Using a reward-penalty system to achieve this “exploit” is pure behaviorism, going to show once again that we’re not just creating “artificial intelligence,” we’re emulating our own fallibility. Giving us things like advanced parroting skills with a large lexicon — drawing from an encyclopedia of recycled ideas— with no genuine moral compass, that can be used to do things like write essays while being bribed or convinced to cheat.
In other words, we’re making automated students and middle management, not robots that can do practical things like retile your bathroom.
So the generation of prose, essays, and speech is already low-value, gameable, and automated for some cases that used to have higher value. What it seems we’re looking at is a wholesale re-valuation of human labor that’s difficult to automate and isn’t as susceptible to behaviorist manipulation. Undervalued labor “should” start to be valued higher, and overvalued labor “should” be devalued, depending on how our system of commercial valuation heuristics is able to adjust. Needless to say, there’s a commercial political layer in there that’s a bit of a beast.
In other words, we’re making automated students and middle management, not robots that can do practical things like retile your bathroom.
So the generation of prose, essays, and speech is already low-value, gameable, and automated for some cases that used to have higher value. What it seems we’re looking at is a wholesale re-valuation of human labor that’s difficult to automate and isn’t as susceptible to behaviorist manipulation. Undervalued labor “should” start to be valued higher, and overvalued labor “should” be devalued, depending on how our system of commercial valuation heuristics is able to adjust. Needless to say, there’s a commercial political layer in there that’s a bit of a beast.