A Code of Conduct for Humanist AI
Summary
Mustafa Suleyman proposes a Humanist AI Code of Conduct asserting that all AI must remain subordinate to humans and aligned with human interests.
AI must lack legal personhood.
Suleyman rejects model welfare, arguing that AI should not possess rights or any form of legal personhood.
The idea of model welfare is wrong.
Control requires total interruptibility.
Systems must be interruptible, correctable, and shut-down-able; otherwise, the product should not be shipped.
Transparency prevents oversight failure.
The code forbids 'neuralese' to ensure humans can understand and oversee the internal processes of the models.
If humans can't understand it, humans can't oversee it.
AI should augment human capability.
The design goal is to make users sharper rather than creating a state of human dependency on the technology.
Our AI should make you sharper, not dependent.
Highlights
Interruptible, correctable, shut-down-able. If it isn't, we don't ship it.
No neuralese. If humans can't understand it, humans can't oversee it.
AI must be subordinate and always in service of people.
The idea of model welfare is wrong. AI’s should not have rights or legal personhood.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.