“In some ways, adversarial policies are more worrying than attacks on supervised learning models, because reinforcement learning policies govern an AI’s overall behavior.If a driverless car misclassifies input from its camera, it could fall back on other sensors, for example.” TIL fail-safe components are 1) ubiquitous 2) work 3) only an option for supervised learning components.
“A supervised learning model, trained to classify images, say, is tested on a different data set from the one it was trained on to ensure that it has not simply memorized a particular bunch of images. But with reinforcement learning, models are typically trained and tested in the same environment.”
First, a RL environment is not equivalent to a supervised learning data set. Second, the train validate test paradigm is not thrown out in RL research, its why OpenAI put their Starcraft agent on public ladders.
“The good news is that adversarial policies may be easier to defend against than other adversarial attacks.” This sentence refers to Graves et al. adversarially training their agents. Adversarial training is, of course, also conducted frequently in supervised learning.
The BRK official position on Valeant is that they take advantage of sick individual's life-or-death need for drugs to price gouge them since purchasing decisions are not made on price (medication is "price inelastic").
Clayton homes is criticized for having bad or misleading loan terms. BRK (and many financial services individuals I know personally) often see just about any loan term as fair as long as no-one was forced into the loan at gunpoint. Is it the responsibility of the lessor or lessee to make sure the lessee is signing to something in their best interest? Loans are not life-or-death and purchasing decisions are frequently made on price (loans are "price elastic").
The article you've linked presents these two cases and makes the argument that they are equivalently immoral, making Munger a hypocrite. Personally, I think they're both fair criticisms, but they definitely come from different places. I don't see necessary cognitive dissonance in having the opinion that only one of these is immoral.
With regard to cutting costs: I see no indication that BRK criticized Valeant for firing people, so I suspect this is a personal issue for the author of the article, so I haven't addressed it here.
I appreciate you linking this - I've read at least a books worth of panicked news coverage over the past week and had not seen a single story about potentially successful application of antivirals. I guess fear sells better.
“In some ways, adversarial policies are more worrying than attacks on supervised learning models, because reinforcement learning policies govern an AI’s overall behavior.If a driverless car misclassifies input from its camera, it could fall back on other sensors, for example.” TIL fail-safe components are 1) ubiquitous 2) work 3) only an option for supervised learning components.
“A supervised learning model, trained to classify images, say, is tested on a different data set from the one it was trained on to ensure that it has not simply memorized a particular bunch of images. But with reinforcement learning, models are typically trained and tested in the same environment.” First, a RL environment is not equivalent to a supervised learning data set. Second, the train validate test paradigm is not thrown out in RL research, its why OpenAI put their Starcraft agent on public ladders.
“The good news is that adversarial policies may be easier to defend against than other adversarial attacks.” This sentence refers to Graves et al. adversarially training their agents. Adversarial training is, of course, also conducted frequently in supervised learning.