If I were going to make a gambling analogy, I'd imagine a casino game as follows: draw a card from a standard playing card deck. If it's a spade you lose. Otherwise you win.

For any one hand you may win or lose, but on average you should still take the "gamble" every time, because the odds are so good you'll win. That's what using AI is like right now.

This is a new state of things - a year ago, the analogy would have been reversed.