AN ADAPTIVE CRISIS

The debate is not about better model testing.

There’s a distinction that was popularised by Harvard prof, Ronald Heifetz, in Adaptive Leadership theory. Knowing the difference between what are technical problems and adaptive problems.

It goes something like this. Technical problems can be difficult, but they are usually solvable within existing systems and with existing knowledge. The challenge is there, expertise already exists somewhere, and the institution or whatever, doesn’t have to change.

For example, a leaky pipe or a blocked drain is a technical problem. Building a bridge, flying around the moon and debugging software are mostly technical problems. They require expertise, for sure, but not fundamental changes to human values or social organisation.

Adaptive problems are different, though. Adaptive problems come up when eg defining the problem itself is contested, when stakeholders disagree about goals, and when the existing frameworks may no longer be adapted to the new environment.

Climate change is an adaptive problem, social media turned out to be an adaptive problem, and AI looks like it’s becoming one too.

Yet much of the mainstream AI governance debate still treats AI problems as if they are only technical challenges. We hear about safety evaluation, model testing, alignment protocols, ‘red teaming’ (researchers deliberately provoking deceptive or dangerous outputs from models etc), release thresholds (the predefined conditions a model must meet before it is allowed to be deployed publicly or given broader access), and oversight frameworks. These are all attempts to make AI legible as a manageable engineering (technical) problem.

But what if the real issue is that AI is now changing the conditions under which human institutions even operate? That possibility changes the conversation.

Universities, for example, have never been just knowledge-transfer systems. Evolutionarily, they are prestige-allocation systems. They certify competence, regulate access to high-status coalitions, and signal things like intelligence, and conscientiousness. AI destabilises all of that. Historically, producing your polished thesis implied years of accumulated knowledge. Now a mediocre student with good prompting skizzles can generate outputs previously associated with strong educational attainment. So, if AI muddles the relationship between visible performance and underlying ability, people soon begin questioning the legitimacy of the sorting system itself.

The thing is, human institutions have never been abstract machines optimised for truth or wisdom. They are adaptive coalitional systems shaped by competition, resource management, coordination pressures, and survival.

In the postmodern Trump-ian era, politics becomes even less about governing material reality and more about controlling narratives, identities, and media attention.

Coherence gives way to spectacle and contradiction ceases to matter because the objective is no longer consistency, but coalition maintenance and dominating attention.

(Trump didn’t invent this dynamic btw, he’s just accelerated and exposed it. He is a highly adapted organism for the media ecology produced by postmodern culture and techno capitalism.)

Both governments and corporations look for strategic advantages. But none of these legacy systems evolved to manage technologies capable of transforming cognition, labour, persuasion, and information flows simultaneously.

This is why so much of the current AI governance discussion feels woefully inadequate. The phrase ‘technological republic’ captures something of this potential new order. It reflects the growing (and justified) unease that the big tech firms are no longer just regular mega-corporations operating within the market for strategic advantage or even political ‘influence’.

More and more, they resemble quasi-sovereign actors hell-bent on colonising the world’s informational and cognitive infrastructure. The whole pie.

The big AI firms now influence communication systems, educational environments, labour displacement, defence capability, and public attention at enormous scale. Most possess more frontier AI expertise than entire governments and so are becoming deeply embedded in national security and infrastructure planning.

From an evolutionary standpoint, this is fundamental because adaptive problems trigger coalition formation. Throughout history, all periods of major technological transition have produced new alliances between political power and economic power. Industrialisation reshaped the relationship between states, industrialists, labour, and finance. Broadcast media altered political persuasion and mass coordination. The internet transformed information flows and institutional authority.

The worry about current governance discussion/arrangements is that they function not as vague ‘safety mechanisms’ at all, but as coalition-building exercises between states and frontier technology firms exploiting a time of massive uncertainty. The public narrative (technical) is safety and responsibility, the deeper (adaptive) logic is about who gets to shape the future cognitive environment of society.

This also helps explain why government and big tech in-bed-together AI governance announcement feels like so much bullshit. Public summits, declarations, safety institutes, and voluntary frameworks create visible signals of coordination and control. In evolutionary terms, these rituals serve an important function. Humans evolved in social environments where displays of competence, cooperation, and legitimacy helped stabilise groups during periods of change. Modern institutions still rely heavily on these ancient psychological mechanisms.

The danger is that adaptive problems are often politically reframed as technical problems precisely because technical problems feel controllable. Technical problems hang on to the assumption that existing institutions remain fundamentally good enough. This is what happened with social media. Initially framed as a technical issue involving moderation systems and platform management, it eventually became clear that social media was reshaping political behaviour, mental health, status competition, public discourse, and even human attention itself.

The technical solutions never fully solved the issue because the problem was never purely technical and so we need to stop AI governance from falling into the same hole.

Civilisations have always thrown up technologies that expand coordination and power beyond the scale their existing institutions evolved to manage. Writing transformed administration, printing transformed religious and political authority, industrialisation transformed labour and class structures and broadcast media transformed persuasion.

AI represents the next stage of that process. That is why the current moment feels psychologically and politically destabilising. Beneath the language of governance frameworks and safety partnerships lies a much older evolutionary struggle over legitimacy, coordination, and control of the human environment.

The debate is not about better model testing, it is about whether the institutions that evolved to govern industrial civilisation can successfully adapt to a world in which cognition itself has become technological infrastructure.

So the paradox is this. The institutions responsible for maintaining social coherence are increasingly incentivised to dissolve coherence for short-term coalitional advantage.

(All ofwhich may go some way to explaining why so many people feel modern politics has become hyper-real. Because the mechanisms through which legitimacy is now manufactured are moving further away from the stabilising functions institutions originally evolved to perform.)

What happens when AI and networked media systems alter the conditions under which human institutions themselves maintain legitimacy, authority and shared reality? Once institutions become optimised for surviving inside attention markets rather than coordinating material reality, their evolutionary function begins to change. They stop behaving as systems for stabilising cooperation and increasingly behave like competitors inside the Spectacle.

This is why so much contemporary governance feels performative at best. The institutions are still speaking the language of industrial-era legitimacy (expertise, procedure, administration, rationality) while operating inside a media environment that rewards theatre and outrage. The result is an institutional uncanny valley, in which everything still looks like governance, but it behaves like content.

So this is not a technical challenge we’re in, it is an adaptive crisis in the relationship between human psychology, media systems and institutional evolution. And we’re in it up to our necks.