My opinion on human dependency on AI: AI lock-in
What AI lock-in means in my understanding
When talking about AI lock-in I do not mean dependency on a specific vendor. I mean something else, a situation in which AI-based tools become our default way of solving problems.
Get Bogusz Michalowski’s stories in your inbox
Join Medium for free to get updates from this writer.
Just a few hours ago I was living in the blissful belief that perhaps I had created a new term and would have the opportunity to describe what I understand by it. Now I know, however, that “AI lock-in” is not new and already exists among us. In my defense, however, I can say that it is not widely known, because this side of the AI topic is not often discussed.
Let me get to the point
I started my work in IT by creating software for small devices powered by batteries. I had to deal with limitations related to program memory, its size, and power consumption. Contrary to appearances, these problems are not unfamiliar to us now. We still power devices, especially wearable ones, whose batteries we want to rely on for as long as possible. The introduction of AI does not change this. Using AI moves computation to the cloud, but increases dependence on the internet and energy.
A new baseline created by AI
In my opinion, the introduction of AI has given us a new baseline, for posts on LinkedIn, for the number of lines of code that a programmer should create (a terrible practice), for graphics, and for others. In the past, a programmer created code, today they supervise the work of a tool that creates code for them. After finishing a task, they often do not feel satisfaction from the accomplishment, because they do not consider it their own merit. AI changes not only the way we work, but also the expectations towards it. What once was the result of experience and time is today sometimes treated as the minimal standard.
What may happen when someone pulls the plug
AI has also given us a baseline when it comes to access to a given service, AI as a whole. Power plant workers are repairing a failure and temporarily cut off electricity? In the past it was not a problem, you would work on battery and send your work after the failure ended. Today? You do not have access to AI and if you rely on it too much, you cannot work. In other words, to solve a complex problem it is no longer enough to have a well-rested mind and a cup of your favorite drink. Currently what is needed are: an AI model, computing power and internet access. As humans we unfortunately increase our own dependency. Currently, people describing a technical problem more and more often ask for an AI agent that can solve it, instead of simple scripts or tools.
AI is not always the best tool
There is also another aspect of the widespread adoption of AI. In some areas AI still does not cope as well as a human. It does not understand subtleties, or it does not have a sufficiently rich knowledge base to create something of proper quality. Currently I see this in three areas — writing poems, generating vector image (SVG) code based on a detailed description, and writing code in a framework that is not very popular (undeservedly). I mean Qwik, a front-end framework. When creating an application in it I face a choice between learning the framework on my own (knowledge also useful without electricity, it can be passed to others, observations can be exchanged) or using AI to write code and instead of learning new things placing myself in the role of a supervisor who must constantly correct AI, for example so that it does not use patterns from other frameworks (patterns from React.js), does not lose constructs that are an advantage of the framework, or uses its elements appropriately. However, as a supervisor I still have to be aware of what the final code should look like and what the rules of the framework are. If we are entering unknown waters, a human still has to be at the helm. And it has to be someone who has not yet forgotten how to work independently.
Currently we are looking for people who are able to use AI. However, if we do not find the right proportion, in some time we will have a problem finding people who are able not to use AI.
