This looks like an article about a cruise ship. And it is! It doubles as a lesson for anyone deploying AI-based systems.

(Source: WSJ, "Inside the Strange and Lonely Test Run of a New Cruise Ship")

The takeaway lesson? If you really care about the customer experience, you have to test.

I mean, really test. Then keep testing, and test some more. And when you find an issue, you have to correct it.

If you're accustomed to running tests for software apps, you're pointing in the right direction. AI systems take that testing to another level because the underlying models can introduce non-deterministic effects. That means a test might pass three times and then fail on the fourth.

If you apply this cruise line's idea of a "shakedown test" to your AI, you'll catch problems earlier. That reduces your downside risk exposure and leaves you open to the upside gain that AI has to offer.