Every computer guru dreams of a computer or AI system that can code itself. This is no longer a just a dream, as it is now becoming a reality. A team of researchers from Microsoft and the University of Cambridge has developed an AI system, namely DeepCoder, with the ability of coding itself.
DeepCoder uses a technique called program synthesis to write its own codes. Primarily, it can write new programs by studying the codes of other software. It also has the ability to evaluate lines of code to determine which one is more relevant to the program it’s writing. The main reason behind this is to train a neural network to determine and predict properties of the program that generate the output from the inputs.
Coding could be for everyone
During a test, it was realized that DeepCoder has the ability to evaluate pieces of code more complexly and better than a human coder. It was also able to combine pieces of code faster and more intelligently than a human coder. Due to its dependency of the deep learning algorithm, its ability to combine and evaluate pieces of code increases with every new problem.
A researcher and member of DeepCoder creation team from Microsoft, Marc Brockschmidt, outlines that the purpose of this development is to give non-coders a chance to describe a program and leave all the building work to the system.
Currently, DeepCoder can only take part in programming challenges with only five lines of code. But its developers are optimistic that in future, it would be able to develop complex programs that can perform routine tasks. They also claim that this kind of technology is not to replace humans, but to handle more complex programming areas so that human programmers can focus on the less complex areas.