The New Literacy: Thinking With AI Without Surrendering Our Thinking

From cognitive offloading to productive friction, AI is changing more than how we think. It's changing how we learn to think.

This is Part 1 of The Human Reckoning: A Future of Trust Series

I grew up in a world where if I wanted to know something, I had to go looking for it.

I wrestled with ideas, searched for what others had already considered, and worked through my own conclusions. I learned to leave room for debate. Curiosity energized the process, even under the pressure of deadlines.

There was no Google. No answer waiting in my pocket. Certainly no artificial intelligence capable of explaining almost anything I could think of asking, then answering my next question before I had fully formed it.

By the time the internet changed how we found information, I had already spent my formative years learning how to think without it. My research, writing, and decisions were mine, and so were my mistakes. Technology changed my tools, but much of the underlying machinery was already in place.

A child born today may never remember a world without artificial intelligence. AI will be there as they learn to write, research, solve problems and make decisions. It will be part of the environment in which those capabilities develop.

I think about that distinction more and more, particularly because my own experience with AI has surprised me.

In the beginning, like many people, I was largely discovering what it could do. And it felt magical. The answers came fast, the time savings were profound and the possibilities for discovery seemed endless. Over time, I began looking more closely at those answers. They weren’t always accurate. And even when they were good, I wondered if whether I would have reached a similar conclusion through my own process.

I became increasingly interested in what was happening during the interaction itself. The conclusions AI reached interested me less than what might be happening to my own thinking in the process. I wanted to recognize my voice and perspective in whatever we produced together.

That awareness led me to my current relationship with AI. I challenge it. I disagree with it. I ask what I am missing or where my thinking may be flawed. An answer often sends me in a direction I hadn’t considered and creates more questions.

Now, rather than letting AI do my thinking for me, I find myself reading more, exploring more and learning more. In short, I am more intellectually challenged and more curious than I was before. AI gives my thinking more to engage with, but it took me a while to get here. That raises a question I think is much bigger than whether people know how to use AI.

How do we learn to think with AI without surrendering the thinking to it?

We are talking a great deal about AI literacy. Schools, businesses and governments are trying to figure out what people need to understand about these systems and how to use them effectively and responsibly.

That’s essential. But it is only part of the literacy we need. We also need to understand which parts of being human we want artificial intelligence to support, which we are comfortable allowing it to replace, and which we need to keep exercising ourselves.

That distinction is harder than it sounds. We have been designing friction out of human life for a very long time, and most of us have benefited enormously. Technology has made information easier to find, calculations faster, and navigation almost effortless. I wouldn’t willingly give up those conveniences simply because the old ways required more effort.

Some forms of effort are simply inefficient. Others are building something we can’t see while we’re in the middle of them. Figuring something out for ourselves requires us to stay with uncertainty a little longer, decide what matters, or discover that our first assumption was wrong. Those capacities require practice.

Which leads me to a question we should be asking before we rush to make every human task easier:

Before we design friction out of an experience, should we understand what that friction was developing in us?

Not All Friction Is the Same

There is a body of research behind this distinction. Learning scientists use the term “desirable difficulties” to describe conditions that make learning somewhat harder in the moment but can improve retention and understanding over time. Retrieving something from memory rather than immediately looking it up is one example. Working through a problem before being shown the solution can be another.^[i] The value comes from what the effort asks our brains to do, particularly when it exercises a capability we want to strengthen.

Human beings have always looked for ways to make thinking easier. Writing itself is a form of cognitive offloading. Instead of holding everything in memory, we put it somewhere outside ourselves. Books expanded that capacity. Calculators took over computation. Search engines and GPS moved still more work outside our heads.^[ii]

Each time, we gained something. We also changed what we practiced. I don’t know nearly as many phone numbers as I once did. I sometimes use GPS to get somewhere I have driven before. And I certainly don’t want to return to searching through an encyclopedia when I can find what I need in seconds.

This is adaptation. The more interesting question is what we are adapting toward. AI raises the stakes because it can take on so much more of our cognition. It reaches into the thinking process itself, helping us make sense of information, develop an argument, identify patterns or decide what to do next. That makes the boundary between assistance and substitution harder to see.

If AI writes an email I already know how to write, perhaps very little has been lost. If it summarizes hundreds of pages so I can spend my time thinking about what they mean, it may have increased the value of my attention.

The calculation changes when AI does the work through which someone is still building the underlying capability. Students are the easiest example: a student handed an interpretation of a novel before forming their own is one that comes to mind. In business, a young analyst might produce a sophisticated strategy deck without ever deciding which information mattered enough to put on the page.

The output may be excellent. But what happened inside the person who produced it? That question is going to become increasingly important because the quality of an artifact may tell us less and less about the capability of the person behind it.

For most of my career, there was at least some relationship between the two. A thoughtful analysis generally required someone to do thoughtful analysis, and a well-constructed argument required enough understanding of the subject to construct it. AI changes that relationship.

It can produce evidence of capability without necessarily producing capability itself.

That creates a challenge we have barely begun to address, and it lands hardest on people who are still learning. For organizations, this raises a harder question: if junior work no longer requires junior struggle, how do people develop the judgment that senior roles depend on?

The Difference Between Removing Work and Removing Practice

We need to distinguish between unnecessary friction and productive friction.

Unnecessary friction consumes time or cognitive energy without giving much back. There is little reason to preserve repetitive work simply because humans used to have to do it.

Productive friction asks something more of us. It can be the process of deciding which evidence deserves more weight, staying with uncertainty when two explanations seem plausible, or recognizing that we don’t know enough yet.

And sometimes it is simply making a decision, discovering we were wrong and carrying that experience into the next one. These experiences become practice, and over time, that practice helps develop the judgment we draw upon when we need it.

This is where the conversation about AI literacy needs to expand. Teaching someone how to prompt an AI system is useful. Teaching them to verify an answer is essential. But neither necessarily develops the judgment required to know what should be questioned in the first place.

It (AI) reaches into the thinking process itself, helping us make sense of information, develop an argument, identify patterns or decide what to do next. That makes the boundary between assistance and substitution harder to see.

We need enough knowledge to notice when something doesn’t look right and enough curiosity to wonder what might be missing. We also need the confidence to challenge an authoritative answer and the humility to recognize when our own thinking may be wrong.

These are human capabilities, and they develop through use.

Growing Up with the Answer

Those of us adapting to AI today have an advantage we should recognize. We remember before.

I know what it feels like to stare at a blank page with no one there to suggest the first sentence. I know what it is to research a question without having the answer synthesized for me. I have made decisions with incomplete information and learned through experience which questions I should have asked sooner.

My generation did not necessarily develop better judgment. We simply had a great deal of practice doing certain things without intelligent assistance. Millennials and Gen Z came of age as search engines, smartphones and social media changed how people found information, communicated, and learned. Many Millennials remember childhood before smartphones, while Gen Z grew up with far more immediate access to digital information. Both are now adapting to generative AI as students, young adults, parents, and professionals, but it arrived after much of their early learning had already taken place.

The generations behind them won’t have that dividing line. An intelligent system may be present the first time they struggle with a math problem, write a story, or try to understand something unfamiliar.

The possibilities for learning are extraordinary. Imagine being eight years old and able to follow almost any thread of curiosity as far as you want to take it. A question about why animals migrate could lead to climate, geography or evolution. Each answer can open another door, with the explanation adjusting as the child’s understanding grows.

AI could become an extraordinary engine for curiosity.

It could also make uncertainty disappear very quickly.

A child asks. AI answers. The assignment gets done. What matters is not simply whether the child used AI. It is what the child was doing cognitively while using it. That is where our attention needs to shift.

We Don’t Know What We Don’t Know

Metacognition is our ability to think about our own thinking. It includes recognizing how well we understand something, where our knowledge ends and when our confidence may exceed what we actually know.^[iii] Those capacities become especially important when we interact with AI.

Our blind spots extend beyond the limits of what we know. There is an enormous body of knowledge we may not even be aware exists. We can recognize a gap in our understanding when we know enough to identify it. But sometimes we lack the awareness needed to formulate the question in the first place.

This is one of the possibilities I find most exciting about AI. Much of our curiosity begins with a spark: something we read, a conversation, an unexpected connection. Without that exposure, we might never know there was something worth exploring. AI can supply those sparks, opening doors beyond the boundaries of our own experience. But we have to stay curious enough to walk through them rather than simply accept what is presented.

There is a tension here. To recognize the limits of our knowledge, we need opportunities to encounter those limits. If an answer appears before I have tried to retrieve what I know or formulate my own hypothesis, I may get the information I need without discovering where my understanding ended.

That is why sequence matters. If I form an opinion before I ask AI, and consider how confident I am in it, then I can explore what the system surfaces that I hadn’t considered. I can see where we disagree, and whether the new information changes my view. I can also ask what I hadn’t thought to ask.

I saw the difference this makes with imposter syndrome, a subject I have spent years developing my own approach to. When I asked AI about it without bringing my perspective, I got a well-organized summary of what every other expert has said. Nothing was wrong with it, it simply reflected the consensus my work challenged. When I brought my own methodology and framework into the conversation, the difference was night and day. AI could test my thinking, extend it, and show me where it needed more support. It could only do that because I had already done the work of forming a view.

Forming a view first is a small change, but it keeps me cognitively present while expanding what I can discover. Perhaps that is part of what we should mean when we talk about AI literacy: knowing how to remain an active thinker while receiving increasingly sophisticated assistance.

Designing for the Human

Much of our collective learning about AI has happened through individual experimentation. I’ve certainly learned through trial and error, gradually discovering how to engage with AI in ways that challenge and expand my thinking.

But why should we leave that discovery entirely to chance?

We have invested enormous resources in teaching machines to interact more naturally with humans. It’s time to put comparable attention into helping humans develop the skills and habits that make those interactions intellectually valuable.

The way we learn to use AI and the way AI is designed to interact with us will both shape the capabilities we develop.

That extends well beyond schools. Parents are already deciding when children should have immediate access to an answer. Educators are reconsidering what learning looks like when AI is available. Employers face a different version of the same problem when an entry-level employee can produce work that once required years of experience. And technologists are making design choices that will influence whether AI simply answers us or helps us think.

All of this comes back to a deceptively simple question:

What are we trying to develop in the human being?

Once we know that, we can make much better decisions about what the technology should do. There are plenty of things I am happy to hand over to AI, particularly work that adds little to my thinking. At other times, I want help without giving up the task itself.

And increasingly, I find myself wanting something else from it: resistance. A little pushback. I want it to tell me where my argument is weak or surface something I haven’t considered. I want it to make me rethink the premise when the premise deserves rethinking.

In those moments, efficiency isn’t the goal. Better thinking is.

We have become very good at asking how much human effort technology can remove. We also need to understand which kinds of effort are worth preserving and where AI might actually help us deepen them.

There is enormous territory between preserving old ways of working and handing over every cognitive task we can. That is where we learn to think with AI without surrendering the thinking to it.

What Are We Designing For?

We are still early enough in this transition that many of the norms around how we use AI have yet to be established. I see this as an opportunity.

So far, much of the conversation has focused on what AI can do and how much more productive it can make us. We measure time saved, quality of output and work automated. Those are reasonable measures, but they tell us very little about what happened to the human in the process.

We need another measure alongside them: After the interaction, is the human more capable, less capable, or simply more productive?

Those outcomes are not interchangeable. If AI helps me finish something in half the time, that may be valuable even if I learned nothing from the experience. Not every task needs to develop me.

But sometimes something more happens. AI helps me see something I hadn’t considered, understand something more deeply or recognize a flaw in my reasoning. In those moments I become more curious and ask deeper questions. My output improves, but what excites me more is that I do too. This is the outcome that makes me optimistic about where this can go.

As we consider future generations, we cannot simply hand them extraordinarily capable tools and assume judgment will follow. Judgment requires practice. So do curiosity and the ability to recognize when we don’t know enough. If we value those capacities, we need to create environments in which they can develop.

Technology should remove friction that no longer serves us. But before we design it out of an experience, we should understand what that friction was developing in us.

AI offers an opportunity to replace some of that effort with something even more valuable: productive challenges that strengthen our thinking. By questioning our assumptions, introducing new perspectives and helping us discover what we don’t yet know, AI can do more than make thinking easier. It can help us think better.

This won’t happen automatically. It will require intention from the people building these systems and from the rest of us deciding how they enter our classrooms, workplaces, homes and lives.

That is where the idea of a new literacy begins. It goes beyond learning how to prompt a machine or recognize when it is wrong. It asks us to understand when assistance expands our capabilities and when it begins replacing the experiences through which those capabilities develop.

We are still learning how to think with AI without surrendering the thinking to it. And as we do, we are also designing the environment in which human capability will develop alongside increasingly capable machines.

We should be every bit as intentional about developing the human as are about developing the machine.

---

^[i] Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Psychology and the Real World (pp. 56–64). Worth Publishers. EBjork_RBjork_2011.pdf

^[ii] Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. Cognitive Offloading - UCL Discovery

^[iii] Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911. Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry.