Do you fear Artificial Intelligence taking over your job? Ironically, programmers, the very set of professionals that build AI, might be the first to be rendered jobless by the invention, according to an update from DeepMind, a subsidiary of Alphabet.
The AI system named AlphaCode is being touted by DeepMind as capable of writing computer programs at a competitive level. All programmers, especially the junior ones, will now perpetually be on their toes if they do not want to fade into irrelevance.
DeepMind confirmed the ability of its AI system by testing it with challenges used in coding competitions. The AI finished within the top 54 percent of coders with flesh and blood. The challenges AlphaCode worked on came from Codeforces, a platform that releases weekly coding problems and ranks coders based on how they perform.
For DeepMind, this result is a remarkable step for unsupervised coding. However, the company admits that the average coder’s skills are not completely replicated by AlphaCode, at least not yet. To be candid, many of the challenges provided by Codeforces won’t even make a difference to an app developer, for example. They are more targeted at algorithms and theoretical computer science concepts that combine logic, maths, and coding expertise.
AlphaCode was given 10 of such challenges, in the same format human coders get them. The AI package then generated multiple solutions and whittled them down, mimicking how humans approach the challenges. Five thousand humans have attempted the same set of problems, and AlphaCode ranked within 54.3 percent of the best responses, giving it an estimated CodeForces Elo of 1238. Of all coders that took the challenges in the last six months, AlphaCode placed with the top 28 percent.
Codeforces founder, Mike Mirzayanov, could not hide his surprise. “I can safely say the results of AlphaCode exceeded my expectations. I was sceptical [sic] because even in simple competitive problems it is often required not only to implement the algorithm, but also (and this is the most difficult part) to invent it. AlphaCode managed to perform at the level of a promising new competitor.”
In an email to The Verge, a DeepMind principal research scientist, Oriol Vinyals, said that AlphaCode is still in its early stages and not ready for prime-time yet. However, with what the package can do right now, DeepMind is closer to actualizing its vision of writing code that only humans can do for now. Vinyals sees a bright future for AlphaCode, “In the longer-term, we’re excited by [AlphaCode’s] potential for helping programmers and non-programmers write code, improving productivity or creating new ways of making software.”
DeepMind is not the only AI company working on autonomous coding. For instance, Microsoft and OpenAI have worked together on adapting GPT-3 for autocompleting lines of code. GPT-3 is a language-generating program created by OpenAI. To illustrate, this is like Gmail’s feature that suggests words to finish your sentences as you type. Coders using Microsoft’s Visual Studio coding environment will also be familiar with this feature.
Autonomous coding has come a long way but has realistically not come close to replacing human programmers. One of the hurdles they face is buggy code and reproducing copyrighted libraries. However, until their abilities develop and all kinks smoothened out, these AI-powered programs can fill the role of an assistant to human programmers by suggesting lines of code, with room for rejecting problematic suggestions.