Toward More Deliberate AI Use

I have been using AI extensively in recent months. I use it to help me write text, develop code, create designs for presentations and posters, and more generally, to answer all kinds of questions about the world around me. I rely on it especially often when I am uncertain about how to approach a task or when I lack energy.

One example of how I interact with AI models is that I copy small paragraphs of text to a chat interface and ask the AI model to “Shorten this slightly,” “Please add…” or “Rewrite this based on the following feedback…”. For code, I often don’t even write my initial implementation – I usually say something like “Here is my code so far: … I would now like to implement…” and then sometimes correct the model by saying “No, please implement this in the following way…”.

This way of interacting with AI has several important issues. Firstly, producing content while using extensive AI assistance is often much less enjoyable on a deeper, long-term level. I feel far less connected to the content I produce by copy-pasting AI-generated text and code. Secondly, some of the content I produce this way is structurally worse. Even though AI models are becoming generally smarter and can hold more information in their context, they still often lack the ability to make good decisions in writing, software development, and many larger projects. Since I am often too lazy (and too focused on details) to revert a design decision once it’s made, the products of ‘my’ work end up being lower quality. Third, I tend to learn much less from creating new things this way – for example, about the specific software framework I use. This sometimes limits my ability to quickly understand and fix issues with current code or text, to iterate fast over multiple possible next steps, and more generally constrains the growth of my skills.

I don’t want to suggest that one should completely avoid using AI models. The ability to get relatively high-quality answers on problems that require substantial context also enables me to learn many new things. I can probably complete more and significantly larger projects with AI’s help, and it’s often very useful to map out multiple approaches to something in detail before deciding on the next step. However, I think the way I interact with AI should become more deliberate.

Going forward, I would like to make using AI a more active decision. I want to think consciously about whether I really want to solve a particular problem using AI, and if so, how I would like to use AI in that context. As a general guideline, I would like to write the first full draft of every text that matters to me myself. When writing code, I would like to write most short snippets of code myself (this might sometimes include rewriting code that an AI model provided in a chat interface, in case I did not know the syntax of the library/framework I am using). Furthermore, every once in a while, I would like to pick some code or text where I am not confident I could have written it as quickly myself, and delete and re-implement or rewrite it.