The Human Cost of Computing

Our capacity to adapt is truly something to admire. Think about the journey: from hunter-gatherers to modern techno-sapiens, we've navigated profound, irreversible shifts in how we act and think. In this long timeline, computers undeniably contributed the most to our recent transformations. And when I say computers, I mean the entire, sprawling ecosystem, encompassing not just desktops and laptops, but the smartphones, tablets, watches, smart speakers, even thermostats that permeate our daily existence. What binds them together is their collective power to reshape our cognition, our actions, and our habits, often in ways that aren't purely beneficial.

Today, countless talented designers and engineers work tirelessly, aiming to make our digital interactions feel seamless, almost natural. But they face an uphill battle, constrained by the weight of legacy technologies. Trying to introduce something fundamentally different encounters immense friction. A world without the familiar structure of files and folders feels deeply unsettling to most. Suggest altering the standard keyboard, a design, rooted in mechanical limitations of early typewriters, and you'll likely spark public outrage. These reactions starkly reveal how deeply we've internalized conventions that, while perhaps necessary once, were never truly natural to begin with.

The result is that our current digital landscape often feels like a patchwork quilt. It stitches together foundational concepts born in the 1960s, layering them with the polished visual gloss of the 2020s. We celebrate faster chips, yet this raw speed hasn't automatically translated into genuinely smarter, more intuitive experiences. We've merely engineered abstractions upon abstractions, building ever higher without pausing to reconsider the shaky foundations below.

Consider the pillars of modern computing. Files and folders are digital ghosts, artifacts echoing the physical limitations of early hierarchical file systems developed, again in the 1960s. Applications are relics of outdated software distribution models, not necessarily the best way to organize capability. The mouse and keyboard remain interfaces optimized for the static world of office furniture, not the fluid, dynamic range of human expression and interaction.

These aren't just neutral tools. They are inherited constraints, shaping how we think and interact in ways dictated by a past we've failed to fully transcend.

The Shape of Thought

Most people underestimate how radically better human-centered computing could be, just as they underestimate how deeply entrenched our current condition truly is.

Let's pretend, just for a moment, that a computer is a person.

Our interactions with people range wildly, from rich, long-form, multi-modal exchanges to brief, constrained text messages. Communication bandwidth between people who know each other well is often far greater than between complete strangers. When you share history with someone, the immediate bandwidth expands. Incoming information grounds itself within a vast environment of shared context, common sense, and unspoken cues. That deep, context-rich connection is the kind of relationship we desperately need to achieve between a human and a computer.

As humans, we think in stories, emotions, connections. Our minds don't neatly file memories alphabetically or chronologically. We remember through webs of relationships, by places and people, by purpose and feeling. Crucially, each of us remembers differently. There is tremendous breadth in how people think, how they learn, and how they communicate. Yet our digital systems relentlessly force us into rigid boxes, like folders, apps, and icons. They demand significant cognitive overhead for even the simplest tasks. This cognitive overhead slowly calcifies into habits, and eventually, these habits overtake our natural ways of being. Personal computers are hardly personal.

To achieve truly human-centered computing, we must first deeply understand what makes human interactions, both with each other and with the world, possible and meaningful. We must step back. Not just a few paces, but far enough to revisit the dawn of computing, armed with the knowledge we have now. Then, guided by these fundamental principles of human interaction, we can begin to build the next generation of form factors, computing hardware, and operating systems from the ground up.

Seeds of Cognition

AI assistants bolted onto a conventional OS certainly have their place. However, they're once again constrained by the underlying technology which defines information access and representation primarily by dividing it into documents and applications.

What we truly need is an operating system that is built differently from the start. By exploring principles of human cognition, we can build a foundation that will feel much more natural for human-computer interaction. A cognitive computer.

At the core of this new architecture is the concept of a kernel. While traditional computers have many modules that are considered foundational, at the lowest level, there is typically a Von Neumann machine reading ones and zeros and executing a fixed instruction set. In a proposed cognitive computer, the kernel serves as the most fundamental unit of interest. These kernels are autonomous, and in aggregation are multimodal. While each kernel specializes in a distinct modality, they can roughly be broken down into the following groups:

1. Sensing kernels perceive the world through text, vision, audio, touch, and even custom interfaces to legacy systems.

2. Learning kernels build models from sensory input, constructing the system's understanding of the world.

3. Motor kernels coordinate movement and attention, guiding sensor kernels to focus on important input data or steering the thinking process in productive directions.

4. Expression kernels generate outputs like text, images, and sound. Think of them as reverse sensing kernels. They are also capable of sending information back into legacy systems.

Together, these kernels create a cohesive understanding of a dynamic world and the system's place within it, not unlike how your own brain combines multi-modal sensory inputs, motor outputs, and introspection.

A cognitive computer has no files and no applications in the traditional sense. There is only the fluid experience of interacting with the world through thousands of kernels that continuously learn and evolve alongside the user. Information exists within the kernels like memories, interconnected and contextual, rather than as isolated items in folders.

To make the transition between radically different paradigms like traditional computer and cognitive computer easier, we must develop a shared language, often using analogies, that helps us grasp the new system in more detail. In the next few sections I will try to explain hypothetical scenarios of how a user might use a cognitive computer and then dedicate a section building parallels between traditional and cognitive computing.

Echoes in the Machine

Imagine a computing experience fundamentally different from managing files or launching siloed applications. Instead, you engage with a system that understands how different parts of your life and experiences connect based on the information you provide.

Consider listening to a piece of instrumental music that evokes a strong, yet indistinct, sense of nostalgia or reminds you of a specific creative period you can't quite place. On a traditional computer, you might struggle to pinpoint the connection, perhaps searching for the track name, then manually searching through notes, documents, or photo folders from roughly when you think you last heard it, hoping to find related items.

With a cognitive computer, the interaction could be profoundly different because there are no separate files to search through. All information, music, images, text, interactions, is part of the system's integrated memory within its kernels. You might ask, "What was I working on when this music was playing last autumn?" or "Show me context around this track from October." The system doesn't search metadata; its kernels access the interconnected memory stream. It correlates the music playing at specific times with other activities or data points experienced or created concurrently. It links the music across your listening history to photos taken during those times, notes written, or even code committed, identifying patterns of co-occurrence within your personal context.

Instead of just listing potentially related files, it might synthesize an insight based on these temporal and contextual connections: "This melody was frequently playing in the background while you were developing the 'autumn landscapes' photo series last October, particularly during your trip to the mountains. You took several pictures of the changing leaves near the waterfall around that time and also wrote a note describing your feeling as 'peaceful and reflective'." It might even generate a small visual collage blending snippets of those photos with colors evoked by the music, helping you reconnect with that specific moment and the creative work associated with it, all inferred from the unified memory within the kernels.

This isn't searching through files; it's the system accessing its integrated memory, leveraging an evolving understanding of your personal history and context (as represented by the interconnected data within the kernels) to facilitate rediscovery, reflection, and creativity in a way current systems cannot. The interaction shifts from giving commands to a passive tool, towards collaborating with a partner that possesses memory, pattern recognition, and the capacity to understand context. This signifies a fundamentally new relationship with technology.

Ghosts in the Architecture

Describing cognitive computing purely on its own terms can be challenging because it departs so significantly from familiar concepts. Interactions change, and the vocabulary is new. To bridge this gap, let's explore how some traditional computing ideas might map onto the cognitive computer's kernel-based architecture, using analogies to aid understanding.

What is a "File"? In traditional systems, a file is a discrete container for data. In a cognitive computer, there are no files in this sense. Information perceived by sensing kernels (like text from an email, pixels from an image, or audio from a recording) is integrated into models maintained by learning kernels. What might appear similar to a file (e.g., a document you are reading) is better understood as a persistent view or representation generated by expression kernels. This view draws dynamically from the underlying, interconnected memory based on your current context and intent. It's not a static object you open, but a fluid manifestation of information relevant to the moment.

What is an "Application"? Traditionally, an application is a distinct program designed for specific tasks (word processing, photo editing, email). In a cognitive computer, these rigid boundaries dissolve. Functionality emerges from the coordinated activity of relevant kernels. For example, "editing a photograph" isn't launching a photo app. Instead, your intent (expressed perhaps verbally or through gesture, perceived by sensing kernels) activates relevant kernels. Learning kernels access the image model, motor kernels might guide focus ("zoom in here"), other learning kernels interpret modification requests ("make it brighter"), and expression kernels generate the updated visual output. The "application" becomes the dynamic constellation of kernels collaborating on your task, seamlessly integrating different modalities (vision, language, action) as needed.

Crucially, this functionality isn't typically "written" or "installed" in the traditional sense. While kernels might possess foundational capabilities, complex behaviors like photo editing or document composition emerge and refine over time. Learning kernels observe user interactions, identifying patterns between intent, kernel activation, and successful outcomes. They learn to orchestrate the sensing, motor, and expression kernels more effectively for recurring tasks. New capabilities can arise from novel combinations of existing kernel functions, guided by motor kernels exploring solutions or even by direct user guidance shaping the system's responses, akin to teaching rather than programming. This contrasts sharply with traditional software development, offering a more fluid, adaptable, and integrated way to access and evolve system capabilities.

Reading an Email: Instead of opening an email client (an application) to view email messages (files), sensing kernels perceive incoming communication data. learning kernels integrate it into your broader communication context, understanding sender, topic, and relevance. Expression kernels present this information, perhaps as a summary or full view, depending on what was determined is most relevant to your current focus. The artificial separation between "email app" and other information vanishes.

These parallels are imperfect analogies, of course. The essence of cognitive computing lies in moving beyond these discrete, bounded concepts towards a more fluid, integrated, and context-aware interaction model, much closer to how we naturally process the world.

Weaving the Cognitive Fabric

Is this feasible now? The necessary components are rapidly aligning. We have powerful hardware like GPUs, NPUs, and emerging neuromorphic chips. We have advancements in sparse representations, sensor fusion, and multimodal AI models. Perhaps most importantly, the mindset is shifting. Users are increasingly ready for digital assistants that can genuinely think and collaborate, not just follow instructions. Critiques of current AI, highlighting the limitations of purely data-driven large models in achieving true understanding or common sense reasoning, underscore the need for architectures like this, grounded in interaction, embodiment, and principles closer to biological cognition rather than just statistical pattern matching on vast datasets. And crucially, we have the intricate blueprint of the human brain, the only example of generalized intelligence we know, serving as our north star.

Ultimately, the journey toward truly cognitive computing isn't just about building a better operating system. It's about exploring the potential for a new kind of digital entity – one that lives with us, grows with us, and learns from us in a more natural, symbiotic relationship. It's an invitation to build technology that adapts to us, finally alleviating the human cost of computing and unlocking new frontiers of thought and creativity.