Very good read along with links within the article. Sure, even with algorithms whose internals are clear unlike NN, often it's hard to predict results or understand why the result was produced. I followed an algorithm from very basic, having read papers which describe mathematical basis of the algo, and thought I fairly understood them. But I still can't predict what it'll produce until I actually run my program. Yes, those threats may be overhyped, but considering growing adoption of ML in every fields, as much safe guard should be put before they actually happen. It's better safe than sorry.
The entire tech industry today would have been defined as malware 25 years ago. They insist on facilitating the means for software to secretly communicate info without the users knowledge of what is being communicated or who it is communicated to. Then when you add to that the deliberate subversion of security protocols the implementation of hidden backdoors, the infiltration of software projects, I can guarentee you, AI will definately be used for malicious purposes and it wont be by small groups of hackers it will be by the same bunch of lying deceitful tech corporations and government agencies that are responsible for the pitiful state of infosec as it is today because we can be damn sure they have no intention of allowing AI to put a stop to the political surveillance, user profiling, data mining network they have spent 20 years creating. I predict AI will be used as a super administrator on all devices, itself networked between devices to keep every human being in the developed world under closer surveillance than ever before, it will profile everyone and alert authorities if we so much as drop a piece of litter never mind actually speak out against any of it.
Rock, that's not "will" but "is". Most prominent example would be Chinese skynet, but if American or European think it's somebody else's problem, that's wrong. The difference is just how publicly or covertly and how widely or limitedly they do it. The most awful part is this: we know how, after 9.11, flood of info they gathered overwhelmed NSA & CIA and made them essentially incompetent, but ML can solve that problem and makes them really worst Orwellian system ever.
Yes I agree Yuki, but at least you can say the Chinese Government are open about it. Aside from those of us who are involved in or take a special interest in infosec, I dont think a whole lot of other Americans and Europeans realize to what extent this surveillance network has been implemented and how we are teetering on the edge of that nightmare Orweillian state.