4 ms·

Without embodiment, these AIs are just maximising some fitness function, which their creators have a hand in by selecting the data and manually tweaking (see th

by bArray 4y ago

Without embodiment, these AIs are just maximising some fitness function, which their creators have a hand in by selecting the data and manually tweaking (see the Fascism vs Communism example). The real answer here is that the "smarter" these AIs get, they more accurately reflect either the political will of their creators or the material they train on (which may by means of the political will of the creators). In some cases it may get some indication you may have a certain political opinion and then give you answers based on this - which would be quite a smart behaviour.

visarga 4y ago
AI can train in complex simulations and games, not just to imitate human text. It needs grounding, and that can come from a variety of sources - if you connect the model to a simulator, a code execution engine, a game or a math solver, it can learn from interacting with these systems.
- lossolo 4y ago
  All of those are inputs and you can restrict them the same way you restrict a text dataset for training to achieve the results you desire - a model that leans right or left, one that wins a game using a certain strategy, or generates code using specific patterns.
- bArray 4y ago
  As far as I'm aware, we are yet to create a complex simulation or game as rich as the world around us. That said, even if we could, games often have somewhat well-defined fitness functions. The other problem with simulated problem spaces is as the other comment suggests, you can on purpose or accidentally introduce all sorts of biases. "Our robot got smarter and it was left-wing", well, you did simulate it living in Silicon Valley. "Our AI got smarter and it was right-wing", well, you did simulate it living in a small Texas town as a priest. As it happens, these AIs are currently being created Silicon Valley companies that are known to have a left-wing bias. So if any bias from the creators was to filter into the training system, we may have some idea in which direction it pushes.

by bArray 4y ago

visarga 4y ago

AI can train in complex simulations and games, not just to imitate human text. It needs grounding, and that can come from a variety of sources - if you connect the model to a simulator, a code execution engine, a game or a math solver, it can learn from interacting with these systems.

lossolo 4y ago
All of those are inputs and you can restrict them the same way you restrict a text dataset for training to achieve the results you desire - a model that leans right or left, one that wins a game using a certain strategy, or generates code using specific patterns.
bArray 4y ago
As far as I'm aware, we are yet to create a complex simulation or game as rich as the world around us. That said, even if we could, games often have somewhat well-defined fitness functions. The other problem with simulated problem spaces is as the other comment suggests, you can on purpose or accidentally introduce all sorts of biases. "Our robot got smarter and it was left-wing", well, you did simulate it living in Silicon Valley. "Our AI got smarter and it was right-wing", well, you did simulate it living in a small Texas town as a priest. As it happens, these AIs are currently being created Silicon Valley companies that are known to have a left-wing bias. So if any bias from the creators was to filter into the training system, we may have some idea in which direction it pushes.