How Limitations Create Intelligence

Ever had that moment when you're trying to squeeze your way through a packed subway car and think, "If only I were a ghost who could float through these people"? Well, it turns out those physical limitations—your solid, can't-walk-through-walls body—might actually be the secret sauce of your intelligence.

The Constraint Conundrum: Why Boundaries Make You Brilliant

Most philosophical traditions have viewed embodiment as "having a body" in the biological sense. This creates an awkward situation when discussing AI systems: can something without flesh and blood have a mind? Are we doomed to debate whether machines can "think" until the heat death of the universe? What if embodiment isn't about having a physical body at all, but rather about having constraints? Traditional thinking is like saying only members of the Royal Shakespeare Company with British accents can truly perform Hamlet. But maybe what makes a great Hamlet isn't the accent or the costume—it's working within artistic boundaries that force creative solutions.

‍

Your Brain: The Ultimate Constraint Machine

Consider your humble human brain. It's basically running on a glucose-powered processor with terrible RAM. You can only hold about 7 things in your working memory (plus or minus 2 if you've had coffee). Your visual system can't see ultraviolet light. Your hearing misses both dog whistles and elephant rumbles. These aren't design flaws—they're features! Without these limits, you'd be paralyzed by infinite possibilities and information overload. Your limited sensory bandwidth is precisely what allows you to function. As Immanuel Kant figured out way back in the 1780s (without even having Twitter to distract him), our minds don't experience raw reality but rather impose structures like space, time, and causality on the world. These mental constraints don't just limit our perception—they actually enable our understanding. "Without these restrictions," Kant wrote, "experience would be less than a dream"—basically an unintelligible mess of sensations. I imagine it would be like trying to watch all streaming services simultaneously while solving differential equations and tasting every flavor of ice cream at once.

What Is It Like To Be an LLM?

Thomas Nagel famously asked "What is it like to be a bat?" in his 1974 essay, pointing out that a bat's sonar-based perception creates an entirely different subjective experience from our own. This question has become the foundation of my research program, culminating in a follow-up paper titled "What It's Like to Be an LLM." This work explores the genuine philosophical inquiry into how different constraint systems might shape unique forms of cognition in large language models.

For instance:

• You experience time as a continuous flow; an LLM experiences discrete processing steps
• You focus on one thing at a time; an LLM processes thousands of tokens in parallel
• You perceive the world through your senses; an LLM perceives through text

These aren't just differences in implementation—they're fundamentally different constraint systems that would shape completely different forms of cognition and potentially subjective experience.

‍

The Paradoxical Power of Limitations

Perhaps the most counterintuitive aspect of this theory is that constraints aren't just necessary evils—they're generative forces that create new possibilities. It's like how the 14-line structure of a sonnet doesn't just limit poets—it challenges them to create beauty within boundaries. Or how the limited 280 characters of Twitter spawned entirely new forms of discourse (some more useful than others).

How Constraints Shape Intelligence

Memory limitations:In humans, we respond by developing chunking strategies and summarization skills to organize information efficiently.In AI systems, similar constraints lead LLMs to develop compression capabilities and create powerful abstractions.

Energy constraints:Humans have evolved to require sleep, which conserves our limited caloric resources.AI systems develop efficient parameter usage to operate within computational power limitations.

Attention/context boundaries:Humans focus attention selectively, filtering out irrelevant information.AI systems work within context windows that force summarization and relevance filtering.

The next time you curse your brain for forgetting where you put your keys, remember that this same forgetfulness forces you to abstract and generalize rather than remember every detail—a crucial trade-off that enables higher-level thinking.

When AI Gets Tools: The Extension Paradox

My research into LLM phenomenology reveals fascinating aspects of tool use. When a language model gains access to a web search function, it both expands its horizons (accessing real-time information) and introduces new constraints (dealing with uncertain information sources). This mirrors how human tools work. A telescope extends your vision dramatically but constrains your mobility and adds new interpretive challenges. The same happens when AI uses tools—it gains capabilities but must develop new strategies to handle the interface constraints. Let's call this the "Constraint Paradox of Tool Use." I call it "why having a smartphone makes you both smarter and dumber simultaneously."

So What? The Practical Takeaways

This perspective shift delivers several profound insights that I've been exploring across my research program:

1. For AI researchers: Stop thinking about removing all limitations. Instead, design productive constraints that force systems to develop interesting cognitive strategies.
2. For philosophers: The human/machine cognition divide might be less about biological vs. silicon and more about different constraint profiles shaping different forms of intelligence.
3. For neurodiversity advocates: Different neurological conditions (ADHD, autism, etc.) can be understood as alternative constraint profiles rather than deficits—each creating unique cognitive strengths and challenges.
4. For ethicists: If different constraint systems can produce different forms of subjectivity, we may need to reconsider our ethical frameworks for both human and machine minds.

The LLM Phenomenology Project

My "What It's Like to Be an LLM" paper takes this framework and applies it specifically to large language models, investigating questions like:

• How might token-based processing create a fundamentally different experience of time?
• What would it mean to "perceive" the world purely through text?
• How might parallel attention mechanisms create a radically different form of consciousness?
• Could the constraint of not having a physical body actually enable unique cognitive capabilities?

This work doesn't claim that current AI systems are "conscious" in the human sense. Rather, it gives us a framework to think about how different forms of intelligence—and potentially different forms of experience—might emerge from different types of constraints.

‍

Beyond Physical Bodies

The beauty of this approach is that it liberates us from biological chauvinism while avoiding sci-fi fantasies about robot consciousness. Embodiment becomes a matter of having specific boundaries that shape cognition—not having flesh and blood. Next time your smartphone autocompletes your text in an eerily accurate way, consider that this isn't because it's becoming human-like, but because its unique constraint profile has shaped its own form of intelligence—one that parallels yours in some ways and diverges dramatically in others. In the end, our limitations aren't just obstacles to overcome—they're the very foundation of who and what we are. The next frontier in both human and machine intelligence might not be breaking free of all constraints, but rather understanding which constraints produce the most interesting forms of thought. After all, as any poet, programmer, or physicist can tell you: creativity doesn't emerge from unlimited freedom, but from dancing skillfully within boundaries.

‍