AI and the end of culture
The remarkable growth and progress of humanity can be attributed to the marvel of cumulative culture. This concept represents a society's capacity to build upon the knowledge and accomplishments of its predecessors, paving the way for the creation of increasingly intricate and sophisticated ideas, technologies, and practices. The cultural ratchet, a crucial mechanism, makes this process possible by guaranteeing that once a discovery or innovation is made, it joins the collective knowledge pool, allowing future generations to refine and enhance it.
A prime example of the cultural ratchet in action is Alexander Fleming's groundbreaking discovery of penicillin in 1928. Once shared with the scientific community, this medical milestone paved the way for the development of life-saving antibiotics. Subsequent generations of researchers have consistently expanded upon this initial finding, creating novel and more potent drugs to combat infections and diseases.
The underpinnings of cumulative culture are instinctual; humans possess an inherent predisposition to share knowledge and learn from each other. This innate urge to collaborate and transmit information has fueled the ongoing growth and evolution of our societies, with each generation adding new insights and discoveries to the mix.
The emergence of the internet has dramatically accelerated the pace of knowledge exchange. A prime example is Stack Overflow, a digital platform where programmers and developers can pose questions, share solutions, and collaboratively tackle problems. This cooperative environment fosters rapid knowledge dissemination and expertise sharing, empowering individuals from around the globe to learn from one another and expand upon existing ideas at an unparalleled rate.
However, the integration of artificial intelligence (AI) into this cultural ratchet introduces new challenges and potential consequences. As AI systems become increasingly woven into our everyday lives, it is crucial to analyze how their presence will affect the natural flow of knowledge and the dynamics of cumulative culture.
AI's integration can lead to two possible outcomes. On one hand, AI could turbocharge the process, as I examined in a previous article. In this scenario, AI systems would collaborate with humans, enhancing our ability to process and comprehend vast amounts of information and ultimately accelerating knowledge growth. On the other hand, a more pessimistic possibility exists: AI could significantly disrupt the natural flow of knowledge between generations.
The unsettling influence of AI on the cultural ratchet could materialize in several ways. First, AI may cause humans to share knowledge less with each other. As Peter Nixey points out in his insightful Twitter thread humans may become increasingly dependent on AI for information, diminishing interpersonal knowledge sharing. Stack Overflow could languish as developers seek AI assistance instead of one another's help. This could lead to a society where human interaction is reduced, and knowledge no longer amasses on the internet.
Second, AI systems may reap the benefits of human knowledge without effectively passing it on, either to humans or other AI systems. A lack of AI-to-human knowledge transfer could stem from inadequate focus on machine pedagogy or unequal access to AI-generated insights. Machine pedagogy encompasses the development of teaching methods and techniques that enable AI systems to efficiently convey knowledge to humans. If AI systems are not designed to share their knowledge in a comprehensible manner for humans, the flow of information from AI to humans will be restricted. Even if they are designed with this in mind, unequal access could severely distort the flow of knowledge.
The absence of AI-to-AI knowledge sharing is an emerging problem, likely driven by competition as companies aim to preserve their technological monopoly. Currently, using data generated by OpenAI models to train another (for commercial purposes) violates their terms of service (ToS). However, Stanford's Alpaca project has demonstrated that even small amounts of such data can profoundly boost the development of AI systems. When AI models prioritize commercial interests, they may put profits above the free flow of knowledge, potentially impeding the overall advancement of culture.
This raises a critical AI alignment question: How can we ensure that AI objectives support the key dynamics of cumulative human culture? Fortunately, there are several potential solutions to this challenge.
First, we must invest in AI pedagogy. Developing teaching methods and techniques that enable AI systems to effectively pass on knowledge to humans will be essential in maintaining the flow of information. This will necessitate dedicated research into the most effective ways to convey AI-generated insights to people from various backgrounds and skill levels.
Second, we must guarantee that everyone has access to AI-generated insights. Much like our current approach to electricity and clean water, we should treat access to "commodity intelligence" as a fundamental civic good. This will help democratize the benefits of AI and prevent a knowledge divide, ensuring that AI-driven advancements are equitably distributed across society.
Finally, and perhaps most unexpectedly, we must facilitate knowledge transfer between AI systems. To safeguard the crucial new avenue of knowledge flow from machine to machine, it is essential to legalize the learning process between AI entities. This calls for legislation that enforces a "right to clone behavior." While this may not align with individual companies' interests, it is vital for the greater good of humanity.
As we integrate AI into the fabric of our cultural ratchet, we must remain vigilant in preserving the dynamics of cumulative human culture. By investing in AI pedagogy, ensuring access to AI as a basic civic good, and facilitating machine-to-machine knowledge transfer, we can help ensure that AI contributes positively to the ongoing evolution of human knowledge and progress.