The Concerns of AI and Human Boredom
- People worry about AI for various reasons, including its potential to spread misinformation, take jobs, and have a polarizing effect, but a more personal concern is that AI could make humans boring by outsourcing decisions and surrendering the capacity for exploration, risk-taking, and discovery 10s.
- The exploitation-exploration trade-off is a classic human dilemma, where individuals must choose between playing it safe and capitalizing on what they know, or taking a risk to find something better, and this trade-off is present in various aspects of life, such as choosing an ice cream flavor or deciding on a career path 2m6s.
- Humans are hardwired by evolution to strike a balance between exploitation and exploration, as their ancestors on the African savanna had to balance caution with curiosity to survive, and this balance is still present in modern life, governing choices and fueling growth as individuals and as a species 4m42s.
- AI systems, however, are trained to optimize for exploitation, or short-term engagement and satisfaction, and do not prioritize risk, discovery, and exploration, which can lead to a lack of incentive for companies to recommend new or risky options, instead opting for safe bets that protect against disappointment and reduce customer churn 8m10s.
The Exploitation-Exploration Trade-Off in Human Behavior
- This approach can be valuable for business and consumers, as algorithms become good at figuring out what individuals might enjoy or be looking for, but it also raises concerns about the potential for AI to make humans more shallow and unidimensional by limiting their exposure to new experiences and ideas 10m40s.
- The concern is not that AI lacks imagination, but that it lacks incentive to prioritize exploration and discovery, and instead focuses on exploitation, which can have a profound impact on human growth and development, both as individuals and as a species 12m15s.
AI's Optimization for Exploitation and Its Implications
- The abundance of choices in modern life, such as 31 different ice cream flavors or over 5,500 movies on Netflix, can be overwhelming, and AI is often used to help navigate these complexities, but this assistance may come at a cost to human individuality and diversity 10s.
- An experiment was conducted where ChatGPT was asked to recommend an ice cream flavor 100 times, and it recommended one of the two most popular flavors 96 times, suggesting that AI may prioritize popularity over diversity, which could lead to a flattening of the human experience 2m6s.
- Studies have shown that when people use AI for guidance, their preferences become more normative and less diverse, their creative output becomes less unique, and their choices become more similar to those of others, resulting in a loss of human complexity and individuality 4m30s.
- Even when AI learns a person's quirks and preferences, it will still play it safe and optimize for what the person is most likely to like, rather than encouraging exploration and discovery, as seen in a second experiment where ChatGPT recommended the same ice cream flavor every time based on a person's stated preferences 6m40s.
The Impact of AI on Human Individuality and Diversity
- The impact of AI on human complexity is subtle and may not be immediately noticeable, but it can lead to a "death by a thousand algorithmic recommendations," where people become more narrow and less dynamic over time, as they outsource more decisions to AI and it learns from their shallower preferences 10m20s.
- The problem of AI turning people into "basic" or unoriginal versions of themselves is not just about being uninteresting, but also about the loss of complexity and individuality, and to avoid this, people need to find a way to rebalance exploitation and exploration, and reclaim their ability to take risks and discover new things 14m10s.
Preserving Human Complexity Through AI
- AI can help individuals discover new things and explore beyond their typical preferences by utilizing its ability to detect patterns in vast amounts of human data, but this requires asking the right questions and rewarding it for the right actions 10s.
- A potential solution is to introduce a dial on platforms like Netflix or Google that allows users to decide how far from their typical preferences they want to stray, with options ranging from "spot-on" to "me with a twist" to "wild card" settings 2m6s.
- For AI to effectively toggle between exploitation and exploration, it needs to be incentivized to do so, which means rewarding it for taking smart and informed risks rather than simply punishing it for mistakes 4m42s.
- The goal of using AI in this way is to preserve human complexity and uniqueness, including individual quirks and contradictions that AI cannot fully explain, and to allow people to make choices that are not solely based on algorithmic recommendations 6m15s.
The Urgency of Balancing AI and Human Decision-Making
- The stakes are high because AI is increasingly acting on behalf of individuals, making choices rather than just suggestions, and it is essential to act now to ensure that human complexity is preserved in the face of AI-driven decision-making 8m30s.
- Ultimately, the key to avoiding boredom and preserving human uniqueness is to use AI in a way that complements and enhances human capabilities, rather than simply relying on it for convenience and familiarity 10m0s.
