AI, please automate my job

Today was one of those difficult days at work. I wanted to timestamp this moment, and my sentiment towards AI.

It seems there are two main camps on this subject. The polarization that reigns in politics and culture has also reached to technology. The two positions can be summed up as:

A) AI is just a fancy parlor trick. It fundamentally changes nothing. B) AI poses an existential threat to humanity.

I remember when my family got a dial-up modem in the 90s. I was hooked. I was 10 or so and knew it would change everything. I remember being 12 and ordering by email magic tricks from a store in another city 500 kilometers away from home. I knew that this would catch on. It was so convenient that it was a matter of time… this way of doing stuff would storm over the world. And it did.

Soon is gonna be two years since I first used ChatGPT. And I really wanted to be blown away by it. I wanted to feel the same I felt about Internet, the smart phone, the Kindle and so many others wonders technology has gifted humanity. If so many smart people are talking about AI being an existential threat then no doubt it can automate my job, if not completely then at least most of it.

I really tried, many many many times, different AI iterations, different application contexts. I keep hearing: this model performs 99% better than the previous. And then, some months later, the same story. And then all those stories about fake demos… Recently I tried an RGA tool with openAI paid tokens. In it, I hooked a simple pen-and-paper RPG ruleset to see if I could get a chatbot that answers questions for the game director during sessions. And again: the moment you know something about the subject you ask about, the answers look like one of those papers you present in school without studying and that you improvised in the hallway five minutes before handing it over. At first sight, they look ok… until you start paying attention. (Why AI can’t sometimes answer: “Hey, I’m not sure about this”?)

One of the things I like the most about reading and listening to pundits is collecting their predictions. I keep a list of predictions from each of my favorite economic analysts and then as time goes by I like to revisit them. What interests me is not the result they predicted but rather the reasons they put forward to support their predictions. I’m not a luddite, I’m completely into technology, and I want to be wrong, but so far I feel more in the ‘parlor trick’ camp than in the ‘Skynet’ one. Let me share my reasons.

This post is meant to be a time capsule in which I express some hypotheses about what might be going on:

a) I’m stupid. I simply lack the proper cognitive/technical skills that would enable me, as a developer, to unleash the power of AI. In the next couple of years better, simpler tools will be launched and my day to day will be dramatically impacted. I could probably get this result now, but again, I’m a limited person.

b) AI was initially promising (ChatGPT release back in November 2022 when it made world headlines), got the whole industry into a bubble and now everyone has the incentive to remain bullish. And by the whole industry I mean from the very big names up to the content creators that are happy about the clicks and views AI bring (and all those snake oil salesmen that sold/still sell juicy workshops/consultant services to so many companies). They thought this would improve exponentially and when it didn’t they panicked and decided to double down and keep pretending that the king was not naked. A bit like Sabine Hossenfelder explains theoretical physicists do as they keep collecting increasingly more money for increasingly bigger colliders to discover new particles that maybe (probably) don’t exist.

That’s it. It has to boil down to these two. As of today I assign a prior probability of 70% to a and 30% to b. In two years I will adjust these in a Bayesian manner.

If it’s Skynet what lies in the future, then ok, we will all die or go back to harvest potatoes. In its own way it will suck, but at the same time I will be amazed. Forget about pet projects code for online stores, to-do list apps, instagram or snake game clones. If AI can automate the absolute irrational chaos of my daily job as a Java back-end developer, if it can make sense of the hundreds of tables, dozens of repositories, deprecated documentation, ad-hoc unspoken agreements, arcane, idiosyncratic workflows stemming from the collaboration and ever changing needs of a myriad teams and hundreds of employees chipping in over the course of many years. All that human mayhem that I’m positively sure is part of every company comprised of human beings, then so be it… in awe I will march to the fields and get on down with those potatoes.