Attention
In our previous articles, we explored how machines learn and, in the process, what humans might relearn about our own thinking. We examined how agency can slip, how anxiety can narrow judgment, how patterns form over epochs, and how loops shape how responses to events become patterns. In this installment, we go a step further. Rather than focusing on what breaks down when thinking fails, we turn our attention to what needs to hold together for coherence to be possible in the first place—for that is what drives thought processes.
The word “coherence” comes from a root meaning “to stick together.” We tend to use it to describe clarity or logical consistency, but coherence is far more demanding than either. It is difficult for both humans and machines precisely because it requires three elements to come together simultaneously. “Attention sustains focus, context provides meaning, and purpose directs attention.” Together, they operate as part of a scaffold that builds sequentially, lending structure to the next. But these elements are not additive; they are interdependent. When attention falters, context cannot form, and without context, purpose inevitably collapses.
Purpose, in this sense, is not simply intention or motivation. It is the stabilizing aim that gives direction to attention and meaning to context, answering not only what we are doing, but why it matters over time.
For humans, coherence functions like a scaffolding built from attention, context, and purpose. If any of these supports weaken, the structure wobbles and eventually collapses.
For machines, coherence depends on the instructions they are given through prompting and training. Even when a system is optimized to focus on relevant inputs, coherence can still fail if context is incomplete or purpose is misaligned.
The “Attention Mechanism” in machine learning is a technique that allows models to focus on the most critical parts of input data when making predictions. It assigns different weights to different elements, thereby helping the model prioritize relevant information rather than treating all inputs equally. Yet even when the attention mechanism is optimized, coherence can still fail if context is incomplete or misaligned, leading to purpose degrading into pattern completion rather than meaningful action. In this way, the challenge of coherence is shared by both humans and machines, who can process information efficiently, yet still lose alignment when attention, context, and purpose fail to hold together.
Even with the best intentions, alignment breaks down when attention, context, and purpose fail to hold together. A helpful way to see how these elements interact is to imagine water flowing through a system. Attention is the pressure that moves the water, context is the pipe that directs it, and purpose is the reservoir it is meant to fill. If attention is scattered, the pressure drops and the flow weakens. If the context is cracked or misaligned, the water leaks or runs in the wrong direction. And if the purpose is unclear, there is no container at the end. The water keeps moving, but nothing is ever collected.
Potarazu and Knoop: ‘Know pain, know gain’: On how ambition turns pain into currency — and why we must learn to spend it wisely (September 13, 2025)
Coherence in action
Coherence is not an exceptional state; it is our natural baseline. We are responsible for its collapse. Our physical, emotional, and intellectual faculties are represented by attention, context, and purpose, respectively, and when they work in harmony, attention becomes focused, context becomes relevant, and purpose becomes intentional. In those moments, perspective feels clear and unforced, as if effort and meaning are aligned. Yet for most of us, these states are brief, not because coherence is rare, but because attention is so difficult to sustain. Minor disruptions accumulate, and coherence erodes slowly rather than collapsing all at once.
A clean way to see coherence in action is to watch Carlos Alcaraz in the fifth set of the French Open. In that moment, his attention is absolute, locked onto his opponent Jannick Sinner across the net, the spin of the ball, the precise point where it will land. It isn’t 99% focus; it is total presence. At the same time, he remains aware—without reacting—to the roar of the crowd, the tension in the stadium, even the subtle gestures from his coach. Nothing extraneous pulls him out of the moment. The context is unmistakable: in the deciding set of a Grand Slam, every point determines outcomes. He is not thinking about the next tournament or his legacy; he is entirely focused on what matters now. That clarity of context allows purpose to emerge naturally, shaping each stroke with intention.
Intention vs fragmented attention:
One place we sometimes wish we could perform at such a level is with some of our most crucial relationships. For most of us, coherence fails in more familiar ways. For example, a parent rushes home after work to be there for their kids. The purpose is clear. Yet attention remains fragmented, pulled by emails and texts. Context blurs as work concerns seep into the evening. The parent is physically but not mentally present. Coherence fails not through a single mistake, but through small losses of attention and context that drain the purpose of its intent.
The sensory paradox
When we suppress or outsource our own sight and hearing through constant device use, we weaken the very sensory inputs that sustain attention. As attention degrades, context, things, and purpose begin to drift. Research shows that when we engage with information on small screens rather than in the environment around us, our visual engagement changes in ways that reduce comprehension and alter brain activity. Hyperfocus can increase anxiety—because it might signal to the brain the need to focus intensely on a point. Most of the materials we seek and are served are negative. In one study, reading from a smartphone elicited different patterns of brain activity and fewer natural physiological responses and was associated with reduced comprehension, suggesting that the visual environment itself can interfere with sensory processing and attention integration.
Social media has undoubtedly brought benefits, allowing us to connect across distances and share moments that would otherwise remain unseen. It will enable us, even now, to come together around ideas like those in this column.
At the same time, it creates a different kind of distance. When interaction is mediated through screens, we miss cues that would otherwise anchor attention and understanding, leading us to miss the tears in someone’s eyes or the subtle tremor of fear in a friend’s voice.
Machines face a related limitation. Their “attention” is confined to fixed context windows, and anything outside those boundaries is effectively invisible. Without continuous sensory input, machines cannot perceive the broader situation in real time. This is why researchers like Yan LeCun are pursuing “world models” that integrate context, prediction, and feedback. The idea is to include sensory inputs into machine learning models —systems that can simulate or experience the world more like humans do, integrating broader context with sensory signals and ongoing feedback. Ironically, humans themselves are increasingly sensory-deprived in essential ways. When attention is narrow and context is constrained, human and artificial systems struggle too.
Transference of attention
As we spend more time scrolling through the streams of notifications, reels, and alerts, attention that once belonged to the people physically in front of us is increasingly redirected elsewhere. Often it is handed to machines designed to pay attention on our behalf, just like we do with puppy or nanny cams or macros on dating sites—deciding what to surface, what to prioritize, and what deserves attention.
In doing so, we surrender a portion of our own agency. Attention is lost twice: first, when we fail to look up and honestly engage, and second, when we outsource judgment about what matters. What is most unsettling is not that machines are becoming better at attending, but that we are becoming less practiced at doing so ourselves.
Machines fail in similar ways. Large language models do not understand in the human sense; they rely on holding context, reading signals from structure and constraints, and moving toward an implied goal. A prompt is, in effect, an externalized form of human attention. When it lacks clarity, context, or purpose, machines respond much like distracted humans: they produce output without understanding, which might be fluent but misaligned. Under cognitive overload, people fall back on habits; under vague instructions, machines fall back on patterns. In both cases, coherence narrows as attention narrows.
Attention grabbers
Modern digital designs are designed to capture and hold attention by keeping us in a state of constant switching — scrolling, swiping, and reacting to the next stimulus. Each new piece of content delivers a small dose of novelty, even when the material itself is shallow or repetitive.
Over time, this trains the brain to expect frequent rewards and rapid change, weakening our capacity for sustained focus. Instead of staying with a single idea long enough to understand it, we learn to sample, react, and move on. Neural circuits that once supported deep attention are increasingly diverted toward scanning and rapid decision-making, making stillness feel uncomfortable and focus challenging. The result is not simply distraction, but a gradual rewiring where constant switching becomes the default, and the ability to remain present and intentional slowly erodes.
As Johann Hari argued in “Stolen Focus,” our declining ability to pay attention is less a personal failing than a predictable response to how modern environments are designed. Constant interruptions, rapid context switching, sleep deprivation, chronic stress, and digital systems optimized to capture and monetize our focus rather than protect it. When attention is repeatedly pulled away, the brain loses the conditions required for deep thinking: sustained presence, emotional regulation, and a sense of meaning. What fractures first is not intelligence but coherence.
Potarazu and Knoop: Humans vs. AI: Thinking in epochs (Part 1) (
Attention residue
Constant switching does not just fragment attention in the moment; it leaves behind what psychologists describe as “attention residue.” When we move rapidly from one task, idea, or stimulus to another, a portion of our mind remains stuck on what we just left, reducing the mental resources available for what comes next. Over time, this residue accumulates, making it harder to think deeply, hold context, or stay fully engaged with a single purpose. If Alcaraz kept harping on a point he had lost in the previous game, he would compromise his attention in the present moment because of the residue.
Machines fall into a similar trap through the way we prompt them. When prompts are layered, rushed, or constantly shifting in scope, models are forced to juggle partial goals and incomplete contexts, leaving behind their own form of residue—fragments of prior instructions that interfere with coherence. Just as humans struggle to think clearly when attention is divided, machines produce muddled or overconfident outputs when we ask them to pivot too quickly without resetting context or clarifying intent. In both cases, the failure is not speed itself, but the absence of deliberate pauses that allow attention to disengage and refocus fully.
Attention residue shows us that the cost of distraction is not simply lost focus, but lost footing. When fragments of past tasks, emotions, and stimuli linger in the mind, they interfere with our ability to inhabit the present moment fully. Context becomes harder to see, purpose harder to hold, and coherence increasingly fragile. We may continue to act, respond, and produce output, but the underlying alignment begins to slip. This is where the deeper problem emerges. Attention alone is not enough to sustain coherence; it must be anchored by context and guided by purpose.
To increase coherence in our 2026, we might regularly ask ourselves:
What are we paying attention to right now, and why this, and not something else?
What might we be missing about the larger context of this moment and its influence on us?
What are we actually trying to move toward, and how will we know that we have learned enough to move on?
In Part 2, we examine the roles of context and purpose within the coherence scaffold. Coherence is something we assemble again and again, often under pressure, often imperfectly. It depends on our ability to hold attention long enough for context to form, and on our willingness to let purpose guide action rather than react to threat or noise.
We wish you a happy, coherent new year!
(Sreedhar Potarazu, MD, MBA, is an ophthalmologist, healthcare entrepreneur and author with more than two decades of experience at the intersection of medicine, business, and technology.
Carin Isabel Knoop founded and leads the Case Research & Writing Group at Harvard Business School and is an author whose work focuses on human behavior, leadership, and organizational life. She is the co-author of Compassionate Management of Mental Health in the Modern Workplace (Springer).