How would the world deal with AI-equipped semi- or fully-autonomous nuclear weapons?
The uncertain situation in Iran and conflicting reports about its current proximity to nuclear capability bring this frightening question to the fore.
AI-interfaced nuclear weapons would present very uncertain outcomes.
For example, AI models analysing sensor data might misinterpret natural phenomena such as sunbeams or weather patterns, seeing them as missile launches.
AI lacks the contextual awareness to question such anomalies, potentially triggering retaliatory strikes before humans can intervene.
Target misidentification would be amplified by AI’s reliance on imperfect training data.
In addition, AI models are inevitably affected by the biases of their trainers, who carry their own political or racial agendas.
AI systems are also notoriously susceptible to hallucination, presenting false or misrepresented information as fact.
The danger bias and hallucination would represent for nuclear weapons systems is obvious.
Moreover, AI systems would be susceptible to sophisticated cyberattack. Adversaries could spoof AI sensors or poison training data to manipulate outcomes as per their interests or those of their allies.
There are numerous other potential pitfalls. The ethics of AI in warfare have hardly been addressed, not least the problem of accountability.
If, for example, an errant AI causes a potential or actual missile launch, who should be held accountable?
Should it be political overlords, military chiefs, AI developers or hardware designers?
There are questions, too, regarding the degree to which military command structures, particularly in developing countries, would involve big tech corporations in the procurement of their AI hardware or software.
What backdoor leverage would this offer corporate heads who ultimately pursue corporate profit above global security or the common good?
Would contractors build failsafes into their products, or simply offer weapons built to buyers' specifications?
There are myriad other ethical questions to be asked and answered. Until they are, governments of all stripes must pull back from developing AI nuclear weapons systems.
Western governments must take the lead, ensuring that the world is not subjected to a new, hyper-powered AI nuclear arms race.
AI's potential still represents unknown territory. Engineers coined the phrase "deep learning" to cover areas of AI activity and output even they don't understand.
With the eventual possibility of artificial general intelligence - or even, one day, artificial super intelligence - lurking in the shadows, humanity cannot afford to open the door to nuclear weapons powered by machine intelligence.