Each time your Maps app calculates the fastest route to your destination, it relies on an algorithm Edsger W. Dijkstra invented decades ago. As a pioneering computer scientist, the Dutch mathematician laid the groundwork for much of modern software design, but there’s another area that he made important contributions to: our thinking about thinking itself.
You might imagine that one of the founding fathers of computing would have relied on machines to solve the hardest problems. In reality, it was quite the opposite. He insisted on doing the vast majority of thinking away from a keyboard. Much of his work started out meticulously handwritten with a fountain pen.
The question of how we think about thinking, and more specifically, which part of our thinking should be left to computers, is particularly relevant today. AI is rapidly approaching a point where we can offload the substance of our work to it. We’re well past a world of merely outsourcing physical labor; today, we are increasingly able to let AI handle the bulk of our creative and analytical work.
That’s a nice thing, because we get what we need with little effort. But when we outsource a task wholesale, without really understanding what went into the output, we’re missing some of the real benefit.
Whenever we work, we produce two kinds of output: the work product and the learning that comes with the process. This then helps us solve similar issues faster the next time around, as well as recognize situations where we can deploy our new skills.
If we outsource problem-solving entirely, we certainly get the work product, but we miss out on building the internal mental models required to apply those lessons later, making it much harder to integrate that new information with what we already know.
The work we don’t notice
One temptation when using a new tool is to assume that everything the tool does was wasted effort in the first place. Why waste mental energy remembering anything when your phone can do it for you?
Cognitive offloading is useful in many situations, since we’re using tools to reduce the amount of information we have to hold in our heads. Writing down a shopping list or putting an appointment in a calendar frees up working memory for other tasks. But sometimes, it’s tricky to know what part of a task is more useful than others.
In 1998, philosophers Andy Clark and David Chalmers published a paper called “The Extended Mind,” arguing that the boundary between thinking and the outside world is not always where we assume it is. Their famous example involved Otto, a man with Alzheimer’s who uses a notebook to store information he cannot reliably remember. So if Otto consults his notebook in the same way another person uses their memory, Clark and Chalmers argue, it makes little sense to treat the notebook as merely an aid. For practical purposes, part of Otto’s mind is the notebook.
The idea is that the human mind naturally forms a “coupled system” with its environment. Our body optimizes for efficiency. If an external tool can reliably hold information or solve a problem, our cognitive systems can stop processing that information. This is a brilliant feature of human intelligence, but it reveals why our choice of tools is so high stakes: Adopting a new technology heavily shapes the kind of cognitive strategies we rely on, and the skills we allow to die.
Paying attention
One reason why people like Dijkstra don’t always use a computer for the deep thinking phase of their work is that writing on a keyboard is very different from writing on paper.
A keyboard introduces an environment that dramatically lowers the cost of error. You can type a sentence, realize it doesn’t quite work, and backspace it into oblivion. This seems purely beneficial because it frees us from the anxiety of getting things perfect on the first try. But it also subtly changes how we approach a problem.
When the cost of making a mistake is zero, we no longer have to meticulously plan our structure in advance. Instead, we shift from intentional design to reactive tweaking. We throw half-formed ideas onto the screen and let those guide us.
This brings up the concept of “desirable difficulties,” a term coined by cognitive psychologist Robert A. Bjork. His theory suggests that introducing certain kinds of friction actually improves learning. For instance, when we are physically forced to slow down while taking notes by hand, we cannot transcribe verbatim. We have to synthesize and be selective, which forces conceptual understanding. When a laptop removes that friction, we often default to mindless transcription.
Arriving at your destination
When we offload the friction of figuring things out, we might get to a finished product much faster. But we arrive there as a tourist. Because we took a shortcut, we lack a deep, structural understanding of what we just created.
If we only occasionally arrive at a destination as a tourist, there is no real harm done. But if we constantly rely on tools to generate the substance of our work while we merely tweak the edges, we will eventually end up with a polished output that contains none of our actual understanding.
Decades ago, while working on the Manhattan Project, physicist Richard Feynman noticed something he called the “computer disease.” He watched as brilliant people became so mesmerized by what the new punch-card tabulators could do that they started playing with the machines instead of doing their actual physics work. They mistook operating the tool for understanding the problem.
Feynman understood the difference better than anyone. At the time of his death, he left a single sentence written on his blackboard: “What I cannot create, I do not understand.”
