The Economist July 10th 2021 pp71-72 |Science & technology|Automating programming “The software software engineers” “Artificial intelligence is transforming the way computer programs are written”
Source of Image news.mit.edu April 15, 2021
Read The Economist for all detail.
Summary provided by 2244
Imagine being able to enunciate software requirements in spoken English and then having an artificial intelligence (AI)-enabled coding program respond by producing a draft program with those capabilities. This is one of several coding solutions AI-enabled programs are starting to create. Besides coding AI-enabled applications like GPT-2 and its successor GPT-3 from OpenAI (San Francisco) were “trained on an unprecedented mass of text to teach it the probability that a given word will follow preceding words.” Other uses include answering customer questions about products and “powering the speech of fictional characters in virtual worlds.” Of course, the survivalist in us sends up Red Flags about the potential for misuse in the wrong hands such as creating chaos by infusing massive amounts of fake news or flooding “social media with ‘trolling and griefing’ messages.”
Coding is clearly a big application for AI by freeing up the human experts from the boring, tedious and repetitive work of coding. After all everything is derivative anyway and billions of lines of code have been written already for many common tasks. There’s efficiency to consider as well with estimates as high as 50% savings in time but likely at least 10-20%. An example is highlighted from Tabnine (Tel Aviv) using their eponymously-named software. The coder starts with code “to handle...purchase orders"and Tabnine then suggests “code to display product names and prices as well as code to create fields to be filled with quantities, payment and delivery data. It works even though Tabnine has never been specifically instructed to do that.” Others have or are developing similar capabilities including Microsoft with “Visual Studio.”
Even more exciting is the ability of AI-enabled coding tools to detect inadvertent software bugs and malicious code. As it is humans are not very good at finding software bugs. Training for de-bugging software from GitHub (A Microsoft Subsidiary) was based on “at least a billion snippets of code identified as harboring a bug.”
Opmerkingen