AI within five years may be used for mass deception, weapons development, public manipulation, political abuse and cyberattacks, according to an MIT study.