Garbage in, garbage out: mitigating risks and maximizing benefits of AI in research