Growth In The Age Of AI: One Computer Scientist’s Reckoning (Part 1 — How AI is Different)

This is the first of four posts on what AI means for my future growth as a computer scientist. If you want the map of where the series is headed, start with the introduction.

Before we explore what it means to grow in the AI era, we have to understand what makes AI so different from any other automation before it.

Software tooling and libraries have always kept improving, and automation got far easier than it was in the early 2000s. What took a month for me to write in C would later take an afternoon in Python. I still remember the first time I wrote a neural network from scratch for a class assignment during my Master’s in 2009, and then again in 2020 using PyTorch. The experience was wildly different. In 2020 I was happy I could focus on the larger research problem at hand instead of worrying about whether my neural network implementation was correct and optimized. Each improvement and new tool made me faster at things I had already learned the hard way. And I was not worried, PyTorch was simply letting me focus on the bigger problem.

One may argue that AI is doing the same thing, just on a larger scale. Perhaps. But here are a few differences I keep coming back to.

• AI is the first tool reaching higher up the stack, toward the ideation, the reasoning, the intuition, and the judgment I always thought of as the real value I bring. That is what makes the unease different this time.
• AI impacts multiple skills at once. A calculator does arithmetic for you, GPS navigation does routing for you, an AI-based polyp detector flags polyps. Each tool atrophies one specific skill. AI is a single tool that touches nearly every skill I have at once, so the atrophy isn’t localized, it’s broad and simultaneous.
• Earlier I would struggle first, and the understanding would follow as a result: the friction produced the understanding. AI delivers the answer now, and understanding becomes optional, something I would now have to go back and reconstruct on purpose if I am motivated enough. For the first time, I can produce “correct” work I don’t fully understand. This, in my opinion, is the one thing that, if left unchecked, will cause significant loss of skill over the next few years.
• AI fails in the most dangerous possible way. A compiler error or a memory bug is obvious. AI’s failure mode is almost invisible: it’s usually a confident, wrong answer engineered to look right. That’s the hardest kind of failure to catch, and it’s precisely the kind that rewards turning off my own judgment, the very skill I most need to keep it in check.
• I use AI constantly, all day. So whatever atrophy is happening is compounding faster than with any automation tool before it. I have been using it for only two years, and I can already feel that I have lost some of the rigor I used to bring to my work. The dopamine release that comes with seeing working code and beautiful UIs created at the press of a button is so satisfying that I quickly forget the due diligence required to make sure the results are actually correct.
• Moreover, AI is evolving at a rate the field has never seen before. The list of what it can do better than me keeps growing every few months.

So that is the unease. And here’s what makes it more than just my own anxiety talking: this exact pattern, a tool that makes experts more productive while quietly eroding the skills underneath, has already played out in other fields, and we have studies to prove it. We will explore that next in Part 2 of this series.