There's a larger, nuanced point here: statistical ML based "AI" is simply an API over aggregated samples of human intelligence. The sampling suffers from bias (

There's a larger, nuanced point here: statistical ML based "AI" is simply an API over aggregated samples of human intelligence. The sampling suffers from bias (the "white men" example in this thread), gaps (the OP's example, which relies on "shirts without stripes" being an uncommon, though perfectly acceptable phrase), and many more such shortcomings of what is essentially "rote learning".