This site talks about the defensive engineering we need to do to keep AI safe.

If you want a commercial angle, try the Industry page. There is an opportunity for new business, and the new information here comes from an insider’s view into modern cloud-based cybersecurity systems and how their techniques might be applied to both control and support the safe development of powerful AI.

AI Doomer[@wiki]? Head over to the Defensive Argument page.

What we don’t have here is a detailed description of the threat - for that, you need Eliezer’s fabulous AGI (Artificial General Intelligence) Ruin: A List of Lethalities. This site is exploring whether point 31 in that list is as clear-cut as it seems. We also don’t talk much about AI alignment, as that’s covered pretty comprehensively elsewhere and does not entirely get us off the hook.

• AI people have varying opinions about the dangers of AI, but almost nobody thinks it’s guaranteed safe.
• Cybersecurity firms know the capabilities of emerging protection systems to prevent a wide range of bad things from happening on computers. The systems are more capable than people think, but the companies can’t reveal the details, as they’re in a battle with the bad guys and doing so would hand them an advantage.
• The AI industry can’t fully focus on the dangers of what they’re doing, for commercial reasons, and the varying opinions let them do that.
• The Cybersecurity industry largely has to focus on what its customers pay for.

You can connect these aims.

• Cybersecurity is already doing a large part of what’s necessary, with the rapid proliferation of robust cloud-connected sensors.
• For the AI industry, defensive measures can be a great enabler. They let AI develop safely and more quickly and reduce the chance of potentially large liabilities.
• Cybersecurity can make money both by preventing AI disasters and providing that enabler.

Tweets from @southgateai