Here’s DeepMind’s Incredible Code Writing AI Coder: AlphaCode

DeepMind is a well-known industry giant in developing cutting-edge artificial intelligence (AI) solutions to many problems. While many think that only some automated jobs will be replaced by artificial intelligence, AlphaCode may point to a different possibility. Imagine what would it be like if your coder was also a code itself. While program programming sounds like an all-creating idea, it might not be very far away. The technology is still new and insufficient, but it is going in small steps.

Pixabay

What does DeepMind’s AlphaCode do?

It is an artificial intelligence made to analyze existing solutions, analyze and filter them through machine learning to create a solution for a given coding challenge in competitive programming. It creates code snippets from its analysis to a problem with a clear strict definition, simulating the standard critical thinking of a coder. You can find how it approaches a question from DeepMind’s own page for attention visualization. The company claims to have tested it on Codeforces’ competitive programming challenges in their blog post. And they have received astounding results for a baby artificial programmer, with an estimated performance in the top 54% range.

The company also mentions how the program simulates the real code writing process of a competitor. The program thinks about the existing knowledge from GitHub and its own dataset, carefully complies with several solutions, thinks about the top candidates from its solutions dataset, and makes the final submissions after testing and seeing the results. It is really an interesting concept if you think about how it simulates a real coder’s algorithmic thinking. It’s algorithmic thinking, in the end, a coder’s best weapon, but still an algorithmic one.

Pixabay

Some side-thoughts

While some programmers think that an automated coder will be invented eventually and preach that coders should switch to “less automatable” fields in computer science, it still seems too early for that. Even the company mentions that AlphaCode is more like a super-advanced helper to a programmer rather than an experienced coder itself.

Writing elegant and superb code with no vulnerabilities requires more than algorithmic thinking. Another irreplaceable factor in most coders’ eyes is the ability to transform human ideas and projects to the coding environment, which is even hard for normal engineers in terms of providing healthy communication between managers and coders.

However, it should also be said that most people thought it was impossible for computers to achieve certain things back in the day. It is a huge industry with ever-growing research as you may remember from my previous article titled “Meta’s Powerful AI Supercomputer Will Be Fully-Built in Mid-2022“. Therefore, the industry should also take its lesson in saying “impossible”.

In the end, if one sector ends, another opens. Simple codes may be written by machine learning algorithms, but maybe data analysts and AI developers will be in demand, who knows? However, as a developer myself, the future of coding sure seems exciting if we are competing with always-eager-to-learn AIs!