The big news this week was a call from tech luminaries to pause development and deployment of AI models more advanced than OpenAI’s GTP-4—the stunningly capable language algorithm behind ChatGPT—until risks including job displacement and misinformation can be better understood.
Even if OpenAI, Google, Microsoft, and other tech heavyweights were to stop what they’re doing—and they’re not going to stop what they’re doing—the AI models that have already been developed are likely to have profound impacts, especially in software development.
It might not look like a regular business deal, but Alphabet’s agreement to supply AI to Replit, a web-based coding tool with over 20 million users, is something of a seismic shift. Replit will use Google’s AI models, along with others, in Ghostwriter, a tool that recommends code and answers code-related questions in a manner similar to ChatGPT. Amjad Masad, Replit’s CEO, tells me that Google has “super cool technology” and that his company can get it into the hands of developers. Through this partnership, Google will also make Replit available to users of Google Cloud, helping it reach more business customers.
The move is particularly significant because Alphabet is squaring up to Microsoft and GitHub, which are likewise using AI to assist coders with Copilot, an add-on for Visual Studio. The same AI that makes ChatGPT seem so clever works on computer languages. When you start typing code, tools like Copilot will suggest a way to complete it.
Alphabet’s move also signals what could be the next big battleground for large tech companies. While so much attention is being paid to ChatGPT parlor tricks and Midjourey 5 versions of Donald Trump, the bigger story is about which company can offer developers the best AI tools—and the new software that developers will build with that AI by their side.
Research from Microsoft suggests that developers can perform tasks over 50 percent faster when using an AI assistant. Companies that offer cutting-edge AI can draw developers to their coding tools and get those users hooked on their clouds and other stuff. Amazon has developed an AI coding tool called Code Whisperer, and Meta is working on one for internal use too. Presumably, Apple will not want to be left behind.
As well as helping developers write code, AI is starting to change the way code is put together. Last week, OpenAI announced that the first plugins for ChatGPT have been created. They will make it possible for the bot to perform tasks like searching for flights, booking restaurants, and ordering groceries. Incorporating AI into code can also accelerate software development. This week Masad of Replit shared a neat example—an app that will turn voice commands into working websites. “We think a lot of software projects will start that way in future,” Masad says.
With things moving so quickly, it’s worth considering what the consequences of rapidly incorporating AI into software development might be. AI tools can reproduce vulnerabilities in the code they suggest that developers may not notice or might be unable to spot. Perhaps developers will become more complacent, or see their skills atrophy, if they rely too heavily on AI. And what kind of “technical debt” might emerge if programmers need to go back and fix software that no human has ever closely examined?
It may not be long before we find out.