Spray Your Garden With AI, I Trust You Won’t Hit The Neighbours

This message is not for the inspired. It’s for the hopeless. For those who fear that the things they know and love about today will not be available tomorrow.

Because, right now there are a lot of people I know and interact with who are terrified about how AI will replace “them”. The adoption of AI will undoubtedly change things. And while change is not always bad, it usually involves an uncomfortable form of parting ways, to some extent. This is the keyboard I have used to author Software for the last 4 years of my 15 year career.

It has been my experience that since the release of ChatGPT, the panic and tension levels in the software companies I have worked at has been extreme. So I thought I’d share my perspective as a software engineer who has seen a variety of technology implementations and team dynamics over the years, in both the New York and London tech scenes.

Someone recently told me that teachers are extensively using AI to form lesson plans. And my first thought was that if memorization-based education is so easily replaced, why are we still teaching it? And this is not a new perspective from me. I didn’t come to the conclusion that education is somewhat irrelevant today due to a weak reliance on real-world problem solving because of AI. That is a position I have discussed openly for quite some time.

So here is a disclaimer: the following paragraphs are not made up of balanced or objective opinions, they are dramatized in order to form a more effective call to action for anyone who feels that the emergence of AI is a reason not to push on with the vision and passion for software development they once had.

and here goes..

There are so many human problems in how technology is applied today. Mishandled projects that focus on the immediate accomplishments rather than a grand system, tech debt projects that throw away company knowledge in the pursuit of “cleanliness” without diagnosing what was actually dirty to begin with.

Machine learning has been under-utilized for several decades, and while the new breed of “transformer” based generative systems have potential. Adding more data to a system that is predisposed to inaccuracy is like putting your finger on the garden hose to make it go farther. And the idea that more data solves that precision problem, is like increasing the water pressure to try to prevent the unfortunate event in which water sprays over your walls and ruins your neighbour’s barbecue.

And there is no way that AI itself solves that type of problem. If you have driven your car off the road, a bigger engine will not help you. It would be more sustainable to review why, you as the driver, still have a valid driving license.

Large Language Models are the grand babies of Somewhat-Smaller Character Models (spell checkers). And spell checkers are Awesome! no doubt about that… but… you still have to write something.

My advice if you are feeling unnecessary in this day and age is to sit down with your mechanical keyboard and build something intricate. There are lots of technical problems to solve that are not getting nearly enough attention. The implementation challenges of using complex systems to solve equally complex situations is anchored in the ability to leverage your knowledge over the space. Because if the space is complex, the solution will require an equally intricate amount of analysis and attention, before it can be automated.