Is the end of programming night?
If you ask Matt Welsh, he’d say yes. As Richard McManus wrote on The New Stack, Welsh is a former professor of computer science at Harvard who spoke at a virtual meetup of the Chicago Association for Computing Machinery (ACM), explaining his thesis that ChatGPT and GitHub Copilot represent the beginning of the end of programming.
Welsh joined us on The New Stack Makers to discuss his perspectives about the end of programming and answer questions about the future of computer science, distributed computing, and more.
Welsh is now the founder of Fixie.ai, a platform they are building to let companies develop applications on top of large language models to extend with different capabilities.
For 40 to 50 years, programming language design has had one goal. Make it easier to write programs, Welsh said in the interview.
Still, programming languages are complex, Welsh said. And no amount of work is going to make it simple.
“It doesn’t seem likely to me that any amount of work on improving type systems or syntax or any of that debugging is going to suddenly crack that nut and just make programming suddenly easy,” Welsh said. “We’ve been at it for a while. It’s not improving. So this is where I think there’s going to have to be a kind of a quantum shift to not programming anymore as the way to talk to computers and instruct them.”
It’s comparable to when, for example, only a few people could read books.
“Well, if computing becomes, let’s say, democratized, because now you don’t need to be like a wizard in a tower, who understands how to write Rust code, to instruct a machine, that’s going to completely change that dynamic,” Welsh said. “Anyone will be able to do it. And I actually think that’s a really good thing. You know, there’s all kinds of people in the world and places in the world that could benefit from computing that simply don’t have access to it, because the skill level, the skill set required is just way too high.”
As for computer science, it has always been about humans taking a problem and turning it into instructions for a machine, Welsh said. That’s the definition of computer science. It’s the art of science, mapping problems onto what machines can do. Now that models are getting larger, it’s no longer an x86 CPU running the machine instructions.
“So now your computational core is no longer an x86 CPU running machine instructions,” Welsh said. “It’s an AI model that is solving problems. And, you know, operating and working in the way that like a human might, in a lot of ways.”