Will ChatGPT and AlphaCode replace programmers?

Close-up of a person typing code on a laptop

AIs compete with humans to write code.Photo credit: Getty

Artificial intelligence (AI) researchers have been impressed by the capabilities of AlphaCode, an AI system that can often compete with humans in solving simple computer science problems. Google sister company DeepMind, a London-based AI powerhouse, released the tool in February and has now published its findings in Science1showing that AlphaCode beats about half of the people in code contests.

And for the past week or so, social media users have been intrigued by the ability of another chatbot, called ChatGPT, to occasionally produce meaningful-sounding (and sometimes sublimely ridiculous) mini-essays — including short computer programs — on demand. But these cutting-edge AIs can only perform fairly limited tasks, and researchers say they’re a long way from replacing human programmers.

ChatGPT, the latest version of a natural language system from San Francisco, California-based software company OpenAI, was released on November 30th. Both ChatGPT and AlphaCode are “big language models” – systems based on neural networks that learn to perform a task by processing huge amounts of existing human-generated text. In fact, the two systems use “virtually the same architecture,” says Zico Kolter, a computer scientist at Carnegie Mellon University in Pittsburgh, Pennsylvania. “And while there are of course small differences in training and execution, the main difference, if any, is that they are simply being trained on different datasets and therefore different tasks.”

While ChatGPT is a general-purpose conversational engine, AlphaCode is more specialized: it was trained solely on how people answer questions from software writing contests. “AlphaCode was designed and trained specifically for competitive programming, not software engineering,” says David Choi, research engineer at DeepMind and co-author of the Science paper, told Nature in an email.

READ :  7 Jobs with Canada's Best Employers for 2023 to Apply for Now

Human needs

Researchers have pointed out that a lot of the work that goes into any major software development project — designing a web browser, for example — is understanding the needs of the people who will be using it. These are difficult to describe with the simple, machine-readable specifications that an AI can use to create code.

Kolter says it’s unclear if it will ever be possible for machines to create large software systems from scratch. But “my best guess is that tools like this, which can create parts of a program, will probably become ‘second nature’ tools for programmers,” he says.

“We hope that further research will lead to tools that increase programmer productivity and bring us closer to problem-solving AI,” says Choi.

Kolter adds that there are already some AI tools good enough to make life easier for programmers, such as Copilot, a code auto-completion service launched last year by code repository GitHub and on based on OpenAI technology.

Leave a Reply

Your email address will not be published. Required fields are marked *