Part 2/9:
Shapiro begins by emphasizing the importance of heuristic imperatives—rules of thumb guiding AI behavior in complex, real-world scenarios. Unlike strict formal rules, heuristics rely on instinct, experience, and creative thinking to solve problems efficiently. He asserts that for safe and benevolent AGI, embedding such imperatives as core objectives can help ensure the AI remains aligned with human wellbeing.
He uses GPT-3’s latest Davinci engine to demonstrate how prompts about definitions and behaviors can illustrate these principles in action. The process underscores that advanced language models already possess nuanced understanding, which can be harnessed to promote benevolent outcomes.