These slides from “How to recognize AI snake oil” by Arvind Narayanan does a pretty good job of putting AI – really Machine Learning, ML – capabilities in perspective.
There’s a great example in them of machine learning being given about 13,000 data points on a large number of individuals and still not faring better than simple linear regression done on 4 of those data points.
I’ll also restate my fifth law:
Negative consequences of data linking will be disproportionally larger for the poorest, those less able to defend themselves.
Henning's blog 