AI is making things better, but worse
We thought AI was going to make life easier, giving us time back by automating away the monotonous work. But just like the Industrial Revolution, factory automation, and robots didn’t lead to giving workers a life of ease and luxury, knowledge work automation through agentic and generative AI is no Caribbean vacation; it tends to leave us with dopamine “crashes” and mental fatigue. Both of those lead to increased brain fog, exhaustion, and irritability.
Before we go any further, let me preface what follows with the caveat that I do not consider myself a neuroscience enthusiast, let alone a professional. This is my rudimentary understanding, and it is likely missing nuance or is simply incorrect. If you have more experience in this field than me, and you find my explanations lacking, please send me an email and I will update this post accordingly.
When you click the button to send off the generative AI to go do its work, it’s a little bit like pulling the lever of a slot machine. What you get out of it could be terrible, or it could be great. Life changing, even. More often than not, it’s just good enough that we feel like we are winning: The report numbers are crunched, the software is written, the email is polished, the artwork is generated—in quantities far beyond what we could previously accomplish. But there is a cost. We pull the lever, getting close to what we really want, but never quite reaching the peak of excellence. It’s good enough that we go back for more, but not satisfying enough to get us to quit. And so we pull the lever again, and again, and again.
My understanding is that this is called a variable reward schedule1. After a prolonged period of operating in this mode, our brains calibrate themselves to this, and the rewards for normal routine activities are dull by comparison. Irritability, exhaustion, lack of motivation, and poor or impulsive decision-making set in, and it takes time to recharge and let our brains calibrate back to normal before we can be decent human beings again.
A related second problem is simply mental fatigue. When you make a series of high-impact decisions, perform qualitative analysis on a large volume of content2, or attempt to multitask3, it has a similar effect, at least in my experience. I know less about the biological and psychological mechanisms on this one, but they’re real.
AI work involves all of these leading causes of mental fatigue. Entrepreneur and software engineer Steve Yegge, in his widely circulated Medium post, The AI Vampire, writes:
I’ve argued that AI has turned us all into Jeff Bezos, by automating the easy work, and leaving us with all the difficult decisions, summaries, and problem-solving. I find that I am only really comfortable working at that pace for short bursts of a few hours once or occasionally twice a day, even with lots of practice. … I’m convinced that 3 to 4 hours is going to be the sweet spot for the new workday. … Assume that exhaustion is the norm. Building things with AI takes a lot of human energy.
Because we automate the busy work, it doesn’t give our brains time to recharge from the fatigue. Furthermore, we’re constantly switching from task to task as our agents happily work away. I’m currently learning how to protect that energy and keep some reserve in the tank for my family during evenings and weekends.
You could try to sustain your multiplied output at full capacity and end up disconnected from reality, exhausted, angry, and unhappy. I’ve recently been filling my evenings with LLM conversations, music generation, and vibe-coded side projects only to yell at my kids and say rude things to my wife because my brain is too tapped out to be compassionate.
We have to find the right balance between productivity and rest—both at work and at home. We can’t go full bore at work and then expect to be able to have the energy and motivation to accomplish the things we need to get done at home. What is that balance? It obviously will be different for everyone. But we have to be intentional about it. We can’t just let it happen to us, or we’ll burn out and be left with nothing.
-
B. F. Skinner showed that when rewards come after an unpredictable number of tries (he called it a variable ratio schedule), people keep going much longer. Because the next reward might come at any time, it encourages repeated behavior. This is why slot machines, doom scrolling, social media reactions, and now generative AI are so addicting. This is different from the variable interval schedule, which is time-based. Because the number of tries is proportional to the number of rewards—albeit variably—the variable ratio schedule is much more likely to increase engagement. ↩︎
-
John Sweller’s Cognitive Load Theory, developed in the 1980s, is helpful here. It’s the idea that we have a limited capacity for working memory, and when we exceed that, it increases stress, reduces learning, and our attention fragments. It’s the same reason a small code change is easier to review than a large one. Generative and agentic AI defaults to producing large changes. There are probably ways that we can tune our context so that the models return smaller changes at a time, but then we drift further into rapid task switching (see footnote #3) and reduce our overall productivity. ↩︎
-
Psychologists generally distinguish true multitasking from rapid task switching. When people switch back and forth between tasks, performance slows and errors tend to increase because the brain has to reorient each time rather than doing both demanding tasks at once. Furthermore, while the research is still in the early stages, there are indications that the negative long-term effects include weakened memory, reduced concentration and increased anxiety. ↩︎
Reader Comments
Evan Lutz:
I agree 100%. I briefly mentioned it in footnote #1, but the slot machine analogy certainly applies to those technologies as well. I think the reward from AI is stronger, though—the content is original (for some definition of that word) and personal—so the negative effects of that slot machine are stronger as well. Combine that with winning more often than your typical slot machine, and that makes generative AI a powerful drug.