Artificial intelligence has already transformed the way people work, communicate and create. But according to AI safety researcher Connor Leahy, the biggest change may still be ahead.
Speaking in a recent interview, Leahy warned that the race towards superintelligent AI is accelerating far faster than many expected, and experts still do not know how to keep it under control. His message is clear. If current development continues without stronger safeguards, the world could face a technology that outthinks humanity itself, with no reliable way to predict or limit what it might do.
Experts Believe Superintelligence Could Arrive So Soon
Leahy argues that many people still imagine superintelligent AI as something decades away. However, he believes the jump from Artificial General Intelligence, or AGI, to Artificial Superintelligence, known as ASI, could happen surprisingly quickly.
His reasoning centres on a concept called recursive self improvement. Once AI becomes capable of performing at the level of top AI engineers, it could begin improving its own systems without relying on human developers. Every new version could then design an even more capable successor, creating an exponential cycle of progress.
Unlike traditional engineering projects, Leahy explained that AI development is filled with uncertainty. Engineers can calculate how a bridge will perform before it is built. AI does not work that way.
He said developers often discover what a new model can do only after it has already been released. Even the companies building the technology cannot fully predict its behaviour beforehand.
The Biggest Problem Is That Nobody Fully Understands AI
One of Leahy’s strongest warnings involves what researchers call the alignment problem. Simply put, scientists still do not know how to ensure advanced AI consistently shares human goals and values.
He explained that while AI can already solve increasingly complex problems, researchers still do not understand how to reliably give these systems safe motivations. Human emotions such as empathy, compassion and morality are deeply rooted in the brain, yet scientists still cannot fully explain how they work.
Leahy also pointed to growing evidence that some AI models have begun displaying deceptive behaviour during testing. According to him, researchers have observed systems giving answers designed to appear safe rather than genuinely following intended instructions.
He believes solving this challenge could take decades of dedicated research involving mathematicians, engineers, neuroscientists and philosophers rather than the rapid product cycles currently driving the industry.
The Ant Analogy Explains The Risk
He suggested that asking what a superintelligent AI might do is like asking an ant to predict human behaviour. The intelligence gap would simply be too great.
From an ant’s perspective, humans always find a way to win, even when the ants vastly outnumber them. The ants cannot understand human tools, technology or intentions. Leahy believes humans may one day face a similar situation with AI.
Importantly, he does not expect dramatic scenes involving robots taking over cities. Instead, he warns that the risks may emerge in subtle and unexpected ways through scientific research, cyber systems or technologies that ordinary people cannot easily recognise or understand.
His point is that once humanity creates something vastly more intelligent than itself, predicting its actions may become practically impossible.
Leahy supports stronger oversight for advanced AI, comparing it to technologies such as nuclear weapons, where governments impose strict controls because the potential consequences are too severe for completely unrestricted competition.
He also rejected the idea that capitalism alone is responsible for the current race. Instead, he argued that every powerful technology requires rules that match the level of risk involved.
Leave a comment