Could AIs increase the risk of nuclear war?
Analysts from the RAND corporation think modern technologies and AIs are dangerously weakening the principle of mutually assured destruction and thus increase the risks of nuclear war.
Operation Upshot-Knothole – Badger
By Photo courtesy of National Nuclear Security Administration / Nevada Site Office Public domain, via Wikimedia Commons
Artificial Intelligences are THE thing of our time. They can be great helpers in almost any field. But it also shows that they reveal all our fears. Based on what the prestigious research institution – RAND Corporation – think AIs could potentially destabilize the fragile balance of mutually assured destruction and bring the world closer to a nuclear cataclysm.
And why exactly? AIs could potentially evaluate the countries situation as dangerous (in terms of its nuclear arsenal) and force either it (in the case of the country lagging behind) or another to take drastic measures. Another danger lays in the fact that an AI with bad data could convince the leaders of a nation's military to use nuclear weapons when not necessary. Last year RAND ran a series of workshops which resulted in a report entitled “Security 2040: How technology, people, and ideas are shaping the future of global security”.
Let’s say that AIs increase the ability of Country A to aim at and destroy the nuclear arsenal of Country B. And Country B is lead by paranoid leaders. That could lead to a completely different decision-making process and change of strategy. That could mean building many more nuclear weapons and in an extreme case, even a first strike strategy. The workshops essentially show that AIs aren’t only a useful tool but also a dangerous one. They improve defense system, attacks system, potentially improve the decision making process. But the question is how the opponent will respond.
Sources:
I don't trust AI at all, all of these sientists who say it's safe are just guessing
I think it will for one simple reason:
All factors can't (yet) be quantified, and every blind spot increases the risk that a sampling bias in your data will generate incorrect results. AI models may map close to reality, but those hidden variables will always make over-dependence on them risky. As nation states lean more and more on the advice of computers and less on a human understanding of the complexities of geopolitics, we risk miscalculated plays kicking off a causal chain that could lead to nuclear war with the wrong leaders in power at the wrong time.