Making AI accountable easier said than done, says U of A expert