Researchers at DeepMind have developed an algorithm that generates computer programs from natural language requests.
Reinforcement Learning
They describe an approach for generating artificial intelligence (AI) software using a technique called reinforcement learning. Their paper was published in the same issue of Science magazine as a perspective from another researcher who explains some of the challenges involved in developing AI programs.
Creating computer code to create a software program or app that carries out the desired function may vary from simple to extremely complicated.
Over time, humans have streamlined the approach to creating more sophisticated programs in an effort to enable individuals or groups to create them efficiently.
Recently, computer scientists have started considering the possibility of having computers do programming.
If computers were able to write code on the fly, anyone would be able to ask their computer to write specific code just for them. However, scientists must first figure out exactly how to accomplish that feat before they can begin working toward making that happen.
Google Brain
The current hope is for deep-learning neural networks to learn how to program themselves. If they can teach themselves how to write programs, then they could review their own output and the outputs of others, and use that knowledge to get better over the long term. That’s basically what Google Brain is trying to do.
AlphaCode
A software development company has developed an AI system called AlphaCode that listens to a human reading out instructions and then thinks about them before doing something.
Then, it must turn the user’s request into a set of instructions that can be translated into computer language. Once it has done so, it must then turn these instructions into a program that can be run by a computer.
We have some exciting new things coming up. Keep reading!
This whole process sounds familiar to computer scientists, who use similar techniques every day when writing programs.
Instead of teaching their systems how to code by giving them examples of good coding practices, the researchers instead let them observe real programmers working together and then taught them from there.
It seems to be working so far, but we’re not sure if it will actually get us any results.
What is your view on this technological development? Tell us in the comments section below.
Visit our blog for even more amazing tech happenings!