Andrea Corbellini: It will take decades to undo the damage done by “AI”

Many business owners in the software development industry are investing a lot
into LLM bots, marketed as “generative AI”. Some of them go as far as forcing
software developers to use such tools under the threat of being fired if they
don’t comply. After all, why shouldn’t they? This technology has a catchy name,
it produces very convincing output that sometimes is correct (or close to being
correct), and companies producing these tools are very good at downplaying
their limitations and at promising that the “next version” will be
astonishingly better than the one before.
My day-to-day experience, and scientific research, however show a quite big
problem: while senior developers get little or no gain from “generative AI”
(1,
2,
3,
4),
junior developers get massive boosts in productivity. Why am I describing a
boost in productivity as a problem? Well, it’s simple: the new generations of
software developers (the ones that are just entering the job market, and the
ones that are still in school) are relying heavily on these LLM tools for
nearly all work-related tasks. And, again, why shouldn’t they? The entire
world is telling them to do so! And as a result of that, they are advancing
without gaining any actual skill.
Me and other (ex-)coworkers have seen it first hand: a junior developer
completes a task using an LLM bot. More senior developers find major problems
with the result. Junior developers go back to the LLM tool asking for a fix,
wasting hours or days without a positive outcome. Junior developers are
becoming unable to perform tasks independently.
In addition to that, there’s a problem that I feel like is not talked about
extensively: these LLM tools can only regurgitate what they’ve been trained
with. They’re good at finding patterns and re-applying strategies that have
been used in prior work, but by their nature they cannot create or
innovate. So, if this trend continues, the new generations of software
developers are not just going to be skill-less, they’re also going to be
incapable of solving new problems that haven’t been seen before.
My prediction? Within 10 or 20 years, the older generation of software
developers will be asked to come out of retirement to fix the unmaintainable
mess created by the newer generations, and to make advancements in the
information technology sector. A bit like old COBOL developers are asked to
come out of retirement to maintain banking systems, but on a much larger scale.
This, unless there will be advancements towards true Artificial Intelligence,
or unless the current trend in education is broken.
Now don’t get me wrong: I’m not opposed to LLM bots (although I’m against
calling them “AI”, because they’re very far from fulfilling the AI promise),
and I think they can be very powerful tools. What I’m worried about is that new
generations are being told to rely on them almost exclusively, and this can
only lead to an evident skill gap. In fact, I think that if there’s a country
that in the future will be able to harness the power of the LLM bots and at
the same time maintain a good enough level of education, thus addressing the
skill gap problem, they will be the dominant economic power of the future,
because they will be able to automate tasks that are time/energy-consuming for
humans, while at the same time use the human brain to innovate and advance.