Compute growth doesn't make a software intelligence explosion more likely
ai
LessWrong researcher Karthik Tadepalli says a popular argument about artificial intelligence has a flaw. The argument: computing power is growing exponentially, so a software intelligence explosion becomes more likely. Tadepalli's response: that logic doesn't hold up. An intelligence explosion would require improvement rates to accelerate at an increasing pace—what's called super-exponential growth. When computing power grows exponentially, the acceleration stays flat; it doesn't multiply. To trigger an intelligence explosion would require computing power itself to grow super-exponentially, through some feedback loop in chip production. That's a much heavier lift than simply building more chips. The takeaway: more compute does speed up AI research generally, but it doesn't specifically move the needle on whether an intelligence explosion is likely.
Source: https://www.lesswrong.com/posts/PMkBrDcXHoShHak7y/compute...
Listen to this story
Hear this and more stories in a personalized audio briefing.
Open The Chonkerton