expert involvement to enter enough rules and data that is needed. As time goes by, the software itself can link data according to the rules that the software has without human expert involvement. More rules that the software has, the result is also more satisfied and accurate.

In my opinion, a machine that can learn must be developed further, not just to investigate a problem. We can widen the need of learning machine to other aspects. For example, a robot. In first use, the robot needs human expert to control robot behavior: what robot must do when meet a kind of obstacle, what robot must do when meet another kind of obstacle. As time goes by, no matter where the robot is placed, robot knows what to do according to its experiences before (read as: according to the predefined rules). Moreover, the robot can make its own rules according to predefined rules that the robot has before, make the robot smarter and smarter.

Learning machine isn't as terrible as we see in the Terminator movie. Instead, it will help us very much.