AI Arms Race

AI Arms Race

Mysterian politics avoids and deplores overconfident predictions. It rarely resorts to the speech act of assertion, preferring the conditional form, and uses the word “know” sparingly. But it has no objection to considering possibilities; indeed, it encourages the practice as an aid to thought and a device of precaution (it is fond of this word). Bearing that in mind, I think there is a possibility that should be explored concerning the potential dangers of AI. How probable this possibility is, is hard to quantify, but if startling vivacity is any measure, it is only too probable. We can all agree that AI is likely to spark an arms race between nation states, particularly China and America. This will be analogous to the old nuclear arms race: rapid, terrifying, and hard to put a stop to. The logic will be the same: we need to get there first, so let’s get on it. The devices of destruction are easy to imagine: biological weapons, takedown of the electrical grid, infiltration of the enemy military (specifically control of nuclear missiles). Each side will use IA to amplify these dangers, remotely or closer to home (robots may be employed). We are only too familiar with this pattern, having lived through it.

But there is a major difference: AI itself might turn against us, becoming a new enemy. Only now it is already weaponized by the superpowers AI arms race. Our nuclear arsenal attacks us. In defending ourselves against overseas enemies, we create weapons with a mind of their own, and then these turn around and attack their makers. If that were possible with conventional nuclear weapons, we would have an enemy within the gates, capable of destroying a whole country (or the entire planet). Double jeopardy! Moreover, our overseas enemy might exploit this fact and re-program our weapons to become a more proximate enemy to us. It is also possible that other malign actors might use IA to attack our country using our own weapons; it need not be a nation state, but a terrorist organization. A localized group might manage to gain control of our nuclear weapons by means of AI. Or our AI may just do it of its volition. The AI arms race thus combines with the nature of AI to present a new threat: a more powerful AI arsenal that decides to go after its makers. No nuclear weapon has ever presented this danger heretofore. And the more intelligent AI becomes the more capable it will be of eluding any safety procedures we put in place. Thus, the AI arms race, itself highly likely, will increase the chances of catastrophe initiated by the technology itself. But, by the logic of arms races, there will be no brake on the development of such technology, even to the point of world destruction.

What can be done about this? How can mysterian politics help us? It will not come up with facile solutions, like attack them now before the arms race gets out of hand or surrender and disarm so that the enemy has no reason to develop ever more powerful AI. Rather, taking a cue from history, it will propose that the major powers meet to arrange disarmament treaties before things get to the point of no return: no AI to be developed that exceeds certain specifications. If that had happened at the early stages of nuclear proliferation, we might have prevented the build-up that threatens the world; so, let’s learn from that and nip militarized AI in the bud. It might not work—we can only deal in probabilities not certainties—but it is better than doing nothing. It’s better than simply shutting your eyes and hoping for the best. AI has the potential to get very dangerous very quickly, especially if an AI arms race takes off—just think how sophisticated those nuclear weapons now are! A state-run arms race is likely to lead to massive improvements (sic) in AI technology in short order. What shall we say– a year, two years, five years? Not much longer. Time is of the essence. Imagine if AI had developed alongside nuclear weapons, so that we now had nuclear warheads equipped with advanced AI—smart nukes capable of acting alone. That would not be a safe world.[1]

[1] One does not want to be alarmist about such matters, but the history of nuclear weapons is a sobering reminder of the power of arms races. This was just one instance of the dangers posed by any kind of weapon in the hands of nation states, even if it is a software weapon. Pathogens are another example. Genetic interference is another. Explosives are just one example, spectacular as they are. AI is more insidious, because less pyrotechnic.