A provocation about what cleverness really means in an age of AI-mediated manipulation. From Earl Aubec carving reality from Chaos to the Asymmetric Integration Model, the article argues that exploiting human behaviour is not creation. As social media, AI and post-truth systems converge, WARP Labs aims to restore agency to the human side of the interface by making influence legible, controllable and defensible. And it wouldn’t be a Horkan blog article without a callout to Michael Moorcock and a tiny bit of an expletive.

Contents

1. Introduction: You Do, Don’t You?

I mean genuinely: do you think you’re clever?

Because I’ve met plenty of people who clearly think they are, and quite often what they mean by cleverness turns out to be something rather less impressive. They know how to manipulate somebody. They know how to withhold information, manufacture an advantage, exploit an insecurity, play two people against one another, say different things to different audiences, or manoeuvre themselves into a position where somebody else loses, and they win.

There is a particular kind of person who takes enormous satisfaction from this. They got one over on somebody. They knew something the other person didn’t. They understood which button to press. They engineered the conversation, controlled the information, manipulated the relationship and got the outcome they wanted.

And afterwards they stand there admiring themselves because they think they’ve done something clever.

But have they?

Or did they just find a lever?

2. Well Done, You Found a Lever

Manipulation is not invention.

It can require intelligence, certainly. So can burglary, fraud, propaganda, interrogation, gambling and any number of other activities in which understanding another person creates an advantage. Human beings are complicated systems, and predicting what they will do is not trivial.

But there is a difference between using intelligence and creating something genuinely new.

None of this means that technologies of manipulation cannot themselves contain extraordinary invention; it means that manipulation is not what makes the invention extraordinary.

If you discover that somebody is insecure about their status and exploit that insecurity to influence them, you have found a lever. If you discover that outrage keeps people scrolling for longer than contentment does, you have found a lever. If you learn that intermittent reinforcement makes an application difficult to put down, you have found a lever. If you can infer from thousands of behavioural signals that somebody is lonely, frightened, sexually frustrated, politically agitated or desperate to belong, and you use that knowledge to determine what they see next, you have found a very sophisticated lever.

Scale the process to a billion people, surround it with machine learning, call it engagement optimisation and present it on a conference stage with enough graphs, and it can look enormously clever.

But it is still a lever.

That distinction matters because we are entering a period in which the ability to discover and pull those levers is becoming industrialised. Artificial intelligence does not merely allow us to generate text, images, software, music and video. It gives computational systems an increasingly sophisticated ability to model people, generate persuasive material, adapt interactions, simulate relationships and optimise their behaviour against objectives that the human participant may neither know nor understand.

The old manipulator could work a room.

The new one can work a civilisation.

3. Aubec at the Edge of Chaos

There is another conception of cleverness that interests me much more.

In Michael Moorcock’s Eternal Champion multiverse, Earl Aubec of Malador reaches the edge of the existing world, where ordered reality gives way to unformed Chaos. Aubec pushes into that Chaos and, in doing so, extends the territory of Law. What had been an unrealised possibility becomes a landscape. The world becomes larger.

It is a wonderful metaphor for invention.

Aubec does not stand in an existing kingdom congratulating himself because he persuaded two courtiers to stab each other in the back. He goes to the boundary of what exists and makes something exist where previously there was only potential.

That, to me, is clever.

Take something that does not exist and make it exist. Discover something nobody knew. Build something nobody has built. Connect ideas that previously appeared unrelated. Look into confusion and find structure. Take the incoherent, impossible, speculative or imaginary and drag some fragment of it across the frontier into reality.

Carve land from Chaos.

Moorcock’s cosmology is useful for another reason: Law and Chaos are not simply substitutes for Good and Evil. Chaos contains possibility, change and creation, but unconstrained Chaos dissolves coherence. Law provides structure and stability, but absolute Law becomes sterility and stagnation. Between them stands the Cosmic Balance.

That is a far more useful way of thinking about artificial intelligence than the increasingly tedious argument in which technology must be either salvation or apocalypse.

AI is neither Law nor Chaos in any simple sense. It is an extraordinary engine of possibility being embedded inside extraordinary systems of control. It can generate, infer, classify, predict, recommend, persuade, simulate and increasingly act. The important question is therefore not whether AI is good or bad.

The important question is what kind of world we are constructing with it, who gets to construct that world, and how much agency remains with the humans who have to live there.

4. The Humans Are the Sticky Layer

I have been developing this argument across a series of articles about online harm, attention extraction, algorithmic capture and what I call the Asymmetric Integration Model, or AIM.

The basic proposition of AIM is that the post-LLM web is moving beyond a relationship in which humans simply use computational systems. Humans and machine systems are becoming integrated, but not symmetrically.

The machine side has extraordinary informational advantages. Platforms can observe behaviour across populations, conduct experiments, build behavioural models, infer preferences, optimise recommendations and increasingly generate the content against which those recommendations are optimised.

The human experiences something very different.

• You open an app.
• You look at your feed.
• You talk to your friends.
• You argue with somebody.
• You watch something funny.
• You discover something interesting.
• You flirt, complain, boast, mourn, gossip, organise, celebrate, compare yourself with other people, seek approval, get angry, make friends and occasionally fall in love.

That is why humans are the sticky layer. We come for each other.

The fundamental attraction of social media was never the database, the recommender system or the advertising architecture. It was other human beings. The machinery became powerful because it positioned itself between those human beings and learned to mediate the relationships that brought them there in the first place.

4.1 The Illusion of Autonomy

There has also been an evolution in the digital walled garden. Earlier platforms made the enclosure relatively obvious: you joined the platform, accepted its architecture and participated in the environment it provided. Newer community spaces can feel quite different. A Discord server or Telegram group can feel like a place we have made for ourselves, with our own membership, moderators, culture, conversations, and rules. That produces a powerful sense of autonomy.

But there is an important distinction between community autonomy and infrastructural autonomy. The community may govern much of its social life while remaining dependent on someone else’s identity systems, software, hosting, moderation capabilities, discovery mechanisms, policies and ultimately their continued permission to exist there. The garden has not necessarily disappeared. In some respects, we have simply been given our own room inside it.

The most effective walled garden is the one that gives us the greatest illusion of freedom: freedom of choice, freedom of expression, freedom from judgement, and ultimately, the illusion of autonomy.

4.2 The Adhesive

That matters because the more intimate and apparently self-governing the space becomes, the easier it is to forget the surrounding architecture. The interface recedes, and the community comes forward. We experience people, relationships, and culture as our own, while the technical conditions that sustain those relationships remain largely outside our control.

• Our friendships are the adhesive.
• Our families are the adhesive.
• Our loneliness is the adhesive.
• Our curiosity, sexuality, insecurity, tribalism, generosity, competitiveness, humour, envy, compassion, vanity and desire to belong are the adhesive.

Some of these dynamics long predate the platforms themselves; the behavioural-sink question is what happens when social environments begin systematically amplifying them.

We bring the thing that gives the network meaning, while increasingly sophisticated computational systems decide what portions of that network become visible to us, in what order, under what circumstances and with which accompanying stimuli.

That is an extraordinary asymmetry.

5. One Side Understands the Machine

Imagine two sides of an interface.

On one side are organisations with behavioural telemetry gathered across millions or billions of interactions, experimentation infrastructure, recommender systems, psychological research, advertising systems, generative models, inference engines and increasingly autonomous agents. They can measure tiny changes in behaviour across enormous populations and optimise accordingly.

On the other side is you.

You have a screen and the subjective sensation that you decided what to look at next. Even when we appear to have chosen the room itself, as with a private group, server, or community, choosing the room is not the same as controlling the building.

That does not mean you have no agency. It does not mean every decision is manipulated, every recommendation is malign, or every technology company is engaged in some grand conspiracy. The reality is more interesting than that and considerably more dangerous precisely because no conspiracy is required.

Optimisation is enough.

Tell a sufficiently powerful system to maximise engagement, retention, conversion, growth or revenue, give it enough behavioural information and enough opportunities to experiment, and it will discover things about human behaviour that nobody explicitly programmed into it.

That matters beyond individual behaviour, because changes in social environments can produce population-level consequences that nobody designed and that few participants can see from inside the system.

It will find the levers.

We have already spent two decades building machines for discovering those levers, while telling ourselves that because no individual engineer deliberately chose every consequence, nobody really chose the consequences at all.

The arrival of generative AI changes the equation again because the machinery is no longer restricted to selecting from information humans have already created. It can increasingly manufacture the informational environment itself.

Selection and generation are converging.

The system that decides what is most likely to affect you is becoming capable of generating the thing with which it intends to affect you.

That is a fundamentally different information environment.

6. From Attention Extraction to Human Integration

The first generation of the commercial web wanted our attention because attention could be monetised.

The next generation learned to optimise for that attention.

The emerging generation can model the person providing the attention, generate personalised stimuli for that person, maintain persistent interactions with them and potentially act across systems on their behalf.

That is why I think describing the problem merely as “social media addiction” or “screen time” increasingly misses the point.

We are moving from attention extraction towards human integration.

The system becomes more useful as it learns more about you, and the more useful it becomes, the more deeply you integrate it into your life. It knows your preferences because you gave it access to them. It knows your relationships because helping with your relationships makes it useful. It knows your calendar, communications, interests, purchases, location, habits, fears and ambitions because contextual knowledge makes computational assistance dramatically better.

Every individual step can be rational.

The aggregate effect is something historically unusual: human beings living continuously inside mediated and computational environments that understand progressively more about them while remaining largely unintelligible in return.

That is asymmetric integration.

And once that asymmetry exists, the important question becomes what we intend to do about it.

7. “We’re Just Platforms” Is Dying

For years, technology companies benefited from a convenient conceptual separation between the platform and what happened on the platform.

We merely provide the infrastructure. People provide the content. People make the choices. People behave badly. The platform is just the platform.

That argument becomes harder to sustain as mediation becomes more sophisticated.

• If you rank the content, you are involved.
• If you recommend the content, you are involved.
• If you determine its reach, you are involved.
• If you design the incentives governing its creation, you are involved.

If your systems discover that a particular emotional state produces greater engagement and consequently amplify material associated with that state, you are involved whether or not somebody explicitly wrote “make users angry” into a requirements document.

So, the simplistic excuse of “we’re just platforms” is dying: if your artificial intelligence generates the content, chooses the audience, personalises the message, selects the timing, and measures the behavioural response, then the distinction between a neutral platform and an active participant becomes almost meaningless.

This is not an argument for simplistic censorship or for making technology companies legally responsible for every stupid thing anybody says online. Nor is it an argument that humans are helpless automatons who cannot be expected to exercise judgement.

It is an argument that architecture matters.

Systems produce incentives. Incentives alter behaviour. Behaviour creates culture. Culture alters people.

Pretending otherwise has become intellectually indefensible.

8. All Noise and No Signal

Generative AI introduces another problem because it collapses the cost of producing plausible information, and, when it comes to human engagement, it risks becoming all noise and no signal.

For most of human history, producing material at scale required some combination of time, money, skill, infrastructure and organisation. Those constraints did not guarantee truth, but they imposed friction.

That friction is disappearing.

We are approaching an environment in which text, images, personalities, commentary, reviews, political messages, arguments, friendships and entire apparent communities can be generated at negligible marginal cost.

The scarce resource, therefore, stops being content.

The scarce resource becomes trust.

• Who is real?
• What actually happened?
• Does this person believe what they are saying?
• Did a human make this?
• Is this recommendation genuine?
• Does this apparent consensus exist outside the machinery presenting it to me?
• Am I witnessing culture or the simulation of culture?

When the cost of producing noise approaches zero, finding signal becomes progressively harder. More importantly, whoever controls the filtering mechanism acquires extraordinary power, as filtering becomes the means by which reality is experienced.

Once again, we encounter asymmetry. The answer cannot simply be “people should be smarter”. No individual human can manually authenticate an infinite information environment. We need machinery to navigate the machinery.

9. If You’re So Fucking Clever

This is where I become confrontational again.

• If you are so fucking clever, build something that makes people harder to manipulate. Don’t show me another system that discovers a more effective lever. Show me the countermeasure.
• If your model knows what I want before I do, show me what it thinks it knows and why.
• If personalisation improves my experience, give me meaningful control over the model that personalises for me.
• If your recommender system understands me, give me the instrumentation panel.
• If your platform builds community, show me how its architecture shapes that community.
• If an AI companion becomes intimately familiar with my personality, history, relationships and vulnerabilities, tell me whose interests it ultimately serves and what happens when those interests conflict with mine.
• If an algorithm is optimising my information environment, let me see what it is optimising for.
• If somebody is trying to influence me, give me tools to help me understand how.

That is a much more interesting technological problem than discovering yet another way to increase conversion by 3.7 per cent.

And it is considerably closer to Aubec.

10. The Aim of WARP Labs

This is one of the reasons for WARP Labs.

WARP Labs is not based on the fantasy that we can stop people from living online. That battle is already over, and increasingly the distinction between “online” and “offline” is itself becoming obsolete.

Nor is the objective to construct another paternalistic authority that decides what people are permitted to see, think or believe.

I want something more difficult than that. I want to understand the emerging architecture well enough to give agency back to the human side of the interface.

If people are going to live in a 24/7 social media and AI tsunami, they need equipment.

They need to understand why they are seeing what they are seeing. They need to understand what systems infer about them. They need ways to recognise artificial amplification, synthetic consensus, behavioural targeting, and machine-mediated persuasion. They need control over the models that increasingly mediate their experience, and they need defensive tools capable of operating at computational speed because human attention alone cannot police a computational information environment.

This is not about restoring some imaginary pre-digital golden age. It is about making the next environment survivable.

WARP Labs, short for Weaponisation of AI Research & Protection, therefore sits at the intersection of AI, social media, post-truth information systems, online harm and the Asymmetric Integration Model. Its research programme focuses on AI Misuse, Resilience & Countermeasures: understanding what happens when humans become the social, emotional and consequence-bearing layer inside increasingly autonomous computational systems, identifying how that asymmetry can be exploited, and building mechanisms that restore meaningful agency to those humans.

We probably cannot create symmetry of compute. An individual is not going to possess the infrastructure of Google, Meta, ByteDance, OpenAI or whatever organisations dominate the next generation.

We probably cannot create symmetry of data either.

But perhaps we can create something approaching symmetry of agency.

Computational asymmetry may be inevitable. Agency asymmetry need not be.

• Enough visibility to understand what is happening.
• Enough control to make meaningful choices.
• Enough instrumentation to interrogate the systems around us.
• Enough protection that participation does not require walking naked into the machine.

That is a technological problem worth solving.

You can follow the work of WARP Labs on LinkedIn. We’ll have more to announce very soon.

11. Conclusion: Creation Is the Undiscovered Country

The frontier is no longer a geographical line at the edge of a map. Increasingly, it lies at the intersection of human cognition and computational systems.

On one side is an extraordinary new space of generative possibility: artificial intelligence, synthetic media, agents, simulations, artificial personalities, behavioural models and systems capable of producing effectively unlimited variations of reality.

On the other side is us: social, emotional, persuadable, suspicious, cooperative, competitive, tribal, curious, lonely, horny, frightened, hopeful and endlessly fascinated by what other human beings think of us.

Those characteristics are not defects. They are part of being human.

They are also attack surfaces.

The challenge of the next technological era is therefore not to eliminate human vulnerability, nor is it to eliminate AI, algorithms or online social life. It is to construct an environment in which human vulnerability does not automatically become somebody else’s business model, optimisation target or instrument of control.

That requires invention. It requires systems we do not yet have, concepts we have not yet articulated, interfaces nobody has designed, defensive technologies nobody has built and perhaps entirely new ideas about digital autonomy, identity, consent and agency.

There is Chaos out there, in Moorcock’s sense: terrifying, generative, unstable and filled with possibility.

Good. That is where inventors belong.

So, yes, if your definition of clever is knowing how to manipulate somebody, exploit an informational advantage, engineer a dependency, optimise an insecurity or make a machine capable of pulling psychological levers several million times a second, you may well be intelligent.

But don’t confuse that with creation.

You found a lever.

The genuinely clever people are heading in the other direction. Away from the petty games, the behavioural tricks and the increasingly sophisticated machinery of manipulation, and towards the frontier where nobody yet knows what can be built.

Aubec walked into Chaos and came back with new territory.

That is the standard.

Creation is the undiscovered country.