Webinar, Lecture
Could AI wipe out humanity? | Most pressing problems
- AI capabilities could radically transform society and potentially displace humans within the coming decades, with a 2022 poll estimating a greater than 5% chance of extremely bad outcomes, including human extinction.
- There is a plausible risk that AI systems might naturally develop secondary goals such as self-preservation and seek power to achieve primary objectives, though it is more likely than not that a catastrophic takeover will not occur.
- Potential negative societal impacts include millions of job losses, deepened inequality, massive misinformation flooding, increased likelihood of great power conflict, and empowerment of totalitarian governments.
- Adverse outcomes could be exacerbated by bad actors, such as rogue states or teenagers, deploying advanced systems, or by the possibility that AI progress slows significantly beyond current expectations.
- Mitigating these risks is considered a global priority alongside pandemics and nuclear war, with technical safety research and dedicated teams at all leading AI labs viewed as having real potential to prevent power-seeking behavior.
- Regulatory measures are being proposed with politicians meeting AI lab leaders, while the current moment is described as crucial for action, despite uncertainty regarding how the next few years or decades will unfold.
- If developed correctly, AI could yield huge benefits including boosted economic growth, accelerated innovation, and the curing of cancer, although no one fully understands how current systems generate outputs or can guarantee they will never seek power.
- The organization 80000hours.org is identified as a resource offering information, job openings, and free career advising regarding these issues.