Attention Architecture: Emotionally Intelligent AI

We taught machines to think. We forgot to teach them what to notice.

In 2017, a team of Google researchers published a paper that would reshape the trajectory of artificial intelligence. It was called “Attention Is All You Need,” and it introduced the Transformer architecture, the engine behind ChatGPT, Claude, Gemini, and virtually every large language model in production today.

The core idea was deceptively simple. When processing a sentence, not every word matters equally. The word “it” in “The cat sat on the mat because it was tired” refers to the cat, not the mat. A model that treats every word with equal weight will misunderstand the sentence. A model that knows where to pay attention will understand it.

So they built an attention mechanism. A system that learns, from billions of examples, what to focus on and what to filter out. What to amplify and what to suppress. What matters and what doesn’t.

It changed everything. It was also, without anyone quite realising it, an admission that the most important thing in intelligence isn’t processing power. It’s knowing where to look.

Eight years on, we’ve built systems that can reason, write code, generate images, pass bar exams, and hold conversations that feel uncannily human. We’ve scaled these systems to hundreds of billions of parameters. We’ve fed them the accumulated text of the internet. We’ve made them, by almost any measure, extraordinarily intelligent.

And yet.

There’s something they still can’t do. Something so fundamental to human cognition that we barely even recognise it as a skill. Something that, once you see it, changes the way you think about intelligence itself.

They can’t tell the difference between what matters and what matters to you.

I was once in a section briefing where four crews heard the same set of facts and walked out with four completely different missions in their heads.

The Tornado F3 was a two-seat aircraft. Pilot up front, navigator behind. A section was four aircraft, so eight people in the room for the brief. The mission commander stood at the front and laid out the picture: the threat, the fuel state, the weather, the formation plan. Same words for everyone. Same screen. Same data.

But I watched the room, and I could see it happening. The pilots were locked onto the weather and the formation geometry. The navigators were processing the threat picture and the fuel profiles. And within each pair, the experienced crews were hearing something different from the fresh ones. The experienced navigator next to me wasn’t hearing the threat briefing as new information. He was hearing it as a confirmation or a challenge to the picture he’d already started building before he walked into the room. His attention architecture had already sorted the data into signal and noise before the first slide went up.

Nobody was wrong. Nobody was inattentive. Everyone was paying attention. They were just paying attention to different things. Because attention, the real kind, the kind that determines what you do next in a situation where the stakes are real, isn’t a spotlight. It’s a filter. And the filter is shaped by need.

I was the navigator. My job was to manage the weapons systems, the threat assessment, the tactical picture. My pilot’s job was to fly the aircraft and make the stick-and-throttle decisions. We heard the same brief. We processed it through different architectures. And when things got busy, and they always got busy, those architectures determined what each of us attended to first, what we trusted, and what we ignored until it was too late to ignore.

That was thirty years ago. I’ve spent the decades since trying to understand why that happens. Not as a curiosity. As a structural question. Why do people who are equally intelligent, equally experienced, and equally motivated perceive the same situation differently? And if I can answer that, can I predict how they’ll perceive it? Can I build systems that account for it?

The answer, which took seven psychological theories to properly articulate, is that human attention has an architecture. Not a pattern. Not a preference. An architecture. A structural system that determines what becomes signal and what becomes noise. And that architecture is different for different people because it’s built on different foundations.

Here’s what I mean by that, and I want to be precise because this is where most of the conversation about attention goes wrong.

Attention isn’t a skill you can improve. It isn’t focus, which you can train. It isn’t concentration, which you can practise. Attention, at the architectural level, is a filtering system that operates before you’re aware of it. Before conscious thought. Before choice. Before you’ve decided to pay attention to anything.

You walk into a room and within two seconds, your attention architecture has already done its work. It’s already sorted the incoming data into signal and noise. It’s already decided what matters. And it’s done all of that based on something you didn’t choose and can’t easily override: the fundamental psychological needs that drive your behaviour.

A person whose deepest need is connection walks into that room and their attention architecture immediately scans for social signal. Who’s connected to whom? Who’s been left out? What’s the emotional temperature? They don’t decide to do this. They don’t even notice it happening. It’s just what becomes visible to them, in the same way that a radio tuned to one frequency simply doesn’t receive transmissions on another.

A person whose deepest need is competence walks into the same room and their attention architecture immediately scans for information density. What’s on the whiteboard? What’s the structure? Where are the data points? Same room. Same moment. A completely different perceptual reality. Not because they’re ignoring the social dynamics. Because those dynamics aren’t on their frequency. They’re noise, not signal.

A person whose deepest need is autonomy walks in and scans for energy and momentum. Is this going somewhere or are we circling? What’s the opportunity nobody else has spotted? Where’s the opening?

A person whose deepest need is security walks in and scans for risk. What could go wrong? Are the basics covered? Where are the failure points?

Four people. Four architectures. Four completely different versions of the same room. And here’s the thing that should unsettle you: each person genuinely believes they’re seeing the room as it is. Not as they interpret it. As it is. Because the filter is invisible. You don’t see it working. You just see the output, and you assume the output is reality.

It isn’t. It’s your reality. The person standing next to you has a different one.

This is what I built the STAR Framework to explain. Not as a personality test, not as a labelling system, but as a structural model of how need shapes attention, and how attention shapes decision. The seven psychological theories that underpin it, Self-Determination Theory, the Big Five, Dual Process Theory, Regulatory Focus Theory, Social Identity Theory, Cognitive Bias Theory, and the Appraisal Theory of Emotion, all converge on the same insight from different directions: human beings don’t process reality objectively. We process it through need-shaped filters that determine what becomes signal and what becomes noise. The filters are consistent. They’re measurable. And they’re different.

This isn’t a preference. This isn’t a style. This is architecture.

Now let me show you where this gets interesting, because the connection to AI isn’t a metaphor. It’s structural.

The Transformer architecture’s attention mechanism is the most sophisticated information filter ever built. It can process relationships between words, concepts, and patterns across billions of tokens. It is extraordinarily good at knowing what, within a given context, is probably relevant. It can attend to the relationship between a pronoun and its referent across three paragraphs. It can attend to the tone of a document and adjust its output accordingly. It can attend to patterns that no human would spot without statistical analysis.

It is, by any reasonable definition, a marvel of attention.

But it has a limitation that the industry has been remarkably quiet about, because it undermines the narrative that these systems are approaching human-like understanding.

The attention mechanism is uniform. It doesn’t know who it’s attending to.

When you ask an AI a question, the model’s attention mechanism activates across its entire training distribution. It attends to patterns that were statistically significant across all the humans it learned from. Not to your attention architecture. Not to your need hierarchy. Not to the specific way that you filter the world. It answers the question that the average of all humans would ask, not the question you were actually asking.

And here’s why this matters more than any technical limitation: it means that AI, right now, is giving everyone the same answer. The person who needs connection asks about a new software tool and gets an answer about community and collaboration features. The person who needs competence asks about the same tool and needs to hear about specifications and performance benchmarks. They both get the same answer. And it’s wrong for both of them.

Not factually wrong. Architecturally wrong. It doesn’t attend to what matters to the person asking.

This is why AI can give you a technically correct answer that feels completely wrong. You ask a question and get a response that’s accurate, well-structured, and completely useless. Not because the model is stupid. Because it doesn’t know what you find important. It’s answering the question it thinks you asked, not the question you actually asked. And the gap between those two things is the gap between intelligence and understanding.

But here’s where the lightbulb goes on. Because once you see this, you can’t unsee it. And the implications are enormous.

Imagine an AI that doesn’t just process your question, but processes you.

Not in the surveillance capitalism sense. Not in the “we’ve profiled you based on your browsing history” sense. In the sense that the system has a functional model of your attention architecture. What you attend to first. What you filter out. What makes you lean in. What makes you disengage. What kind of answer will actually land for you versus what kind of answer will bounce off.

A person who needs competence would get answers structured around evidence, logic, and methodology. Not because the AI has been told to do that, but because it understands that this particular human’s attention architecture prioritises structure, and that information delivered without a logical framework literally doesn’t register as useful. The words go in but nothing connects. The filter removes them before they reach conscious thought.

A person who needs autonomy would get answers structured around possibilities, options, and momentum. Not because it’s dumbing down, but because it understands that this particular human’s attention architecture prioritises freedom, and information delivered as a rigid prescription literally feels like a constraint rather than a help. They don’t ignore detailed plans. They suffocate under them.

A person who needs connection would get answers grounded in human impact. Stories. Outcomes. What this meant for people. Because their filter doesn’t process data the way a competence-driven mind does. Data is noise. Stories are signal.

A person who needs security would get answers grounded in precedent, track record, and proven reliability. Not “here’s what’s possible” but “here’s what works.” Because their filter doesn’t process possibility the way an autonomy-driven mind does. Possibility is uncertainty. Track record is information.

Same question. Four different answers. Not because the facts are different, but because the attention architecture of the person asking is different, and an emotionally intelligent system honours that.

This isn’t personalisation as we currently understand it. Personalisation is “you bought a red jumper, here’s another red jumper.” This is something deeper. This is architectural alignment. The system isn’t giving you what you’ve shown it you want. It’s giving you information in the shape that your brain is built to receive it.

The difference between those two things is the difference between a map and a compass. A map tells you where things are. A compass tells you which way you’re facing. Current AI gives you the map. Attention-architecture-aware AI gives you the compass.

And that changes everything. Because when a system understands not just what you asked but why you asked it, what need drove the question, what architecture shaped the asking, then it stops being a search engine with better grammar and starts being something genuinely new. A system that doesn’t just process information but processes meaning.

I see this already in the work I do. I wrote recently about what I call The Invisible First Click, the moment where a student asks an AI assistant to recommend accommodation, and the AI’s answer forms their entire consideration set before they ever visit a website. The operator whose property doesn’t appear in that AI-generated answer has, functionally, ceased to exist for that student.

But here’s the deeper layer, the one that nobody is talking about yet.

The AI that generates that answer doesn’t know who it’s answering for. It doesn’t know whether this student will choose based on community, data, excitement, or security. It gives the same answer to everyone. And that means the answer is, by definition, wrong for everyone. Not factually wrong. Architecturally wrong.

The operators who understand this will be the ones who survive the AI-mediated discovery era. Not because they game the algorithm. But because they build a presence that speaks to every attention architecture. Data for those who need evidence. Possibility for those who need momentum. Community for those who need connection. Security for those who need stability. Not in one piece of content, but across an entire ecosystem of touchpoints that an AI can draw from when it’s constructing an answer for a human it doesn’t understand.

If your digital presence only speaks to one attention architecture, you’re invisible to three quarters of your potential audience. Not because they can’t find you, but because when they do find you, the answer doesn’t land. It bounces off their filter and disappears.

Here’s what I believe, and I say this as someone who built a psychographic framework on seven psychological theories, flew combat aircraft at the edge of what physics and human cognition allow, and has spent the last few years obsessing over how AI and human behaviour intersect.

The next era of AI won’t be defined by who has the biggest model. It will be defined by who builds the best attention architecture. Not the architecture of the model, the architecture of the relationship between the model and the human using it.

We have spent a decade asking: How do we make AI more intelligent?

The question we should have been asking is: How do we make AI more attentive?

Not more attentive to data. Not more attentive to patterns. More attentive to us. To the specific, idiosyncratic, need-driven way that each human being filters reality. To the fact that when you ask a question, you’re not just asking for information. You’re asking for information in a form that your particular architecture can receive.

Because that’s what emotional intelligence is. Not empathy as a feeling. Not warmth as a performance. Empathy as an architecture. A structural capacity to understand that the person in front of you is not seeing what you’re seeing, is not attending to what you’re attending to, and is not motivated by what motivates you, and then building your communication, your decisions, and your systems around that understanding.

The Transformer paper was right. Attention is all you need. They just weren’t thinking big enough about what attention means.

We taught machines to think.

Now we need to teach them what to notice.

But if we do, then what …?

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System: Decode Mindset, Understand Motivation, Transform Human Behaviour, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.