A new, powerful definition of intelligence is emerging out of the AI field: “boundary intelligence.”

I first came across it in this excellent piece To Know Is To Stage by Venkatesh Rao, which appropriately, was co-authored by an LLM.

Let me step back and explain what boundary intelligence means (mixing my own words with paraphrasing from that piece).

Our definition of “intelligence” has already evolved through a few eras.

For most of human history, intelligence was mainly defined as “strong memory.” Information was so hard to access in the first place that your ability to recall specific facts and details from memory was considered the highest mark of intelligence.

When the printing press arrived, that definition changed for good. Memorizing long passages became obsolete and unnecessary, since the most flawless memory couldn’t compete with even a small reference library.

Intelligence started to be defined as the ability to process or analyze information that was stored in written form. That included the ability to cross-reference ideas found in different written works, and to synthesize or distill them into a new understanding.

That definition was only strengthened with the rise of digital technology in the late 20th century. Our main metaphor for human intelligence became “processing power,” in analogy to the computer. Intelligence was something that happened “inside the brain,” as a function of a person’s raw brainpower. 

Just like a computer processor, intelligence was defined mainly in terms of power and speed. An intelligent person was someone who could arrive at novel insights quickly. 

But the rise of AI is once again changing our definition of intelligence. That’s because even at this early stage, it already far surpasses our ability to process information, especially large amounts. Many tasks that AI accomplishes in seconds would take us days or weeks to achieve on our own.

Updating the Definition of Intelligence 

Rao proposes a new definition of intelligence in the age of AI: intelligence is defined by what information can be accessed under constraints of cost, availability, and time.

The reality is that storage is now cheap. Computation is even cheaper. What’s expensive is short-term memory access – the ability to keep the relevant details “in mind” for a given problem.

Let’s examine what makes short-term memory access such a difficult problem.

If we use RAM in a computer as a metaphor, the easiest information to access is whatever was accessed most recently. If you have a certain set of data already loaded up “in memory,” it is instantly and cheaply available, versus data that has to be found and loaded up from a hard drive or server.

Thus, a computer’s “intelligence” is now constrained not by the power of its processor, but by its ability to keep the right fragments of the past (and the imagined future) close enough to inform the present. In other words, the bottleneck of a system’s intelligence is how cheaply it can remember.

If you look at how modern computers perform, you can see this principle at work. A CPU can perform billions of operations per second, but is often stuck waiting for the right information to arrive from memory. Storage is cheap and computing is abundant, but what remains tremendously expensive is getting the right data to the right place at the right time.

It’s not the price of knowing that limits intelligence now, but the price of remembering. And the same is increasingly true of humans, as we co-evolve with our technology. 

Activating a memory in the human brain is an expensive operation. It requires waves of coordinated firing across widely distributed neurons, the expenditure of neurotransmitters and metabolic energy, and of course, it takes time. Our “system” pays a real price to retrieve information, and that price determines what we call our intelligence.

The New Frontier of Intelligence

Another way of saying all this is that the new frontier of intelligence is at the boundary of a system – including a computer or a human brain – where it interfaces with external memory. That is where decisions are made about what information to retrieve, when, and how. That boundary is also a filter, determining which information is allowed to enter the system and at what cost.

Rao calls this “boundary intelligence” – the ability to make good decisions at the boundary about what information becomes “knowable” at any given time.

How is the decision of which information to keep accessible made?

It’s made based on predicted needs => what data the system predicts will be useful in the near futureIt’s made based on access frequency => data that was accessed recently is more likely to be needed again soonIt’s made based on cost => if a piece of info is buried too deep, or would require too much computation or energy to retrieve, it’s deprioritized

This explains why intelligent systems – again including human brains, digital computers, and LLMs – often behave in ways that seem deficient or suboptimal. They are not retrieving the ideal memory; they’re retrieving the affordable one. Intelligence in this view isn’t about optimizing across all known information, but optimizing for accessible information under constraints.

We know that the act of recall in the human brain “reactivates” a memory. And the more we recall a specific memory, the more familiar and accessible it becomes in the future. In other words, if we’ve “paid” to keep a memory warm and active by recalling it frequently, it will be even easier to remember the next time. That is how we might remember a fond childhood memory better than yesterday’s boring work meeting.

The implication is that a truly intelligent system is not one that remembers everything, which is impossible anyway. It is the system that knows how to retain access to what matters at its edges, through filtering inputs, deciding what to retrieve, prioritizing relevance, and managing communication with outside systems.

While it’s important to have a certain level of “internal” intelligence, to be able to think and reason and self-regulate, past a certain point, it is boundary intelligence that dominates outcomes. Here are some concrete examples:

Reading well (interior intelligence) matters less than choosing what to read (boundary intelligence)Arguing well (interior) matters less than deciding when and to whom to speak (boundary)Thinking clearly (interior) matters less than focusing attention wisely (boundary)LLMs trained on more data (interior) matter less than having access to rich context (boundary)Being individually productive (interior) matters less than being able to orchestrate a team (boundary)

Boundary Intelligence Is Fundamentally Social

There’s one final detail in this theory: most of the memory an intelligent system utilizes is not its own.

That’s true of computers: they mostly pull data from external hard drives, local networks, or remote servers. Most memory infrastructure is shared.

It’s also true of humans: we rely on external language, culture, societal norms, rituals, and documents, all of which constitute a collective memory infrastructure that we constantly navigate and draw upon. Our own memory is just a small node in a vast external network of books, browsers, friends, and feeds.

This means that boundary intelligence is fundamentally social. It isn’t just about what to retrieve, but from where and from whom. You have to know who to trust, what information or resources they possess, on what terms you can acquire it, and what is expected of you in return.

To act intelligently, you have to know how to navigate through this shared memory. Each intelligent node, human and artificial, is a small island of limited processing ability floating on an ocean of distributed memory. What separates one island from another isn’t what it contains on the inside, but how it filters and navigates what’s on the outside.

Each intelligent system lives not in isolation, but in a perpetual social negotiation with its environment. To be intelligent is not to know everything, but to know how to traverse memory that isn’t yours.

What We Need Now

What boundary intelligence gives you is persistence through time. In other words, it helps you survive – by sensing your environment, adapting to change, and recruiting allies and assets.

Kei Kreutler, in his piece Artificial Memory and Orienting Infinity, reframes cultural memory systems, such as rituals and archives, not as storehouses of facts, but as technologies of orientation.

What we need now is tools to navigate an overwhelming and constantly shifting landscape of relevance. Memory is thus not about having a perfect record of what happened in the past, but about telling you where you are now and where you want to go next. Intelligence is no longer primarily about logic or speed; it’s about the ability to retrieve the past in service of future survival and flourishing.

This is precisely why practices like annual reviews have become so vital in the modern world. In an age where our daily attention is constantly fragmented by digital devices and endless information streams, those who thrive will be those who can regularly zoom out beyond the 24-hour news cycle or social media churn, and contextualize their lives in longer arcs. 

An annual review is a structured way to exercise your boundary intelligence – to consciously decide what memories to keep accessible, what patterns from the past to learn from, and what future possibilities to hold in your awareness.

In modern computing, CPUs don’t process instructions in the order they were received. They process them “out of order,” prioritizing the ones they can handle now and postponing the others for later (a process known as “random access memory”). In other words, they rearrange time.

This is the same thing we do as humans when we conduct an annual review – we revisit and reframe the past, we defer judgment and anticipate regret, and prepare for future conditions that haven’t happened yet. Our lives are not lived linearly. They are assembled out of fragments, swapped in and out of memory, and run only if and when needed. 

The annual review is an orientation technology for managing this temporal complexity, a ritual that lets us consciously navigate between past lessons and future possibilities.

Our memory doesn’t just enable cognition; it enables temporal agency – the ability to reorder time, to choose when to know, when to feel, when to act. And in a world drowning in information, this agency to consciously curate what we remember and what we pursue may be the most important intelligence of all.

I explore these practices in depth in my upcoming book on annual reviews, where I show how this ancient ritual can be adapted for modern life as a powerful tool for developing the boundary intelligence and perspective we desperately need. Sign up here if you’d like to get updates on it.

The post Boundary Intelligence: Why What You Can Access Matters More Than What You Know appeared first on Forte Labs.