3 thoughts on “Using machine learning to reverse engineer machine learning. Via Alok Tiwari”
I wrote about this several months ago. The full spectrum will include AI’s competing with each other; AI’s partially competing, partially collaborating, AI’s fully collaborating with each other; AI’s networked or merged becoming stronger entities.
Um: “they found they could produce software that was able to predict with near-100% accuracy the responses of the AI they’d cloned, sometimes after a few thousand or even just hundreds of queries.” >Cloning a box black after just hundreds of queries gives me pause that these are small “toy” AIs for proof of concept. Think about cloning Google’s indexing algorithm or Siri.
Perhaps but the point is that they claim to be able to do it faster and with less effort than it took to train the original from training data. If true then that’s a significant finding.
I wrote about this several months ago. The full spectrum will include AI’s competing with each other; AI’s partially competing, partially collaborating, AI’s fully collaborating with each other; AI’s networked or merged becoming stronger entities.
LikeLike
Um: “they found they could produce software that was able to predict with near-100% accuracy the responses of the AI they’d cloned, sometimes after a few thousand or even just hundreds of queries.” >Cloning a box black after just hundreds of queries gives me pause that these are small “toy” AIs for proof of concept. Think about cloning Google’s indexing algorithm or Siri.
LikeLike
Perhaps but the point is that they claim to be able to do it faster and with less effort than it took to train the original from training data. If true then that’s a significant finding.
LikeLike