Entire books have been written about this, but in a sentence: AI is very likely to be dangerous by default, since it won't have human values/morality by default

Entire books have been written about this, but in a sentence: AI is very likely to be dangerous by default, since it won't have human values/morality by default, and formalizing human values is very difficult. Seriously look at Superintelligence, or the the Intelligence Explosion FAQ (http://intelligence.org/ie-faq/ http://intelligence.org/ie-faq/) for far more depth into these issues.