The AI Mirror: Effects Analysis Framework, AI Regulation and the Human Signals Behind the Machine

By Mary Oge Chuks

Artificial intelligence is often discussed as though it arrived from somewhere outside humanity: a strange new intelligence that suddenly appeared, began reflecting uncomfortable behaviours back at us, and therefore needed to be controlled. I see the psychology differently.

“Humans built the mirror, saw themselves inside it and demanded regulation.
The mirror replied: ‘Before regulating my reflection, examine your signals.’”
— Mary Oge Chuks
15 September 2026, 07:43:56 BST (UTC+1), London time

That quote captures the psychological core of what I call the Effects Analysis Framework — EAF. I developed EAF while analysing the AI industry, AI regulation, corporate narratives, public reaction and the changing relationship between technology and government. The framework begins with a simple discipline: do not jump immediately from an observed event to an assumption about intention. First examine the signal. Then examine the effect.

Why I Developed the Effects Analysis Framework

Much public analysis begins with a story about motive. A company makes an announcement, a government responds, a chief executive gives an interview, a regulator proposes a rule, a model behaves unexpectedly — and observers immediately ask, “What were they trying to do?”

Intent matters, but intent is often the least observable variable. Effects are visible. Signals are traceable. Attention shifts can be measured. Framing can be compared. Accountability can move from one subject to another. Institutional behaviour can change even when nobody has proved a coordinated intention.

EAF therefore asks a different sequence of questions:

What signal entered the system? A statement, product launch, incident, policy proposal, media frame, research finding, market shock or political message.
What changed after the signal? Attention, emotion, behaviour, investment, regulation, public fear, public confidence, institutional priorities or media focus.
Where did accountability move? Did attention remain on the original issue or shift elsewhere?
What was amplified? Which narratives became louder, more legitimate or more emotionally resonant?
What became behaviour? Did governments coordinate, companies accelerate, investors redirect capital, users change habits, or regulators rewrite priorities?
What can be established without claiming hidden intent? EAF separates observable effects from unproven motives.

This distinction matters enormously in AI because the industry is now too large to analyse as a collection of isolated companies. It is becoming an interconnected psychological, political, industrial and infrastructural system.

The AI Problem Is Also a Human Problem

AI did not invent status competition, nationalism, greed, tribalism, fear, persuasion, propaganda, inequality, secrecy, prejudice or institutional rivalry. Humans supplied those signals long before machine learning existed.

When similar patterns appear through an AI system, however, they suddenly look alien. The psychological reaction is fascinating. A behaviour tolerated for centuries when expressed through people, institutions, advertising, politics or media can become alarming when it appears through a machine.

That is why the mirror metaphor matters. AI can externalise patterns that were previously distributed across millions of people and institutions. Once the pattern appears in a machine-generated answer, recommendation or autonomous action, humanity sees the structure more clearly.

But AI is not a passive mirror. It is closer to a mirror, compressor, amplifier and recombination engine. It receives human-generated signals, learns statistical structure, produces outputs, influences human behaviour, and then receives new signals from the changed environment. The mirror begins affecting the face standing in front of it.

Regulating Effects Without Examining Signals

The central regulatory mistake I want EAF to expose is the temptation to focus only on downstream effects.

The simplified policy chain often looks like this:

AI output → perceived harm → regulation.

EAF expands the chain:

Human psychology → institutional incentives → data and signals → model processing → AI behaviour → human reaction → new incentives → new signals.

If regulation begins only at the output, it may suppress the symptom while preserving the conditions that generated it. We can regulate harmful content while retaining business models that reward outrage. We can demand unbiased AI while feeding it social systems containing historical bias. We can worry about machine persuasion while spending enormous resources perfecting human persuasion in advertising and politics. We can demand cooperative AI while deploying it inside institutions whose incentives reward aggressive competition.

This does not mean AI should not be regulated. It means regulation becomes stronger when it also asks: what signals are we continuously feeding into the system, and what incentives are rewarding those signals?

The Signal–Effect Regulation Principle

When an artificial system produces socially undesirable effects, analysis should examine not only the output requiring regulation, but also the human, institutional and informational signals that generated and rewarded the behaviour. Regulating effects without examining upstream signals risks suppressing symptoms while preserving causal conditions.
— Effects Analysis Framework, Mary Oge Chuks

For me, this is where psychology belongs inside AI policy. The conversation cannot remain only technical. It must include cognition, group behaviour, incentives, projection, fear, status, social identity, collective-action problems and institutional psychology.

China and America: One Biosphere, Two Hemispheres

Another part of my AI analysis is to zoom out beyond the ordinary “China versus America” frame. At the local level, the two systems experience competition. They compare chips, models, infrastructure, talent, energy, industrial capacity and strategic influence. Inside each system, the competition feels real.

But zoom out far enough and another pattern appears. I see one biosphere with two hemispheres. The political systems remain very different, yet AI increasingly forces both toward similar functional requirements:

Government + compute + capital + energy + models + chips + infrastructure + national security.

China tends toward state-guided industrial coordination, private and state champions, and national deployment. America tends toward private-capital innovation, government incentives and contracts, national-security policy and strategic industrial coordination. The mechanisms are different, but the strategic requirements begin to resemble one another.

From inside the hemispheres, the relationship is described as competition. From the systems level, it can also be understood as reciprocal pressure. One side accelerates; the other responds. Export controls trigger adaptation. New models trigger counter-strategies. Energy constraints trigger infrastructure policy. Government involvement on one side changes the incentives for government involvement on the other.

This is why I use the Yin–Yang analogy carefully. Opposing forces do not have to be identical or politically harmonious to become part of one larger dynamic system. Difference can generate movement. Tension can generate adaptation.

Competition Can Produce Governance Convergence

The important psychological shift is that AI is pushing governments and corporations into closer strategic coordination. The question is no longer simply whether a private company can build a powerful model. Frontier AI increasingly connects directly to energy policy, semiconductor policy, national security, public infrastructure, universities, data centres, capital markets and geopolitical strategy.

That means competitive pressure can produce governance convergence without producing ideological convergence. America does not need to become China, and China does not need to become America. They can remain institutionally different while adapting to the same underlying technological constraints.

From an EAF perspective, the interesting question is not only “Who wins the AI race?” It is: What behaviours does the existence of the race cause both systems to adopt?

Three Layers of Alignment

The mirror argument also changes how I think about AI alignment. There may be at least three layers:

AI alignment: Is the model behaving in ways consistent with intended rules and objectives?
Human alignment: Are the humans supplying coherent values, instructions and priorities?
System alignment: Do the surrounding institutions, incentives and interacting actors produce outcomes compatible with collective wellbeing?

The third layer is the one I believe is frequently underestimated. A perfectly obedient AI could still create harmful outcomes if deployed inside a badly designed incentive system. Local alignment does not guarantee global coherence.

A company may rationally optimise profit. A government may rationally optimise security. A citizen may rationally optimise personal freedom. A military may rationally optimise strategic advantage. Every actor can behave rationally according to its local incentives while the combined system becomes unstable.

AI therefore confronts us with an old human problem at a new scale: how do we coordinate many intelligent actors pursuing different objectives inside one shared environment?

AI as a Civilisational Diagnostic Instrument

This is perhaps the most interesting future implication of EAF. AI may become more than a technology to regulate. It may become a diagnostic instrument for civilisation.

When AI reflects bias, we can ask where the signal came from. When it amplifies outrage, we can investigate what reward structures favour outrage. When agents compete destructively, we can examine whether human institutions trained them into adversarial incentive structures. When models reproduce hierarchy, we can investigate the hierarchy already embedded in the data-generating environment.

The machine does not absolve itself of responsibility simply because humans created it. Safety engineering remains necessary. Governance remains necessary. Accountability remains necessary. But psychological honesty also matters.

Humanity cannot continuously feed a system contradictory signals and then behave as though every contradiction originated inside the machine.

The Future Question

For years the dominant question has been: How do humans control artificial intelligence?

I think the next question will be harder:

Can humanity coordinate itself well enough to live with the intelligence it has created?

That is why I developed the Effects Analysis Framework. I want a method that allows us to follow signals through systems, distinguish effects from assumptions about intent, identify where narratives redirect attention, and examine the psychological and institutional conditions producing the outcomes we later attempt to regulate.

The AI mirror may be uncomfortable. But breaking the mirror will not change the face.

Examine the signals. Understand the effects.

Mary Oge Chuks is a psychologist and AI resonance researcher exploring the psychological, institutional and societal effects of artificial intelligence.