In the summer of 2021 OpenAI released Codex, a system that automatically generates code as a response to specifications written in natural language. By simply telling the system you want an app that does X and Y, the language is parsed and the actual code is generated.
It is impressive how far Codex can get into developing applications based on simple instructions and requirements. Albeit, many of the projects in which Codex was demonstrated were relatively simple programs, such as to-do list applications, it is still quite the accomplishment considering the infancy of the program. So what does this mean for developers? Will they become redundant?
Probably not, but what is certain is that it will have a huge impact on the way developers work. Since this is quite an extensive topic for a blog post, we will look at one application of Codex that is quite close to possibly having a large impact on the lives of programmers in the near future.
And this application is the Github Copilot software. Github is a company on which development teams can work together by storing their source code and keeping track of the different versions of the code. It is the largest hosting platform of source code in the world. The company paired up with OpenAI, the organisation behind Codex, to use the Codex software as well as their own enormous quantities of data to create an AI-pair programmer. In the development industry, a pair programmer is someone (or now something) with whom you cooperate while writing code. You basically code ‘together’. Github Copilot is already in the beta release phase and programmers that have access to the release can use the software. The program can be plugged in, very simply, into the code editor and a developer can fire away. It works by analyzing the code you’re writing and predicts what else is going to need to be written. It does so by having analyzed billions of lines of code, basically having ‘seen everything’ already. The idea is that it will greatly increase the productivity of a single developer. However, not everyone is convinced. Criticism has arisen about the quality of code and the fact especially new developers do not fully understand the code that is being generated, possibly leading to issues later on. But there is also hope, for many programmers coding is a minority of what they actually do: most time is spent on understanding the problems that need to be solved and then looking at how this can be solved most efficiently. Coding is only a tool to get from A to B. Increasing coding productivity will just get you there sooner.
It remains to be seen how the coding industry will change. I, however, believe that most developers, especially senior ones, will greatly benefit from the rise of AI by reducing the amount of boilerplate and other simple code they need to write. It will allow for greater value creation and businesses will be able to operate their IT departments more efficiently. How do you look at the rise of AI in programming? Do you see opportunities or mostly danger?
Howard, J., 2021. Is Github Copilot a blessing, or a curse? FastAI. [online] Fast.ai. Available at https://www.fast.ai/2021/07/19/copilot/. [Accessed 30 September 2021]
Nice piece Jeroen! It sure is interesting to see that developers themselves also can get assisted by their own written programs. On the one side, I see this as a great opportunity. AI should be able to assist developers in writing relatively easy though time-consuming code, giving them more time to spend on more creative tasks. Furthermore, AI should be able to spot errors written by human developers and correct those more quickly than a supervisory developer would. On the other side, the inherent weakness of AI, namely the possible bias that is in training datasets, remains. Many researchers have stated that at some point, developers will lose grip on how AI develops new code. Then, in the end, AI will become similar humans: excellent at what they do through extensive practice, but always looking towards a personal lens.
Is this in line with your predictions?