The Codex AI system is proving itself to be able to tackle coding challenges and other tasks quickly and efficiently. Its abilities are raising the question of whether technology can replace humans for simple coding chores or whether it is simply a tool to make professional coders’ lives easier so that they can complete their work more efficiently.
The range of tasks that Codex undertakes include resolving references, writing code for APIs and performing micro-tasks such as paying out Grande Vegas casino bonuses so that coders are free to move on and perform more complex assignments.
Codex
Codex was built by OpenAI, a research lab which offers insights into the state of artificial intelligence. Codex had noted that, while A.I. technologies continue to improve, even the most impressive systems basically complement human workers as opposed to replacing them. It’s based on a mathematical system called a “neural network” which enables machines to acquire specific skills by analyzing vast amounts of data. For instance, the machines can learn to recognize a cat by analyzing thousands of cat photos.
Many people recognize the technology, if not the science behind it. The neural network is responsible for Internet translation services, powering the identification of street signs and pedestrians by self-driving cars and recognizing commands that are spoken into a mobile device.
Building on this technology, OpenAI and other AI labs have begun to design neural networks that can analyze enormous amounts of material. These include hundreds of thousands of pieces of text that have been posted to the Internet including digital books and Wikipedia articles. The networks work by pinpointing patterns in the text and then using that information to predict the next word in a sequence. Neural networks aim to create systems that will write social media posts, articles, speeches and even poetry.
GPT-3
The system, named GPT-3, powers neural networks by writing computer programs. When OpenAI began their research they saw that GPT-3 could be used to write short and simple programs. This ability surprised the researchers who had created it – they hadn’t anticipated this step. “The thing that was funny for us was to see that the applications that most captured people’s imaginations, the ones that most inspired people, were the programming applications,” OpenAI CTO Greg Brockman said in the video demo of Codex. “Because we didn’t make the model to be good at coding at all. And we knew that if we put in some effort, we could make something happen.”
The researchers moved on the unexpected outcome by training a new Codex system which can understand both prose and code.
Codex is limited. You can tell it to create a code that will show a ball bouncing against a blue background and it will give you that. You can request snow falling against a black background and get that too. But it often makes mistakes, creates programs that don’t run or allows security flaws to infiltrate its programs. The programmer often needs to add a digital code in order to gain access to the Internet service that it was trying to enter or tweak a command to suit a specific software setup.
In short, Codex can’t reason like a human. According to OpenAI, Codex produces the correct code 37% of the time. It can mimic or recognize something that it has experienced in the past but it can’t move beyond that to think on its own. It can, however, generate programs in 12 computer languages and translate between those languages. Only an experienced programmer can extract maximum performance from it.
Benefits
Codex is being examined as a teaching tool that could help teach novices to learn to code. One of the high points of the Codex system is that it generates simple programs from brief English descriptions and explains complex codes in simple English.
OpenAI sees the main benefit of Codex as a tool that facilitates programmers’ work. It helps them work faster. It has the potential to point programmers toward new ideas or give them building blocks to create their own programs. One online service for programmers, GitHub, has been using the GPT-3 technology to power its Copilot tool — an upcoming line of codes that operates similarly to the way that autocomplete tools suggest the next word as you type an email or a text.
Jeremy Howard, founder of the Fast.ai artificial intelligence lab and partner in creation of the language technology on which OpenAI work is based said “It is a way of getting code written without having to write as much code. It is not always correct, but it is just close enough.”
Limitations
But human intervention is still a must.
One client, Gado Images, had been struggling to build a system that would sort through photo archives of libraries and newspapers, write tags and captions automatically, resurface forgotten images and share the photos with other businesses and publications. Gado used Codex to move the project along but the technology didn’t solve the problem of how to find the best or most important photos or how to tag the photos accurately.
In the end, Gado combined the Codex AI with the services of a seasoned archivist. “We thought these tools were going to completely remove the need for humans, but what we learned after many years was that this wasn’t really possible — you still needed a skilled human to review the output,” said Thomas Smith, co-founder and CEO of Gado. “The technology gets things wrong. And it can be biased. You still need a person to review what it has done and decide what is good and what is not.”
While Codex can extend a machine’s abilities, having humans at the controls is still vital to success. “A.I. is not playing out like anyone expected,” said Brockman to the New York Times. “It felt like it was going to do this job and that job, and everyone was trying to figure out which one would go first. Instead, it is replacing no jobs. But it is taking away the drudge work from all of them at once.”
Source:Plato Data Intelligence
