Until relatively recently in history (relative meaning, the last few centuries), people slept in two segments. One began at dusk that lasted until the middle of the night, an hour or two awake, and then a second round of sleep after that. According to the BBC, this wasn’t orchestrated or unusual but a regular part of life, with people using the wakeful gap for everything from chores to prayer to socialisation.
Apparently, our 3am restlessness isn’t restlessness at all. It’s our light bulbs, that let us stay up far past dusk. (And our blackout curtains letting us sleep past dawn.) We have mostly adjusted, though middle-of-the-night insomnia is so commonplace it is literally more accurate to call it normal. But we’ve adjusted, more or less.
So if AI is forcing some new adjustments to fairly primal instincts, there’s evidence that we can do it. And, it’s worth acknowledging the insomnia is the echo that persists.
One HBR study points to a potential shift in our cognitive limits due to AI — a so-called ‘brain fog’ that comes from the fatigue of managing AI, at least when we stack multiple tools. There was an already known capacity limit, which I’ve written about before. That’s the underlying thesis behind the famous leaders who reduced their wardrobe choices and other minor decisions to preserve our apparently finite daily limit on critical decision-making. So it aligns, really, that AI — which introduces endlessly more angles, asks us more questions, pushes more decisions on us — would run us up against that limit faster.
Because we should assume that, not unlike AI, we have only so many tokens. Our adaptation, I’d guess, has evolved even more as we moved from physical labor outputs to mental ones (a smaller number of people in industrialized nations simply move object from point A to point B, we are overseeing machines that sew, not doing sewing). Decisions became the heavy load for leaders, not actions, and so we adapted even more to protect and preserve those tokens for the most impactful scenarios. But we also had built-in limits on how many choices were even available to us, bounded by our own imagination. Now, we use two or three (or more) AI tools, and the number of choices is multiples more.
We’re all going to have to learn to budget our decision-making tokens.
So the result of all that is that we might use a day’s worth of decisions by 3pm. That creates a new disconnect between our mental fatigue and our physical fatigue, which used to more naturally converge. (By the way, I am focusing on just work-related AI and other tasks. If you already did a load of personal work, especially via AI before you got to the office, assume the depletion began far earlier in the day.)
We never noticed the convergence because we never needed to, and most likely we assumed it was as much causation as correlation. If you’re physically exhausted, it makes a kind of sense to be cognitively tired too. It’s like trying to do math while sleepy. But now we are making so many decisions that we are essentially using up an equivalent of human tokens — because the AI version of what we used to do gives us more data, more choices, more loops. If you’ve ever spent several hours with ChatGPT (come on, I know it’s not just me), you’ve noticed how easy it is to go deeper into any inquiry than you actually needed to. It’s entertaining. It may also be more cognitively expensive than we realized.
That depletion, when it hits at 3pm instead of nine in the evening, can be disconcerting. I’ve written before about how our need to explain it can make us think it’s a complexity problem, when it’s actually a capacity one. We may simply be reaching our daily limit on complex decision-making because we’re running too many second and third iterations — because AI makes them easy to run. Or, at the very least, we’re putting AI against decisions that never needed it.
The more favorable read is that it’s a bit of both, and that we — and AI — will evolve. AI will get better at not leading users down rabbit holes that don’t serve them. And users will get more disciplined about what they ask. (I’m having a sudden flash-forward to the future: ‘prompticians’ who will build perfectly bounded prompts with decision making maximums built right in.) But I think this is solvable, for the same reason we mostly adjusted to a single long seven-or-eight-hour sleep. We adapted once; we’ll learn to better allocate our decision-making again.
We are in the early days of a revolutionary way of working. Humans need time.
We already have seen leaders reduce wardrobe choices to keep noncritical decisions off their plates, as mentioned. Perhaps we are entering an era where everyone must. Companies could do the same with AI.
That could mean use it less in the early stages, where junior team members were always meant to do the first data collection and synthesis — which would also spare the people with the least experience from running endless AI queries they don’t yet have the grounding to evaluate. Or apply the discipline at the top, where leadership teams shouldn’t be reinventing the wheel with AI on plans that already rolled up through layers of people who hypothesized and tested them. That keeps a leadership team from burning its human tokens overloading on AI, when the decisions reaching them have already been vetted and tested.
Perhaps most reassuring – since no one reading this actually lived through the growing pains of an eight-hour sleep – is that there is confidence already, that we will adapt…eventually. One author of the Harvard brain fog study said in an interview with CNN that the people suffering brain fry did not seem prone to burnout. Unlike the latter, which is usually used to describe long-term fatigue, those with brain fry experienced acute exhaustion. As soon as they took a break, they recovered. Maybe it’s time to bring back a second sleep.
