The game world could include a sign stating the rules/objective of the game in English. Presumably then (perhaps helped by GPT-3) the agents might well learn to find & read it, as the best shortcut to working out what to do. And figure out both the meanings of individual words (nouns, adjectives, verbs) and grammar (word order, conjunctions, etc.) so as to understand whole sentences.*
And thus make the leap from GPT-3 as clever spinner of meaningless symbols like R, E and D to those symbols having a real-world meaning like ‘RED’. Or even ‘Keep the red cube within line of sight of the yellow pyramid, but don’t let the blue agent see either of them’.
Hence leaping from Chinese Room to something more mind-like?
(*Though they might need help, e.g. a pre-coded ‘language instinct’, as trial and error may take a v long time to figure these out.)
The game world could include a sign stating the rules/objective of the game in English. Presumably then (perhaps helped by GPT-3) the agents might well learn to find & read it, as the best shortcut to working out what to do. And figure out both the meanings of individual words (nouns, adjectives, verbs) and grammar (word order, conjunctions, etc.) so as to understand whole sentences.*
And thus make the leap from GPT-3 as clever spinner of meaningless symbols like R, E and D to those symbols having a real-world meaning like ‘RED’. Or even ‘Keep the red cube within line of sight of the yellow pyramid, but don’t let the blue agent see either of them’.
Hence leaping from Chinese Room to something more mind-like?
(*Though they might need help, e.g. a pre-coded ‘language instinct’, as trial and error may take a v long time to figure these out.)