Collaborative research on AI safety is vital

York: So, there’s been a lot of chatter about AI safety lately, especially with Geoffrey Hinton, the so-called ‘Godfather of AI,’ raising alarms about the risks. He’s worried that AI could pose a serious threat to humanity in the next few decades. But here’s the thing: many believe that the best way to tackle these concerns is through collaborative research on AI safety. It’s all about getting everyone involved, including regulators.

Right now, AI is mostly tested after it’s been developed. They use “red teams” to try and find flaws, but that’s just not enough. We need to design AI with safety in mind from the get-go. This means tapping into the know-how from industries that have been doing safety right for ages.

Hinton thinks the risks aren’t something that’s intentionally built into AI, so why not take steps to avoid any potential disasters? While I don’t fully agree with his take on the level of risk, it’s clear that we should be proactive.

In fields like aviation, there are strict limits on how quickly safety can be compromised because of the physical systems involved. But with AI, there’s no such limit, which is why regulation is crucial. Ideally, we should assess risks before deploying AI, but the current methods just don’t cut it. They often overlook important factors like the specific application or how widespread the deployment is.

Regulators need the authority to pull back AI models if necessary, and companies should have ways to prevent misuse. Plus, we need to focus on gathering data that helps predict risks rather than just looking at what’s already happened. It’s a tough challenge, but if Hinton is right about the dangers we face, it’s something we must tackle head-on.