Coding isn't dead, but how it's taught needs to change, says Google DeepMind research scientist

Google DeepMind research scientist Stefania Druga wants to encourage kids to use AI to "co-create" rather than to cheat.

And while there's been much debate about whether "learn to code" is still good career advice, she told Business Insider that she believes coding skills will continue to be valuable — but how they're taught in school needs a makeover.

Discussions about AI in education are frequently marked by the potential negatives, including cheating and skill atrophy. Young people, Druga said, are frequently heavy AI users — but they're not always leveraging it for the best ends.

"The way they're using it right now, in my opinion, is really disappointing," Druga said. "But it's not their fault. It's the way these technologies were designed. It's not the best thing that we could do with AI for learning, to make it solve our homework, or write our essays, or help us pass a test."

Druga said she believes part of the problem is the kind of assignments that students are expected to complete.

"We should be changing the whole framework. First of all, if an AI can solve a test, it's the wrong test," she said. "And then second, for like, generating essays — if you have a tool that always gives you pre-baked information, and there's no back and forth of any sorts, that's when you have issues of over-reliance and lack of critical thinking."

Druga said she first realized the need for AI education nearly a decade ago — long before the average person knew what an LLM was.

"Half of the households in the US would have a voice assistant, and people would ask things first to Alexa before asking their parents," Druga told BI.

Druga — also partially responsible for Scratch, a drag-and-drop programming language used to teach kids how to code — created Cognimates as part of her master's thesis at MIT. The program encourages AI literacy by way of engaging children in projects that pique their interest, including building games and programming robots, along with learning how to train AI models.

The idea is to provide kids with a platform to safely experiment with technology that will likely be a huge part of their lives for the foreseeable future.

"It's kind of creating this sandbox or playground for kids to engage in the scientific process, because they formulate hypotheses like, 'This is why Alexa responds in this way.' And then they have a way of testing that hypothesis very quickly," Druga said. "And the same would apply in the era of Gemini, ChatGPT, and large language models. How do we allow young people to create their own GPTs?"

Druga said she designed her own platform to be more socratic in nature — it poses questions to lead users along the right track, without feeding them the answer. And the kinds of problems that children are asked to solve, she added, mean that they're frequently proud, and consequently possessive, over their work.

"They're very attached to their project. It's a big part of their identity," Druga said. "So they're like, 'It's my project. I don't want AI to do it for me.' But when they get stuck, they would love to have someone to help them debug, or to help them find the right block, or to help them even navigate the platform."

According to Druga, the approach to preserving critical thinking while also making sure children begin developing AI literacy "as early as they can speak," should be two-pronged: When used in educational settings, AI models should support "co-creation" rather than helping students reach a finite answer as quickly as possible — and assignments themselves could be designed to be less cut-and-dried.

"The burden shouldn't be on them to always make the right choice, because I think that's too much to ask," Druga said. "If you were given boring homework and a tool that can do it for you, why wouldn't you use it? I don't blame them."

"We need to change how we teach and assess," she added. "But we also need to change how we design these tools in order to make room for young people's agency, young people's creativity."

Coding isn't dead — but coding education needs a refresh

Druga said Cognimates is teaching children skills that will serve them in a labor market that doesn't yet exist. And though she frequently hears that "coding is dead" (which she doesn't think is true) — she still views an education in the fundamentals, which now include an understanding of AI, as extremely useful.

"I think a problem with CS education and computing education for the longest time was that it would focus on market, and kind of preparing young people to get jobs in tech," Druga said. "It was this promise, like, 'Oh, if you have a CS degree, you're gonna have a cushy job and not have to worry about anything."

That's no longer true, Druga said, given the uncertainty in the job market due to sweeping layoffs and concerns about AI replacing human workers, but she believes that was the "wrong kind of goal" to begin with.

"What we're seeing with AI right now — in large language models and other architectures that are coming after large language models — is that the technology is changing so fast that if your entire value proposition for how you train people is to prepare them for specific stacks or jobs, that's going to become obsolete very fast," she said.

Ideally, Druga said preparation for working in tech involves transferrable skills, particularly teaching people to adapt to ambiguity, as the only constant she foresees is "change, and rapid change" — a view that Google DeepMind CEO Demis Hassabis shares. While the traditional tech career path may not be viable by the time the children currently using Cognimates are ready to enter the workforce, Druga believes it also may not be necessary.

"I know personally, at least, like 20 people who are 19 and went from a prototype to having a fully financially sustainable product in couple of months," she said. "They don't even need VC money or a job in tech. They're just building. They identify a problem, they build a solution, they launch it, and they get paid for it. It's profitable. So talk about the future of jobs, right?"

Regardless of what children eventually end up doing with their skills in AI — Druga believes that what's important is that they develop them in the first place.

"The goal is to make sure that everyone feels like this is for me, and they don't feel intimidated, or they don't feel like, 'Oh, I need to have all of this background knowledge to even get started,' because things are changing so fast," she told BI. "I think the AI engineer or AI scientist frameworks are really challenging the way we thought about labor market education, and kind of the pipeline before."