How was it predicting the brexit vote? It was a relatively close vote. I think as the article points out there was a lot of luck involved in this specific case.
It's not hard to see why in that moment the risk-reward would look good to a computer. It's not optimizing for individual trades but very much in the aggregate. So Brexit would look to any decent algo like a solid bet: the polls were close but the market didnt accurately reflect that, promising serious potential upside for the contrarian position.