AI Coding Technology Doesn't Pose a Threat to Human Workers

intersexs profile photo

AI Coding Technology Doesn't Pose a Threat to Human Workers A new coding technology, Codex, writes code with AI to tackle programming challenges that would otherwise be difficult for humans to solve. However, according to Tom Smith, who oversees an A.I. startup called Gado Images, the technology doesn't pose a threat to professional coders. In fact, he sees it as a tool that could "make a coder's life a lot easier". Built by OpenAI, one of the most ambitious artificial intelligence research labs, Codex sheds light on the current state of artificial intelligence. Huge leaps and bounds have been made, though most AI systems complement human workers instead of replacing them. With the consistent rise of technology called neural networks, machines can now learn skills by looking at large amounts of data. By statistically computing and analyzing cat photos, a neural network can recognize a cat. This is the same technology that recognizes your voice, translates between languages, and drives self-driving cars. Years ago, researchers at similar labs to OpenAI built neural networks that analyzed large amounts of text, including digital books, articles, and other text on the internet. After computing patterns in the text, the networks were able to predict the next word in the sentence. With some short seed text, the AIs could produce full paragraphs by completing the thought. GPT-3, created by OpenAI, could write Twitter posts, speeches, poetry, and news for example. Surprisingly, the program could even write its own computer code, though the programs were short and simple. The AI had learned from code on the internet. OpenAI took the project a step further, and created Codex, training a new system on an enormous database of prose and code. The resulting system understands both prose and code and cooperates with user input. If you tell it to do something, it will do something. Codex can generate programs in 12 languages and translate between them. But it only writes the right code 37% of the time. It makes mistakes, and it can't reason like a human. It doesn't really think on its own. From a beta testers' perspective, the code that Codex produces is impressive. But it needs some tweaking to work correctly. That means that codex is only useful to an experienced programmer. Using the technology, GitHub, a popular online service for computer programmers, now offers a tool called Copilot that suggests the next line of code, like how autocomplete tools suggest the next word when texting or emailing. It's a way of writing code without writing a lot of code, at least with GitHubs product. It's not always perfect, but it is close enough. Codex could help novices learn to code, and professionals code faster. With his start-up, Gado Images, Mr. Smith set out to sort through photo archives of newspapers and libraries, captioning and tagging the photos before sharing them. But technology could only do part of the job. Its output still needed to be manually reviewed and edited before it worked. And a seasoned archivist still needed to find the best and most important photos. These tools don't completely remove the need for humans. They can be helpful at making our jobs easier though. AI isn't taking all the jobs, instead, it's making all of them easier.

Pinterest logo
Tumblr logo VK logo -

© Uglek, 2021

Terms of Use and Privacy Policy