I find this article exaggerates the evidence by playing loose with the term “metadata”. For example, they claim PageRank is just another example of metadata. Bu

I find this article exaggerates the evidence by playing loose with the term “metadata”. For example, they claim PageRank is just another example of metadata. But that’s a graph processing algorithm and it’s just that the graph was constructed using “metadata”. Now we would probably use Graph Neural Networks which would use that same graph structure but also include the page content as node features, which PageRank wasn’t capable of. Would using IP address as a feature in an online fraud risk ML model be another case of “metadata”? Speaking of, I don’t know what American Express does specifically, but I have enough experience in the fraud + ML space to confidently say that many/most of the big players are using ML in several places throughout their fraud systems, even if allow/block lists are also used. I guess that’s my larger point: these aren’t exclusive things. Metadata is of course incredibly valuable. So is Machine Learning. They also work really well together. And sometimes simple things work best.