On 2018-02-25 11:13, 10r wrote:
Hi. I wonder if there has ever been a topic about AI threats against humanity. If not, I would like to propose this discussion. Should we think of models / agents that only work on encrypted information such as numer.ai or should we just think about how to develop such agents / models safely (if that is possible)?
One AI topic I wonder about is, how might it be applied for profit and social control in domestic situations. Imagine an Amazon Echo-like device in the home, surveilling family/roommate members' interactions and arguments, and reacting with evaluations and advice about what the family/roommate members should do differently -- including which purchases to make -- based on whichever therapy paradigm was being imposed, whatever values were assumed to be true. What things might go really well in this situation? What things might go disastrously wrong? For example, could the device be at all sensitive to subtext, to power dynamics (who controls the definition of each situation?), or would it just have a very surface level take, like hey, this one family/roommate member's anger exceeded the permissible limit, that person now needs to apologize and swallow a pharmaceutical. I also wonder what kind of key words/phrases to plug into search engines to find out more about this possible scenario, and what kind of sources I should be looking at -- trade journals? Are any industries, like the psychology/medical industry, developing AI along these lines? The closest I've found so far is the concept/field of "affective computing": https://en.wikipedia.org/wiki/Affective_computing If anyone has thoughts or suggestions, please share, thanks!