AI hallucinations will be solvable within a year, ex-Google AI researcher says—but that may not be a good thing