Tuesday, June 20, 2017

AI/DL/ML win or fail?!

So, I recently received the following recommendations from Quora of topics / people to follow:
This merits some work to figure out what is going on here. Note the first recommendation to follow "Jewish Theology and Philosophy." Expected recommendation: I am Jewish and I 'follow' several other Jewish topics on Quora. The Ariel Sharon recommendation is, however, thought provoking. It is true that Ariel Sharon, deceased in 2014, was Jewish and was a leader of the State of Israel. Might be reasonable to suggest this recommendation based on my Judaism. Maybe Quora has access to some of my personal data (unlikely?) that shows I've been in Israel several times (I got married there). Quora says it is making the recommendation, "Because you follow Sharon, MA." At first glance, you might think this is an AI/DL/ML fail confusing the name of the person with the name of the town. I follow Sharon, MA because that is where I grew up. If you know a little more about Sharon, you know that it has (or had) a large Jewish population. It also has had a number of youth make various trips to Israel and a Jewish, Sharon teen was killed in Israel not too long ago. So, now, it's looking like the AI/DL/ML algorithm is really smart and has examined far more data that we would think. Or not. As we start to develop AI/DL/ML in medical and medical imaging contexts, we're going to run into similar situations where we are not quite sure if the AI/DL/ML is brilliant or a moron. Of course, we're all used to dealing with the 'idiot savant' resident who occasionally blurts out a brilliant answer. Interesting times ahead.

No comments: