Is artificial intelligence ruining attention?

When you clicked to read this post, was that a rational decision? Or was it intuition?

A classical psychological topic is rationalisation. That is, we tend to construct the reasons why we did something after the fact.¹ Not before. When we present our thoughts, we present the course of events in the opposite order. We say, "I thought about this and then I took action." As if one's actions were merely the logical response based on a coherent view of the world in our mind. Nothing could be further from the truth. famously divided the mind into System 1 for immediate responses and System 2 for the thoughtful and effortful thinking responses.² The insight is that most of life happens in System 1 with the occasional System 2 interjection. And for good reason: it takes much more mental effort to think than not to think. And not thinking in this case is acting on emotion or intuition. A well-trained System 1 can do wonders.

Developing as a programmer takes time. Some part of that time, especially in the beginning, is spent tinkering. Initial tasks are often more hands-on, at the keyboard implementing something rudimentary. Humble beginnings. When the more or less mechanical parts are in place, attention is turned to how to structure and best implement whatever thing is going to be turned into code. At university, with some ten years of programming experience already, I spent long hours pondering different aspects of the problem at hand and of the implementation. What was the optimal solution? Of course there is no such thing. There are only variants that fail in different ways and with different frequencies. But a lot of time was spent on finding abstractions and comparing different abstractions. Maybe there was another way that was better in some shape or form than the immediately obvious one? Suffice to say that the actual software took a while to come to fruition.

I've often looked back at this time in anger. Well, not really anger, but not understanding the process I was going through. In a more senior mindset, thinking too much and striving for the optimal solution became an anti-pattern. If there's no implementation at all, we are as far from a perfect solution as we can get. But really this was pivotal in using System 2 to train System 1. To spend a lot of effort on a problem to enable intuition and generalisation. That is, spending less effort and energy coming up with and implementing a solution. With AI agents, this might have gone a step too far. When I have a problem to solve, programming or not, I tend to hand it off to an AI agent instead of imagining a set of potential solutions or strategies to reach a solution. And I certainly don't produce a number of potential solutions and ponder their different properties and which I should engage with.

Recently I've observed a new trend in my daily life. I find myself actively or passively waiting for the AI agent to finish whatever it is the AI is doing. When I get tired of waiting, I switch to or start another task. I've waited hours and sometimes days for AI training to complete for various projects. Sometimes I notice that I'm busy with nothing at all; my cognition needs rejuvenation, but I kick off another task, and then another task again. This mind shifting, hopping between tasks, splits attention and focus, spreading them thin. When writing this, I'm momentarily out of tokens from the great AI provider in the cloud. So I thought I could say I'm not task shifting at the moment, at least. But alas, I am mistaken. I have an AI agent cleaning up my download folders and another looking through media and documenting it. Both served by a local Qwen.

Working with AI agents, ChatGPT and Claude has eroded my attention. That is, I have ruined my attention by not interacting with AI agents in a way that's more suitable for me as a human. Instead of fruitful thinking, I'm often finding myself engaging with outputs, sifting through the resulting tokens and evaluating a ready-made answer. Deciding what to do with it, deciding if it's what I need. Is it in the realm of the response I wanted? Do I need to iterate or restart to get the kind of output I need? Post hoc steering. Far from upfront thinking and planning. Am I creating my own cognitive debt? I get a lot of things done, but if you ask me how, I might not know how we got there.

There are two parts to this mind puzzle: the first is that I skip thinking deeply about the problem and get instead to attend to managing the outcome. The other is that handing off problems to the machine makes my mind restless and wanting to attend to something. Since I have handed off the task at hand, the mind wants to attend to something else, which implies shifting to another task. Task shifting is (in)famously draining. Sustained, demanding cognitive work can induce mental fatigue.³ Multitasking does not come for free either: when it involves switching between tasks, it can slow performance, especially when the rules are complex.⁴ There is a counterpart in computer science: Amdahl's law. For a fixed workload, the part that must be done sequentially limits the speed-up available from parallel processing.⁵ The analogy is uncomfortable. I can run several agents in parallel, but the work of deciding what to ask, reviewing their output and deciding what to do next still comes back to me. Perhaps I am the sequential part. And switching between their outputs adds a cost of its own.

There is a clip of Simon Willison at the Pragmatic Summit where he says that switching between three projects with coding agents leaves him mentally exhausted after about two hours.⁶ He is so productive and uses his mind's abilities to their limits. He needs to take a break and do something else. And he really should, considering that he has staggering productivity and probably does more than ninety-nine percent of the population, at a relentless output pace.

When I complete a task or solve a problem I am somehow convinced that it was deliberated, planned and executed by myself. But was it? Something is missing, or at least I am missing something. I used to think, then do. I want to get back into that mindset.