Software engineers don’t have to be worried about losing their job yet as computers become more adept at writing code.

DeepMind is a U.K.-based artificial intelligence laboratory that was acquired in 2014 by Google. It announced Wednesday that AlphaCode, its piece of software, can program just like a human programmer.

The London-headquartered firm tested AlphaCode’s abilities in a coding competition on Codeforces — a platform that allows human coders to compete against one another.

DeepMind, the team behind the tool, said “AlphaCode placed approximately at the level the median competitor. It marks the first-time an AI code generation systems has reached a competive level in programming competitions.” said in a blogpost.

Dzmitry Bádanau, a computer scientist, wrote that “human-level coding” is still light years away on Twitter.

“The [AlphaCode]System ranks behind 54.3% of participants,” he stated, noting that most participants are students in high school and college who are still learning their problem solving skills.

Bahdanau claimed that the majority of people reading his tweet can “easily train for AlphaCode to outperform them.”

Although researchers have tried to teach computers how to code for many decades, the idea has not yet become mainstream. This is partly due to the inflexibility of the AI tools meant to create new codes.

CNBC interviewed an AI researcher scientist who preferred anonymity because they couldn’t talk about the subject publicly. He said AlphaCode is an incredible technical achievement. However, careful analysis of which coding tasks it succeeds on and what it struggles with is necessary.

According to the scientist, AI tools like AlphaCode may change the nature software engineering roles in some way as they mature. But the complexity and human role means that machines will not be capable of performing all of these roles for a while.

Gary Marcus from New York University told CNBC that it could serve as an assistant for a programmer, in the same way that an old calculator may have served an accountant.

“It is not one-stop shopping, that would replace an actual programmer. That is a decade away.

DeepMind is not the only company that has developed AI tools capable of writing their own code.

This was June 2013. MicrosoftAnnounced AI system that can recommend codeSoftware developers can use this information as they are working.

The system is called GitHubCopilot and draws from source code submitted to the code-sharing site GitHub (which Microsoft purchased in 2018) as well as code posted on other websites.

Microsoft and GitHub created it together with OpenAI, an AI research startup that Microsoft purchased. backed in 2019. In 2019, the GitHub Copilot depends on large amounts of code written in multiple programming languages, and huge Azure cloud computing resources.

Nat Friedman is the CEO of GitHub. He describes GitHub Copilot to be a virtual version what software developers call “software creators”. a pair programmer — that’s when two developers work side-by-side collaboratively on the same project. This tool examines existing code, comments and offers one or more lines of text to be added. The model becomes smarter as programmers make and reject changes.

Friedman said that it makes programming faster. Friedman stated that hundreds of developers from GitHub use the Copilot feature every day to code and most of them accept suggestions.

DeepMind published a separate research paper on Friday that stated it had conducted similar tests against OpenAI technology.

Samim Winiger, an AI researcher in Berlin, told CNBC that every good computer programmer knows that it is essentially impossible to create “perfect code.”

He stated that all programs have flaws and would eventually be destroyed by bugs, hacks or other complexity.

Computer programming, in critical circumstances, is fundamentally about creating fail-safe systems that are accountable.

IBM declared in 1979 that “computers cannot be held accountable for anything” and that “computers must not make management decisions.”

Winiger claimed that code accountability has been mostly ignored in spite of hype about AI coders performing better than humans.

Do we really need complex, opaque, un-introspectable autonomous systems that are incomprehensible for most people and incalculable to all in order to manage our critical infrastructure?” He referred to the financial system, food supply chain and nuclear power plants.

— Additional reporting by CNBC’s Jordan Novet.