Climate, nuclear and pandemic policy can guide us through the AI threat

The buzz last week over artificial intelligence was the latest revelation that it could place the future of humanity at grave risk . The news invoked nightmarish scenes of AI taking over the world, using agents to carry out its own ends independent of human control like in “The Terminator” and other Hollywood epics.

If AI is potentially existential, then at least three other developments that threaten society loom as examples: climate change, massive pandemics and thermonuclear weapons. Consideration of each may offer insights on AI and how to control, if possible, its hostile potential and harness its value for humanity.

Climate change brought a great ice age. Whether a future ice age or global heating threatens mankind, it has in the past without AI.

Massive pandemics cannot be dismissed. It is not beyond the realm of possibility that AI could produce a fatal disease that is highly contagious and incurable. Frank Herbert’s “ The White Plague ,” written in 1982, posited a bioengineered disease that only killed women, suggesting something similar could be manufactured.

Perhaps the most relevant threat comparable to AI was the creation of thermonuclear weapons. It led atomic bomb creator Robert Oppenheimer to quote from Bhagavad Gita — a Hindu scripture featuring a prince at war who expresses “misgivings about the justice of killing so many people” — to show his pessimism about what lay ahead.

Thermonuclear weapons are indeed existential; they are potentially 1,000 times greater than nuclear weapons. But thermonuclear war has been avoided, partially because control was bifurcated over military and civilian use. President Harry Truman created the Atomic Energy Commission in 1947. It originally oversaw civilian usage but was replaced by the Nuclear Regulatory Commission in 1975. Both the Departments of Defense and Energy oversaw the nuclear weapons complex.

Man-made pandemics and nuclear weapons require vast, complicated infrastructure that’s unlikely to be duplicated by individuals. And while severe climate change could be catastrophic, efforts to control greenhouse gases and other pollutants might have helped to deter some of the more dire consequences.

However, AI could make it easier to create man-made pandemics and could also impose huge damage to critical infrastructure such as power and water, which would threaten the food supply. AI in its current form could not likely build a nuclear weapon, but individuals using AI to tamper with nuclear weapons and reactors could prove disastrous. And hacks against AI infrastructure unprompted by humans are also frequent occurrences.

One solution, as with control of nuclear weapons, is to create an AI regulatory agency that puts out guidelines and is empowered with oversight of the industry. The degree of regulation will be a political decision. And, as with the International Atomic Energy Agency, this model could be replicated under the United Nations.

The Departments of Defense, Energy and Commerce, along with the CIA, would determine possible national security threats from AI and the appropriate responses to those threats. AI negotiations similar to arms reduction and control agreements could follow if properly managed. That would be far more difficult than nuclear agreements, as only a handful of states possess that capability. AI is more fungible and indeed could be exploited by individuals rather than nation-states.

What is needed is a federal framework to check, oversee and control the existential threat AI might pose to society. With the right foundations and leadership, just as a thermonuclear war has never been fought, perhaps a parallel outcome could be obtained for AI.

Harlan Ullman is chairman of a private company, and principal author of the doctrine of shock and awe. His next book, co-written with Field Marshal The Lord David Richards, former United Kingdom chief of defense and due out this Fall, is “Who Thinks Best Wins: How Decisive Strategic Thinking Will Prevent Global Chaos. “