I have a Friendly AI proposal at http://www.fungible.com/respect. It got some good feedback on the SL4 list some time back, so I have a list of bugs to fix in the proposal that have not yet been fixed. They do seem fixable, though.
Assuming they are fixable, we’d have a specification for a Friendly AI that would be mathematically well-defined, but not easily implementable.
It takes an ordinary approach to cause-and-effect, so if it were confronted by Newcomb’s problem it would either self-modify or lose, depending on whether the entity in control in that story manages to read its mind before it was able to self-modify.
Furthermore it assumes that there is a real world and that the humans it is responsible for care about the portion of the real world that they interact with, so humans concerned about simulations or about events happening to copies of themselves that are unreasonably far away won’t be well served by it.
(There are lots of more practical issues on the bug list, too, and frankly I’m not very concerned about the issues stated above.)
I have a day job that can’t plausibly be called research.
I have a Friendly AI proposal at http://www.fungible.com/respect. It got some good feedback on the SL4 list some time back, so I have a list of bugs to fix in the proposal that have not yet been fixed. They do seem fixable, though.
Assuming they are fixable, we’d have a specification for a Friendly AI that would be mathematically well-defined, but not easily implementable.
It takes an ordinary approach to cause-and-effect, so if it were confronted by Newcomb’s problem it would either self-modify or lose, depending on whether the entity in control in that story manages to read its mind before it was able to self-modify.
Furthermore it assumes that there is a real world and that the humans it is responsible for care about the portion of the real world that they interact with, so humans concerned about simulations or about events happening to copies of themselves that are unreasonably far away won’t be well served by it.
(There are lots of more practical issues on the bug list, too, and frankly I’m not very concerned about the issues stated above.)
I have a day job that can’t plausibly be called research.