“AI is still underhyped and underestimated”

The global race for AI dominance is about more than technological superiority – it will define geopolitical power plays, underpin national security, influence resource-based conflicts and have far-reaching economic outcomes.

And yet AI remains “underhyped and underestimated,” according to Airmic’s opening keynote speaker, Dr Keith Dear, founder and CEO of AI strategy business, Cassi, who warned that most humans are failing to comprehend the scale and speed of “the most profound revolution in the history of humanity.”

Anthropic’s Claude Mythos evolution is a case in point, he said: the AI model – banned from widespread use until recently – is considered “so capable it represents an active threat to humanity and society,” with superhuman cyber capabilities and the emerging risk of enabling biothreats.

The global race for supremacy: US leading while Europe lags

During a fireside chat with Airmic CEO, Diane Maxwell, Dear warned that the US is “in the lead” as it actively pursues global AI dominance in the frontier to match and exceed human intelligence.

China is six to nine months behind the US, he said, but is, in parallel, adopting a strategy focused on open-source platforms and dominance in robotics that will see it capitalise on its vast industrial base.

Continental Europe, meanwhile, is “regulating its way into irrelevance,” as cultural and legal barriers continue to hinder investment, he said. The UK is striking a more balanced approach, “somewhere between the totally unregulated US and the over-regulated EU”.

The geopolitics of AI matter. “Everything we care about in defence, national security, in business – products, services, strategy, tactics, weapons – are the products of intelligence,” said Dear.

“The returns on intelligence are enormous. If a company or country has access to a superior intelligence, they're going to dominate you by outperforming.”

The Trump administration is already seeking to leverage US leadership in AI, as demonstrated by its decision, just days before Airmic's conference, to order Anthropic to block foreign nationals from accessing its most advanced AI models:

“That is extremely controversial. Now every new model release has to go through the US government,” said Dear.

A “jobs apocalypse” is possible

While the geopolitics of AI may feel one step removed from day-to-day lives, the replacement of potentially millions of jobs worldwide will have profound economic and social consequences, and Dear believes governments and businesses must prepare for a deep and far-reaching impact.

A “jobs apocalypse” is not an impossible scenario and could happen “sooner and faster” than people expect, he said. Such a scenario would have an uneven effect across societies, impacting white collar jobs disproportionately, while potentially driving demand in skills such as engineering.

Even in less extreme scenarios, AI will demand widespread organisational transformation in which “entire workflows will have to be redesigned,” he added, going beyond simple replacement of existing processes.

For business leaders, understanding AI’s capabilities and limitations within the context of individual organisations is critical.

Companies should clearly define what tasks they believe humans will always perform better than machines and those that may soon be automatable. Establishing benchmarks to test AI performance against these tasks is a useful strategy, he advised.

The teams doing the testing must be carefully incentivised to prove machine capabilities – otherwise the pressure, implicit or explicit, can be to slow-roll, or validate assumptions of what AI will never be able to do.

Turning to limitations, AI systems can’t tell you what you should want or value:

“AI can’t tell you what you want, it has no desires. That’s where humans are irreplaceable – only we can decide what we want and the trade-offs we’re willing to make to reach our goals.”

For that reason, Dear says, philosophy is going to become increasingly important as organisations grapple with what they should optimise.

Regulating what?

The ethical, data and workforce challenges associated with AI raises questions about the role of regulation.

There are two big challenges when it comes to regulating AI. The first is deciding what, exactly, is being regulated. AI is not a discrete industry or product, Airmic’s Maxwell noted. At its core, AI is mathematics and computing power applied to data. Regulating AI therefore raises an awkward question: can you really regulate mathematics?

The technology is also borderless and deployed globally. That makes it difficult to determine who should regulate it, how rules should be enforced and whether regulators have the technical expertise to keep pace with a rapidly evolving field.

The second challenge is competitiveness. Every new regulatory requirement carries a cost, diverting talent, time and capital away from innovation and into compliance. The question for policymakers is whether they can manage AI's risks without choking off the innovation needed to remain competitive:

“Too much regulation means that the AI companies don't make progress, and that’s why European AI companies are struggling,” said Dear.

Dear’s overarching message was clear. AI is transforming lives in a way that most humans and businesses cannot yet comprehend at an unprecedented pace, and societies, governments and businesses must prepare:

“Are we building a risk? Yes we are, but the far bigger risk that most organisations are taking is not thinking about where AI is going, not building for that future. And the risk likely won't be visible to many until it's too late.”