In examining the future of the like button amidst advances in artificial intelligence, Max Levchin, PayPal co-founder and Affirm CEO, identifies a significant opportunity for leveraging like data to align AI conclusions more closely with human decision-making.
One issue in machine learning is that computers, when presented with a clear reward function, utilize relentless reinforcement learning to optimize performance. However, this often leads to outcomes differing from those a human might achieve through judgment. To address this, AI developers employ reinforcement learning from human feedback (RLHF), integrating human preferences into AI training. A challenge remains in sourcing sufficient human preference data, as RLHF can be costly if human supervisors and annotators are needed to provide feedback.
Levchin proposes that the like button could offer a solution. He considers the vast repository of like data, particularly held by Facebook, as invaluable for developers training AI on human preferences. He suggests this could be one of the most significant assets on the internet, providing crucial data on content preferences for AI model training.
While Levchin’s vision includes AI learning from human preferences via the like button, AI is concurrently influencing how these preferences are shaped. Social media platforms use AI to not only analyze likes but also predict them, potentially diminishing the button’s necessity.
This raises an interesting point, as conventional thought suggested AI’s impact would predominantly be felt in altering the like button’s role. Instead, platforms are applying AI to enhance algorithms. For instance, in early 2024, Facebook experimented with AI to redesign its video recommendation algorithm, leading to improved watch times—a key performance metric.
YouTube co-founder Steve Chen also weighed in on the subject, contemplating a future where the like button might become redundant if AI can accurately gauge user preferences through viewing and sharing patterns. Although the like button has been a straightforward tool for platforms, the ultimate goal is to optimize data use for precision and simplicity.
Chen noted, however, that the like button may remain relevant for accommodating temporary changes in user preferences due to life events. Additionally, he highlighted the button’s importance to advertisers, as it serves as a simple connector among viewers, creators, and advertisers, providing instant feedback and engagement evidence.
