The conversation revolves around the power dynamics among institutions and political systems, the incentives in distributed economic systems, and the psychology of engineers and leaders who deploy AGI, with a discussion on human nature and our capacity for good and evil at scale. The potential for AI to empower humans to create, flourish and succeed is undeniable, but it also raises concerns about the vast difference in intelligence between people and AI systems.
The AGI Conference is the primary gathering of scientists and researchers who are strong advocates of the pursuit of AGI and not just deep neural net and deep reinforcement learning. They believe large scale nonlinear regression is not the only path to AGI.
The negative consequences of poorly defined objectives in AI can already be seen in social media and relatively weak machine learning algorithms. The AI safety community's main objective is to properly define objectives to avoid catastrophic consequences.
In this podcast, the speakers discuss what they would ask an AGI system and how it could help identify blind spots humans have when it comes to big problems in the world. They also explore some possible novel solutions to these unexplored problems.
As we test and improve new technology like AI, our understanding of its safety concerns and how to address them can rapidly evolve over time, leading to a developing philosophy for AI safety. However, many prior predictions about AI have turned out to be incorrect.