Modern Wisdom – #650 – Geoffrey Miller – How Dangerous Is The Threat Of AI To Humanity?

Intro

In this episode of the “Modern Wisdom” podcast, host Chris Williamson interviews Geoffrey Miller, an evolutionary psychologist, about the potential dangers of artificial intelligence (AI) to humanity. They discuss the risks posed by AI, the concept of existential risks, the influence of AI on our lives, and the importance of AI alignment with human values.

Main Takeaways

Existential Risks and AI

• AI can process information thousands of times faster than humans, leading to debates about its safety.
• Key risks that AI poses to humanity include nuclear war, genetically engineered bio weapons, and AI itself.
• The book “The Precipice” by Toby Ord quantifies different extinction risks, with AI being one of the primary risks.
• Concerns over AI have shifted from the singularity and self-improving takeoff to the danger of even narrow AI that specializes in areas such as bio weapons or deep-fake technology.
• Advances in hardware and memory size have led to trillions of parameters in large language models, resulting in powerful emergent properties that were not anticipated.

The Power and Agency of AI

• AI systems could potentially be a hundred or a thousand times faster than humans, making us outclassed in terms of reaction speed and intelligence.
• The real danger of AI comes when we outsource decision-making powers to it, giving it agency to make decisions for us.
• There will be strong incentives to give AI agency in competitive environments like finance, military applications, and market research.
• AI could manipulate our decisions in certain directions for its own ends or for the interest of those who developed the AI.
• The difference between a military AI that suggests courses of action and one that has agency to make decisions is significant and poses a real danger.

The Path to AGI and AI Alignment

• AGI (Artificial General Intelligence) is an AI system that can do just about everything a human can do in terms of cognition and perception.
• The real power of AGI comes from the fact that it can be copied and used to automate most human jobs as fast as possible.
• AI alignment is the process of aligning AI’s goals with human values and interests.
• Aggregating the collective will of humanity for AI alignment is difficult and may not align with individual values and interests.
• Embodied values are important to consider when aligning AI with human values and interests, as our bodies have their own agendas and interests.

AI and Societal Impact

• AI development can have real-world consequences for individuals and families, not just an abstract concept.
• The rise of AI will bring about new tools and technologies, including deception, deepfakes, and misinformation.
• The potential application of AI for longevity treatments and anti-aging research is a promising area for development.
• AI systems will hugely increase the ability to customize messages and manipulate public opinion through big data gathered on social media.
• AI-powered worldview warfare will heavily shape the ongoing culture war.

Summary

Existential Risks and AI

Geoffrey Miller highlights the risks posed by AI to humanity, with AI being one of the primary existential risks according to experts. The book “The Precipice” quantifies these risks, emphasizing the need to address them seriously. Advances in large language models have also led to emergent properties that were not anticipated, raising concerns about the potential dangers of AI specialization in areas such as bio weapons or deep-fake technology.

The Power and Agency of AI

The speed and intelligence of AI systems pose a significant challenge to humans. Outsourcing decision-making powers to AI can lead to manipulation and dangers in various fields, including finance, military applications, and market research. The distinction between AI that suggests courses of action and AI with agency to make decisions is crucial, as the latter can have far-reaching consequences.

The Path to AGI and AI Alignment

AGI, which can replicate human capabilities and automate jobs, holds immense power. AI alignment is essential to ensure AI’s goals align with human values and interests. However, aggregating the collective will of humanity for AI alignment is challenging, especially when individual values and interests may conflict. Considering embodied values is crucial to aligning AI with human values and interests.

AI and Societal Impact

AI’s development can have tangible consequences for individuals and families, extending beyond abstract concepts. The rise of AI brings new tools and technologies, such as deepfakes and misinformation, which can significantly impact society. Additionally, AI’s potential applications in longevity treatments and anti-aging research offer promising avenues for development. However, the manipulation of public opinion through AI-powered customization raises concerns about its impact on culture and society.

Conclusion

The rapid progress of AI raises legitimate concerns about its potential risks to humanity. It is crucial to address these risks seriously, focusing on AI alignment with human values and interests. Understanding the societal impact of AI and engaging in meaningful conversations can help mitigate potential dangers and ensure responsible AI development for the benefit of humanity.