
No matter how many efforts we make, computers can’t act and think like humans.
After repeated tests, researchers have found new way to improve Artificial Intelligence, based on humans’ learning techniques. Scientists from the Massachusetts Institute of Technology aim to create the first AI algorithm that can actually recognize handwritten letters without taking a long period of time to memorize them.
No matter how many efforts scientists make to create humanlike technology, we still can’t create a program that is capable of thinking. The available algorithms are nothing but copycats of human behavior; therefore, they will never be able to accomplish the same tasks as a human person.
To solve this problem, researchers tried to identify first the underlying causes of this computer drawback. Tests have revealed to them that the slow learning process of computers is the cause preventing artificial intelligence machineries from learning like a human person.
Humans don’t need much time to register and acquire new knowledge. It is enough to show an item to a baby and explain what that object is to him/her and the baby will immediately learn it. AI machines, on the other hand, have to view an object thousands of times until they can actually register it. Even so, we can never be too sure, the computer won’t make mistakes when recognizing it.
Such experiment have helped scientists in the development of the new available technology. Thanks to them, we now have accurate speech and face recognition technology, but the database is still limited. For this reason, MIT researchers used the Bayesian Program Learning algorithm.
The new program was inspired by the Bayesian statistical model, according to which, acquiring new knowledge is made through inference rather than through repetition. This program is much closer to humans’’ learning process; hence, scientists’ decision to choose it.
Unlike previous used AI programs, the Bayesian Program Learning technique was capable of accurately identifying handwritten symbols after seeing them only once. During tests, the program used probabilities to infer the strokes of pen that the computer had to make in order to reproduce the respective symbol.
There are many applications and uses for the new Artificial Intelligence algorithm. Researchers hope the Bayesian model will help them create humanlike computers that can recognize input and provide real output, not just access a database of automatically generated answers.
Image source: www.pixabay.com