Commentary: Certainly, there are some very detrimental consequences of inexplicability. What is the morality of in/transparency? In which domains do we need public or multi-stakeholder transparency to mitigate errors, biases, abuse or manipulation, etc.? Conversely, are there areas where it's justifiable that explanatory functions should be in the hands of only one party? - Public health decisions for which mass panics are a risk, individual or group identity issues where psychological tensions could arise, competitive decisions in corporate strategy, national security or intel analysis, crime investigations, international trade negotiations, dating / trading / shopping agents, personal bouncer bots, etc? Are the rules for the AI-age different than in pre-AI era, because it penetrates our lives more deeply, pervasively, and with more 2nd and 3rd order unintended consequences?
Original article: http://medcitynews.com/2017/08/darpa-researchers-want-know-machine-learning-algorithms-get-wrong/