A University of Cambridge and Microsoft Research team has developed an artificial intelligence (AI) system that can write its own code. The system is called DeepCoder.
Marc Brockschmidt, one of the creators of DeepCoder at Microsoft Research in Cambridge, U.K., said that the system would allow people who don't know how to code to build their own programs by just describing what they want for it. It has solved the basic challenges that programming competitions have set.
According to New Scientist, DeepCoder uses program synthesis, which is a technique that creates new programs by combining lines of code taken from existing software. The University of Cambridge and Microsoft Research team's AI system learns which pieces of code can be used to achieve the desired result.
One advantage of this technique is that it can conduct a more thorough and wide search than a human programmer. This means that the system can combine source code in a way that might have been overlooked by humans.
Moreover, DeepCoder also utilizes machine learning to search databases of source code. This allows the AI system to arrange the pieces according to its usefulness.
All in all, this makes the University of Cambridge and Microsoft Research team's much faster than the others. It was able to create working programs in just seconds as compared to older systems which took minutes to try out several combinations of lines of codes.
Tech Crunch noted that the system was developed by Matej Balog from the University of Cambridge and Alexander L. Gaunt, Marc Brockschmidt, Sebastian Nowozin, Daniel Tarlow at Microsoft Research. They also co-authored the published paper.
Armando Solar-Lezama at the Massachusetts Institute of Technology, who was not involved with the system, said that this technology would make a significant impact in reducing the amount of effort it takes to code. However, he did add that this automation will not replace programmers - it will only allow coders to devote their time to more complicated work.
