The Alignment Problem - Brian Christian | Modern Wisdom Podcast 297
Topics covered
Popular Clips
Episode Highlights
Ethics
The challenge of embedding ethical and moral values into AI systems is daunting, as explains. He highlights the difficulty of translating human ethics into machine-readable language, especially when human values themselves are not universally agreed upon 1. This complexity is compounded by the fact that AI systems, like those used by social media platforms, can inadvertently create harmful feedback loops by misinterpreting user behavior. notes, "There really is this question, which, to borrow a phrase from Nick Bostrom, this is philosophy on a deadline."
There are these open questions in not just ethical philosophy, but cognitive science, neuroscience even. But we don't have time to wait for the answer because these companies are just going and so we're going to have to try to essentially fix the plane mid-flight.
---
He suggests that the AI safety community is working on solutions like coherent extrapolated volition, which aims to align AI actions with what humans would want if they were more informed and rational 2.
Innovation vs. Risk
The rapid advancement of technology often outpaces our ethical and policy frameworks, creating a precarious balance between innovation and risk. and discuss how the current state of AI research is on the brink of solving problems that society is only beginning to understand 3. This disparity is exacerbated by outdated systems still in use, such as those written in obsolete programming languages, which pose significant risks if not addressed. reflects on this technological dilemma:
We have this idea that tech moves too fast for society to keep up, but in reality, a lot of these crappy machine learning systems that were developed in the 2010s are still going to be around, like, in zombie mode 20 years from now.
---
The conversation underscores the urgency of addressing these misalignment issues before they become insurmountable, emphasizing the need for a balance between technological capability and wisdom 4.