How Stanford researchers are designing fair and trustworthy AI systems