Is the nuclear non-proliferation pact a model for regulating runaway AI?
- Written by: Weekend Times
The potential existential threat of artificial intelligence has leapt into the headlines after a departing safety researcher at AI corporation Anthropic posted on social media he believed
there was more than a one-in-ten chance AI could kill everyone.
Other Anthropic insiders went public in support of his views, saying the technology could become advanced and dangerous enough to cause human extinction within ten years.
This all followed renewed warnings from Turing Award-winning computer scientist Yoshua Bengio, who has spent the past year arguing that increasingly autonomous AI systems could develop self-preservation goals humans cannot easily switch off.
Even if the current development of AI falls short of artificial superintelligence or artificial general intelligence, many have warned it might be used by bad actors for nefarious ends.
This includes biological weapons design, cyber-attacks against critical infrastructure and financial systems, and assisting hybrid warfare efforts. More conventionally, we already know AI was used in the planning and execution of the United States raid on Venezuela at the start of this year.
A number of sceptics have argued the latest claims remain highly speculative. But there are still sound reasons to apply the cautionary principle to a technology that is developing at an exponential rate.
Of particular concern is that AI is emerging as central to growing competition between the US and China. Echoing past space and nuclear races, both superpowers fear falling behind, making any form of restraint a strategic risk rather than a virtue.
We have been here before.
Building the nuclear taboo
Early in the Cold War arms race, nuclear weapons were treated like an offensive tactical option. After a number of critical defeats during the first year of the Korean War (after China sided with North Korea in 1950), key US military figures – notably General Douglas MacArthur and Secretary of State John Foster Dulles – saw nuclear weapons as a way to regain control.
Fortunately, calmer heads prevailed. But the fact they were seen as a legitimate option shows how relatively recent nuclear arms control is.
That taboo against the use of nuclear weapons had to be built. And it was built out of fear, not goodwill.
It took the Soviet Union closing the technology gap – first with its own atomic test in 1949, then with the development of thermonuclear weapons in the 1950s – for the mentality on both sides to begin to shift.
Only once both powers had something to lose did nuclear weapons start being treated less as tools of war-fighting and more as instruments of deterrence. They became seen as a defensive last resort rather than an offensive first move.
The Cuban Missile Crisis in 1962 is often remembered as the closest the world came to nuclear Armageddon. And while the stakes were incredibly high, given both sides were developing nuclear “triad” capabilities involving aircraft, submarines and intercontinental ballistic missiles, it also helped alter the way the US and the Soviet Union thought about nuclear weapons.
In 1963, the Moscow-Washington hotline was implemented and the Limited Nuclear Test Ban Treaty was signed. This helped usher in a period of relative geopolitical calm known as “détente”.
In 1970, the Treaty on the Non-Proliferation of Nuclear Weapons came into force. While less than perfect – several nuclear weapons states are not signatories – the treaty showed that a degree of international cooperation on a crucial problem could be achieved.
This didn’t end the rivalry between Washington and Moscow, nor did it stop proxy wars from Vietnam to Afghanistan. But it did establish a base line: direct superpower confrontation, and with it the risk of nuclear escalation, became something both sides worked to avoid.
When the stakes are high
If the accelerating development of AI poses a similarly existential risk – and enough people building the technology now say it might – the hope has to be that Washington and Beijing eventually treat the issue with something like the seriousness previously applied to nuclear weapons.
That would mean moving past mutual suspicion towards a basic, shared recognition that an uncontrolled AI race benefits neither side if the technology itself is the risk.
None of this means the potential threat posed by AI will dissipate completely. In the case of nuclear weapons, despite the aspiration of the Non-Proliferation Treaty, rogue nuclear states emerged and intensifying rivalry between the US, China and Russia has seen the gradual erosion of those frameworks.
But if there is a lesson worth drawing from the first nuclear age, it is that even bitter rivals can find their way to a shared understanding once the stakes became unmistakably mutual.
Getting there sooner rather than after a crisis would be the more responsible path this time.
Nicholas Ross Smith does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
