You have 1 article left to read this month before you need to register a free LeadDev.com account.
From ChatGPT to GitHub Copilot, AI is becoming part of the software developer toolbelt. What does that mean for how we teach the next generation of engineers?
“AI in software development is here to stay,” GitLab asserted in its 2023 Global DevSecOps report. Around one in four of those surveyed are already using AI to help them code, with a further two-thirds planning on using it to make their lives easier in the next two years.
By simplifying a variety of programming tasks, Generative AI tools – including coding assistants like GitHub Copilot – have very quickly become useful to software developers of all experience levels, with no sign of things slowing down from here.
This shift puts universities – in particular the computer science department – in a tricky position: Do they embrace the AI revolution and help students use these tools, or should they stick to the fundamentals of coding? And what are the implications for an industry that has traditionally valued elite university education above all other achievements when it comes to hiring junior developers?
This is a debate that is taking place on university campuses across the globe, but can universities move quickly enough to keep up with the technology? “Initially when it came out, some students and staff embraced it immediately,” says Becky Strachan, professor at Northumbria University’s computer and information sciences department in the UK. “Others said they weren’t really sure what it is and would just ignore it for now.”
Stick or twist?
By January 2023, it became impossible to ignore. “Everybody I speak to in the industry uses, almost on a daily basis, various [generative AI coding] tools,” says Strachan. “I don’t think we can ignore it. My personal view is that we should be embracing it.”
Ahead of the new academic year, Strachan and her department had a meeting where some of those holdouts begrudgingly came to the conclusion that they needed to adopt generative AI, or risk being swept away by the wave of change.
“Students and educators need to embrace this change, much in the same way we had to tackle the emergence of Google Search when it burst onto the scene in the late 90s,” says Serge Belongie, professor at the University of Copenhagen’s computer science faculty, and lead of the Belongie Lab there.
Outside of computer science, where adoption of these tools has proved somewhat seamless, Belongie thinks the biggest challenge for higher education will be “the cat-and-mouse game of plagiarism detection software.”
But Belongie has been “pleasantly surprised” at how universities are already adapting to the massive technological shift. Within his own computer science department, the faculty were mostly open to their students using generative AI tools, as long as they “provide proper citation and attribution.”
“The general sense is that there’s no putting this genie back in the bottle, and we owe it to the students to incorporate these tools into their training,” he says.
While the initial reaction to generative AI was to see it as an existential risk to education by lowering the bar to entry, “wide consensus emerged quickly on the fact that AI won’t make us educators useless, and won’t make obsolete the entirety of all our education programmes,” says Luca Maria Aiello, head of the study programme for the master’s degree in data science at the IT University of Copenhagen.
“People accepted that there will be important changes to the way in which education works, but we should embrace those changes and improve our programs accordingly, rather than fighting them.”
That’s what Aiello’s institution has been doing – highlighting the positives of AI to computer and data science studies, while recognizing the drawbacks. AI has the potential to provide a significant productivity boost, Aiello says. “However, more and more education professionals agree that AI tools are way more effective in the hands of skilled and experienced programmers,” he adds.
Programmers without strong technical backgrounds can still benefit from them, but “when it comes to fixing, assessing, improving, combining their output or to integrate it into complex projects, expertise is absolutely needed,” he says.
Just as schools didn’t stop teaching math after the advent of calculators, ITU Copenhagen isn’t getting rid of its fundamental computer science concepts anytime soon, including programming, but that doesn’t mean it won’t have to adapt. There is a clear intention to integrate the use of AI into computer science programmes already, such as teaching students how best to use AI to boost their productivity – though that’s still at an early stage, Aiello admits.
Finding a happy medium
Many universities haven’t yet taken a firm policy stance on the use of generative AI in their teaching and assessment, but are putting together FAQs for students outlining the benefits and risks involved. Belongie points to ETH Zurich’s page on generative AI as an example to follow of helpful frankness in communicating the double-edged nature of the technology.
But for those who have, finding a footing in the new AI-enabled future has involved rewriting how to teach and assess students. For Northumbria University, that’s involved acknowledging that automated tools can often code as well as humans. “You can get something like Copilot to generate code; ChatGPT generates code,” Strachan says. “But what I want to see is students understand the process they’ve been through.” To do that, the university has already shifted a module to include an oral presentation at the end of a project to explain this thought process.
For Aiello and ITU, assessments are also being retooled. “Since AI is perfectly able to solve most simple programming problems, when it comes to examination of students, we will need to make sure that the exams of computer science fundamentals are performed in a controlled environment in which AI cannot be used, to make sure we can appropriately certify competence,” he says.
Further into the future, Strachan believes that actually being able to code will become an extinct skill thanks to generative AI tools. “ChatGPT and other tools can do it better than we can,” she says. “They’re very good at syntax. We get syntax wrong all the time.”
If that is to be the case, it becomes an obvious choice to incorporate how to use the tools – and check their output for errors – into teaching. “We now need to emphasize to students that yes, you can use these tools, but you need to know what it’s doing is okay. And how do you know that if you don’t know the basics?” she says.
Navigating this shift as an engineering manager
Teaching students the basics will be important, but the impact of generative AI on the current crop of computer science students is yet to be seen. If it is able to lower the bar to entry for the software developer profession, what does that mean for the historically ‘safe’, and potentially lucrative, computer science degree?
“Nobody knows, right?” says Matt Welsh, chief architect at AI startup Fixie.ai and a former professor of computer science at Harvard. “We’re going to figure this out as that first generation of graduates starts to come out into the workforce in the next year or two.”
Welsh believes it’s important that using generative AI tools is taught in universities. “I’ve been programming for something like 40 years, and I myself go to ChatGPT all the time with stupid questions,” he says.
He also believes that it could help engineering managers break out of their standard set of candidates. That’s because the old ways of interviewing – getting a candidate to stand at a whiteboard without a computer and hand write code using a pen – will go out of the window.
“A lot of people feel – myself included – that being able to do whiteboard coding does not mean that you can be effective working in a large code base, navigating the complexity of other people’s designs, or using tools effectively,” he says. Instead, hiring managers should look for how well staff can use the tools at their disposal: including those which leverage generative AI.
Generative AI could also finally open up the field to a broader group, and not just those who can afford a university education. “Over time, we’re going to have more and more people entering the field that are coming from very different backgrounds. And I think that’s a good thing,” Welsh said.
