Programmers have a powerful new tool to make the coding process more time-efficient.
In August 2021, the artificial intelligence (AI) company OpenAI announced Codex, an AI that generates code from natural language. With Codex, programmers can now make natural language interfaces for existing applications. Codex is proficient in a wide variety of coding languages, including Python, JavaScript, Go, Perl, PHP, Ruby, Swift and TypeScript.
To put it simply, you no longer need to know coding languages to make applications. A user can give the AI simple commands in natural language (like “hello world”) and get code corresponding to it. Let’s take a look at the pros and cons of Codex, the critiques plaguing the AI and what this AI means for the future of coding.
Advantages of Codex
The first and the most important advantage of Codex is that it makes coding more accessible. Codex can perform tasks such as rendering webpages, launching web servers and sending emails. With AI taking over all the tedious parts, the programming process is becoming much more efficient. As per OpenAI, Codex has a 37% accuracy rate in performing coding tasks.
OpenAI’s co-founder and CTO, Greg Brockman, says that the company sees Codex “as a tool to multiply programmers” by removing all the drudge work from programming. Codex is also great for beginner programmers since it simplifies the coding process and it is free to try, which makes it easy on the user’s pocket as well.
Limitations of Codex
Codex is great at writing code based on user commands, but it does not know how to decompose problems. Task decomposition refers to the process of breaking down a problem into smaller parts before writing code to solve the problem. A programmer would naturally be better at this than AI because it requires complex cognitive skills.
Codex is still a work in progress. Its code contains errors and doesn’t work 63% of the time. Another issue seen with Codex is that the code it generates can often differ vastly from the programmer’s actual intent.
Critique of Codex
One of the questions that loom over Codex is whether it writes its own code or lifts pre-existing code from its training data. The AI has been trained on open-source code repositories scraped from the web.
As a result of this, many coders have expressed concerns that Codex is taking advantage of their hard work. Thomas Smith, one of Codex’s beta testers, says that coders routinely lift functions from open source code, and thus Codex doing the same shouldn’t be a problem.
OpenAI is hopeful about the future of Codex. Brockman believes that it can solve the problem of programmer shortage in the United States. Given its above-mentioned limitations, it is unlikely that Codex will take away the jobs of human programmers. However, if used judiciously, Codex can certainly make programmers more powerful and coding more efficient.
Header image courtesy of OpenAI’s website
