On Artificial Intelligence

Questions of Creativity and Cognitive Offload

[Greetings, friends, from Tablerock, Missouri! I’m fresh off a Founders Only retreat weekend in the Denver mountains and can’t stop thinking about how quickly technology is moving the goalposts of the human experience.

Every Founders Only retreat includes a panel discussion where experts from various industries share insights, tips, and best practices relevant to the audience in attendance. This retreat’s panel featured a discussion on Artificial Intelligence and how creators and brands can (and should) be leveraging AI tools to scale their businesses and personal brands. The expert—a 22-year-old from Canada—was brilliant and provided a wealth of valuable insights.

Probably the most impressive part of his presentation, however, was his recognition of the traps and dangers of over-relying on AI. This is something I’ve been thinking about a lot over the last year or so, especially as a writer and lover of ancient philosophy. There are several articles on this Substack over the last few years that wrestle on the periphery of this question: how does artificial intelligence affect the human spirit? Is there a way to use it to enhance? Or is it only an instrument of erosion and decay? (See, e.g., On the Collapse of Stars, On That Hideous Strength, and On the Sorcerer’s Apprentice.)

Then came the MIT study (link to study here) that has been circulating on social media and the internet at large, which confirmed (to a degree), what many had already intuited: over-reliance on AI has negative cognitive consequences. This conclusion has led others on the internet to leap immediately to the position that all use of AI, in all contexts, is destructive and should, therefore, be avoided entirely and eradicated from education. (A friend of mine, Yash Chitneni, published a thoughtful piece defending the use of AI to “reverse learn” just yesterday… you can read that here.)

I find myself somewhere between the two extremes. I have deep and genuine concerns and questions about how the mass use of artificial intelligence will affect us. But I am also (perhaps naively) optimistic that we can find ways to engage with it that don’t totally compromise our spiritual and cognitive health… perhaps even using it in ways to enhance our creative expression and as a tool for flourishing.

As someone deeply passionate about the purity of sacred art and the health of the individual and collective soul, on the one hand, and the democratization and empowerment of artistic expression, on the other, I am slowly developing rules and frameworks for my own personal use of AI that I thought might be worth writing through for my own benefit (and the benefit of others thinking through the same things).

So it is to Artificial Intelligence, creativity, and questions around the consequences of cognitive offload that we devote the rest of today’s essay.]

Figure: Ulysses and the Sirens by John William Waterhouse (1891)

Introduction: The Cognitive Cost of AI Over-Reliance

A recent study by MIT “focuse[d] on finding out the cognitive cost of using an LLM [Large Language Model] in the educational context of writing an essay”¹ has the world chittering about where society’s widespread use of ChatGPT might lead us.

The MIT Study selected 54 subjects and split them into three groups—Brain-Only, Search Engine (i.e., Google access), and LLM (AI access)—and then asked them to write several SAT essays over the course of four months. While the subjects were working, researchers used an EEG to monitor brain activity and found that, of the three groups, those using ChatGPT had the lowest brain engagement and underperformed at neural, linguistic, and behavioral levels. By the end of the Study, those using ChatGPT declined by degree, with many resorting to copy-and-paste methods by the end of the Study.²

Here’s a relevant excerpt from the Study’s abstract:

This study focuses on finding out the cognitive cost of using an LLM in the educational context of writing an essay. We assigned participants to three groups: LLM group, Search Engine group, Brain-only group, where each participant used a designated tool (or no tool in the latter) to write an essay. We conducted 3 sessions with the same group assignment for each participant. In the 4th session we asked LLM group participants to use no tools (we refer to them as LLM-to-Brain), and the Brain-only group participants were asked to use LLM (Brain-to-LLM). We recruited a total of 54 participants for Sessions 1, 2, 3, and 18 participants among them completed session 4. We used electroencephalography (EEG) to record participants’ brain activity in order to assess their cognitive engagement and cognitive load, and to gain a deeper understanding of neural activations during the essay writing task. We performed NLP analysis, and we interviewed each participant after each session. We performed scoring with the help from the human teachers and an AI judge (a specially built AI agent).³

For those of us who have been monitoring or thinking about this problem for the last few years, the Study’s conclusions won’t come as a shock: participants who used AI to write their essays demonstrated lower brain connectivity and neural activity, reported lower ownership over the material (with a vast majority not remembering a single line from their essay), and crafted essays strikingly similar to others who also used AI.

Back to the abstract:

We discovered a consistent homogeneity across the Named Entities Recognition (NERs), n-grams, ontology of topics within each group. EEG analysis presented robust evidence that LLM, Search Engine and Brain-only groups had significantly different neural connectivity patterns, reflecting divergent cognitive strategies. Brain connectivity systematically scaled down with the amount of external support: the Brain‑only group exhibited the strongest, widest‑ranging networks, Search Engine group showed intermediate engagement, and LLM assistance elicited the weakest overall coupling. In session 4, LLM-to-Brain participants showed weaker neural connectivity and under-engagement of alpha and beta networks; and the Brain-to-LLM participants demonstrated higher memory recall, and re‑engagement of widespread occipito-parietal and prefrontal nodes, likely supporting the visual processing, similar to the one frequently perceived in the Search Engine group. The reported ownership of LLM group’s essays in the interviews was low. The Search Engine group had strong ownership, but lesser than the Brain-only group. The LLM group also fell behind in their ability to quote from the essays they wrote just minutes prior… The use of LLM had a measurable impact on participants, and while the benefits were initially apparent, as we demonstrated over the course of 4 months, the LLM group’s participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, scoring.⁴

To summarize, the Study observed a few things that gave cause for concern among those who used ChatGPT: (1) their essays lacked original thought (or a “soul”) and relied on the same expressions and ideas (which were fed to them by AI);⁵ (2) EEGs revealed lower levels of neural activity and attentional engagement during; and (3) post-session interviews showed a significant deficit in memory recall, suggesting a bypass of deep memory processes.

There is, however, hope. For all the concerns the Study raises, one of the promising observations that often gets lost (italicized above) is the finding from Session 4. After writing the three essays, participants were asked to rewrite one of their previous efforts. Those who had previously used ChatGPT now had to do so without the tool (LLM-to-Brain participants), and they struggled significantly.

But when those who had been Brain previously were allowed to use ChatGPT to rewrite (Brain-to-LLM), they showed an increase in neural activity across all EEG frequency bands.

This gives rise to the hope that AI, if used properly, could enhance learning as opposed to diminishing it. “Brain-to-LLM participants showed higher neural connectivity than LLM Group’s sessions 1, 2, 3 (network‑wide spike in alpha-, beta‑, theta‑, and delta-band directed connectivity)... [suggesting] that rewriting an essay using AI tools (after prior AI-free writing) engaged more extensive brain network interactions.”⁶

Insight: Old Patterns, New Problems

The questions that the MIT Study raises around the use of artificial intelligence are not entirely new; they are, in many ways, different and more magnified variations of problems that have existed for thousands of years.⁷

One is that human beings will always be seduced by shortcuts. It’s why we have escalators to avoid stairs and pills that promise the same results as exercise without the work. Over a thousand years ago, Cicero wrote in an essay On Friendship, how rare it was to find someone who preferred to be rather than to seem. Esse quam videri.

Before him, in the days of Plato and Aristotle, it was the sophists who embodied this philosophy. The sophists were masters of making weaker arguments appear stronger through the use of rhetorical flair and showmanship. They prioritized the appearance of wisdom over the embodiment of it because it lined their pockets (as opposed to Socrates, who never took money for his lectures).

Today’s AI tools have, in many ways, platformed and empowered the sophistry that has always existed. Now more than ever, people can produce impressive-sounding content without genuine understanding. It’s easy to appear knowledgeable without putting in the work of actually becoming so. Anyone can now generate a book on Stoicism, complete with quotes from Marcus Aurelius, Epictetus, and Seneca, and examples of how historical figures have embodied the philosophy.

To be clear: I’m not saying that anybody who uses AI is a dirty sophist. Not at all. I’m simply pointing out that AI has now enabled anybody to produce, with relative accuracy, a handbook on something like Stoicism and then sell it into the world. But there’s a big gap between knowledge and understanding. It used to be that, to write coherently about something, you had to understand it first, at least to a degree. For better, or worse, AI has eliminated that gap.

Of course, this is not all bad. On the one hand, it can serve as a powerful educational tool, helping people access ideas and frameworks they might otherwise have never encountered. It allows for the curious and genuine seekers of knowledge to engage in what’s called reverse learning—that is, starting with a “finished product” and then reverse engineering the component parts to slowly train yourself on how the sausage is made.

But on the other hand, it will spark (and already has sparked) a “renaissance of sophists” looking to monetize and profit from philosophies they have never practiced. The problem arises when we realize it’s a not-so-subtle nudge away from the true treasure of philosophy, which is to have its insights ready at hand and immediately accessible as a tool for thriving through life’s challenges.

Applying the insight from the MIT Study that those who used ChatGPT failed to remember a single line from their essays in after-action interviews, the one who writes a book on Stoicism, or the Greeks, or the Romans, or whatever, relying on AI will not internalize the philosophy and, therefore, miss out on most of the value as a result. The desire to follow this path is based, in part, on the soft lie and temptation that it is more profitable to seem Stoic than to be Stoic.

It also forgets the ageless idea that the work is the reward, and the treasure is who you become in the process. When we bypass the stages of growth to get to the fruit, we toss out any fulfillment from achievement. When we go directly from seed to something sweet, we lose the plot and miss out on the best parts of what it means to be human. Our soul has nothing to gnaw on and digest—nothing to sustain or make it substantial. We used to know that we reap what we sow and cultivate; now we think we are entitled to reap as soon as we plant.

When I think back on my own life, and when I felt most like I was growing, I go back to my days of struggle and strain in law school. Every day, we were forced to wrestle with and recognize the patterns of concepts that seemed impossibly complex… we had to read opinions that sometimes caused hours of confusion. But then something would click. Or you would go to bed and wake with the information somehow (almost magically) more organized in your head. That period, more than any other in my life, felt as if I could physically feel new neural pathways forming—like dormant parts of my brain were activating… like I was opening what Kafka called unfamiliar rooms in one’s own castle.

This highlights a real danger of over-relying on AI. If we’re not careful, then to optimize for efficiency, we might well lose the nourishment of genuine understanding and stunt our growth (which, in turn, eliminates the true joy we feel when growing). When the brain engages with new material in a way that rewires neurons, it is like furnishing your mind and imagination with the best ideas to take with you wherever you go. But this only works if the information is moved through your body and engaged with in a way that activates deep memory processing.

I can almost imagine an addendum or sequel to C.S. Lewis’s Screwtape Letters, where senior demon Screwtape addresses how his mentee, Wormwood, might use AI to corrupt the soul of his “patient”:

The trick, Wormwood, is to keep him focused on how artificial intelligence liberates him from his limitations, shifting his attention away from the chains slowly fashioned around his wrists. The true genius of our Father Below is how successfully he finds new ways to trick the humans into making a fool’s bargain, willfully swapping their self-sufficiency for efficiency and the belief that if they simply do and produce more they will be “happy.” Silly creatures. As long as they mistake access to information for wisdom and efficiency for growth, we can comfortably assume they are advancing in our downward direction.

In The Great Divorce, Lewis depicts hell as a place where people become increasingly insubstantial, to the point where they can’t even bend grass under their feet. The residents have become so obsessed with reputation and appearances and selfish pursuits for so long that they have hollowed themselves out.

Our current AI moment offers a similar choice: we can use these tools to become more real (more genuinely knowledgeable, creative, and wise) or more ghostly (more impressive-seeming but less substantial). AI is, in many ways, just the latest temptation toward the great sin of the Tower of Babel—the desire to be like God and transcend human limitations through our own power. AI’s seductive convenience is like the Sirens of Homer (who use sweet songs and promises to lure sailors to their death) and the Witch of the Underworld in Lewis’s The Silver Chair (who uses sweet-smelling powder and hypnotic music to make the protagonists forget reality and accept her false version of the world).

The danger of AI isn’t that it is inherently evil, but that it makes us drowsy to the reality that genuine growth and fulfillment require effort, struggle, engagement, and endurance. Many will use AI as a substitute or shortcut to personal development, instead of a supplement. Therein lies the problem. And what, you might ask, is the solution?

Conclusion: An Emergent Framework of Intentional Engagement

AI presents us with the same fundamental choice that faces every generation navigating the introduction of new technologies: will we use the tools to become more or less human?

The path forward, it seems to me, lies not in rejecting AI entirely, but in understanding how to engage with it in a way that does not totally subvert the spiritual and psychological value of struggle.

Perhaps an analogy, here, might help. Just as escalators and elevators relieve your legs from climbing stairs and reduce physical load, AI relieves your mind and reduces cognitive load. But escalators aren’t necessarily bad, provided you’ve found other ways to tax and train your legs throughout the day. If you’re taking it on a rest day after a few intense days of hiking or after you’ve already exhausted your legs in the gym, then taking it isn’t going to atrophy your physical health; in fact, it might take you where you wouldn’t otherwise have gone.

The same principles apply to AI. There is nothing inherently wrong with allowing AI to help with some of the cognitive load, provided that you’re still finding ways to exhaust your capacities and get to the cognitive place where new neurons are forced to fire. If AI is the cognitive equivalent of taking an escalator (or elevator), we must find ways to increase (rather than decrease) the cognitive difficulty in the areas we care most about growth. Maybe that’s memorizing poetry, a speech, or a presentation. Maybe that’s doing math without a calculator or writing on pure recall.

It doesn’t much matter what it is, specifically, as long as it involves the type of cognitive struggle that causes more frustration than you’re comfortable with. As we move further into the age of artificial intelligence, we must find a balance between where we might allow AI to alleviate strain and stress and where we must pick harder paths of wrestling with ideas until they become a part of us. Some areas of the human experience cannot be “optimized” without losing something essential about what makes us human.

This is something we will have to figure out together. And whether AI ultimately makes us better or worse depends, as with all technologies, entirely on how we choose to use it. As I try and navigate this territory for myself, there is an emergent set of rules I am attempting to follow. And those are these:

1. Avoid copying and pasting AI-generated content directly. Before anything written by AI arrives on the page, it must move through my fingertips—which means the information must move through my DNA, traveling through my eyes and brain before it comes out of my hands. Not only does this force engagement with each word, it ensures voice is preserved and agency over the final product is maintained. In certain instances, like when I was learning to write dialogue for my historical fantasy book, this practice can also serve a similar function to something great writers throughout history have done: transcribing the works of others to learn and internalize patterns. Ben Franklin copied essays from The Spectator. Hunter S. Thompson retyped The Great Gatsby to “feel the rhythm.” Stephen King copied comics to learn storytelling. If used for “reverse learning,” the trend must always be toward self-creation and non-reliance. (The training wheels must always be falling off.)

2. Try to formulate an opinion or wrestle with a concept before asking AI to help. Use AI as a supplement rather than a substitute... after you’ve done your own thinking. Treat it as a research assistant rather than a thinking replacement. The MIT Study supports this approach with its finding that participants who first used only their brains to write essays and later used AI tools showed increased brain activity comparable to those who used search engines.

3. Always ask for sources and verify independently. Whenever using AI to assist with research, make sure it’s giving specific sources and references, and then verify the information directly from the source. In many ways, this is no different than navigating any number of misattributed quotes on the internet. Everything produced by AI should be met with healthy skepticism that assumes hallucination and is seen as potentially valuable but requiring validation. This maintains intellectual ownership and responsibility and prevents a slow slide into mere passivity.

4. Try to limit the use of AI to (a) routine and trivial tasks, (b) check your blindspots, or (c) enhance something you’ve already formulated and are passionate about. At its best, AI might serve human flourishing if it frees us for higher things and pushes us in our pursuits of truth, beauty, and excellence.⁸ But this means that every offload must have a corresponding onload so that our spirits can experience genuine growth in an area we’re passionate about, for this is the stuff that gives our lives meaning and makes life worth living.

As it looks to me, the fundamental question around AI (that each of us must monitor) is whether it makes us more fully human or less so. Does it help us become more loving, more wise, better servants, and more fulfilled? Or does it lead us into pride, laziness, and a stance in the world

These are the questions we must continue to ask and answer for ourselves as we move further into the AI frontier. We must do our best to discern and pick wisely, not just with our heads, but with our souls also in mind. Because the decisions we make today will echo long after we are gone, affecting us, our children, and our children’s children.

—

P.S. I would love to know your thoughts on this so we can work on an emergent framework together. Think I’m too optimistic? Great—shoot me an email and let me know why. Think I’m too pessimistic? Amazing—shoot me an email and let me know why. We’re all facing the rise of AI for the first time together, and I’m just interested in playing my part. Like that ancient Noah, I just want to help us through the other side.

1. Nataliya Kosmyna (and the MIT Media Lab), Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task (the “MIT Study”).

2. A couple of initial points about the study to note: First, the paper has not yet been peer-reviewed, and its sample size is relatively small. Second, it’s not clear to me what incentives (if any) there were for subjects and participants to perform well on the essay. If there were no adequate incentives, then ChatGPT users would have no reason to do anything other than copy and paste. But regardless of any flaws, the paper elevated concerns and has sparked a discussion that is long overdue.

3. MIT Study.

4. MIT Study.

5. This suggests, to me, that one of the most valuable things in the coming age will be perservering authentic thought and voice—it will be the quickest way to stand out in a world where everyone’s saying the same thing. The fastest way to turn a great writer into a good one is to hand the work over to AI.

6. MIT Study.

7. One of the concerns I don’t address in this essay, but which I nevertheless have, surrounds something called AI “sycophancy,” or the tendency of the systems to endlessly affirm, validate, and support user statements rather than provide critical pushback. We already live in a digital echo chamber where individuals have an infinite amount of information to validate or affirm their beliefs, and AI only pushes us further in this direction, eliminating the kind of intellectual friction that promotes growth. Perhaps a future essay on John Stuart Mill’s “On Liberty” applies in a world of AI is in order.

8. Maybe this is too optimistic, which is why I say might here. Because I’m not sure how this will all play out. But I do think there’s a chance it can be nudged toward a use for good.