3 months ago
- The term 'artificial intelligence' (A.I.) is misleading and potentially dangerous, as it can lead to mismanagement by promoting misunderstanding.
- Many A.I. researchers fear doomsday scenarios, including human extinction, but these arguments are often irrational and vague.
- A.I. should be viewed pragmatically as a tool or form of social collaboration, not as an independent intelligent creature.
- Programs like GPT-4 and image generators are innovative mashups of human-created data, not new minds, highlighting hidden concordances between human creations.
- Flexibility in A.I. systems comes from mathematical processes, not intelligence, and can make technology more human-centered by requiring human choice.
- A.I. has potential upsides, such as reducing digital rigidity and increasing user agency, but also risks of manipulation and loss of control.
- Current A.I. policy debates around terms like 'alignment,' 'safety,' and 'fairness' are vague and may not quell fears, leading to circumvention.
- There is broad agreement on labeling deepfakes and automated interactions to ensure transparency and user choice.
- The 'black-box' nature of A.I. systems must be addressed through transparency, though this is challenging due to the complexity of large models.
- 'Data dignity' proposes tracing human contributions in A.I. outputs, potentially creating new economic models and social institutions.
- Data dignity could help mitigate economic displacement by valuing and rewarding unique human contributors, rather than relying on universal basic income.
- Implementing provenance and context for digital information is key to preventing harm, such as deepfake chaos, and controlling A.I. systems.