The enthusiasm for machine learning, and deep learning in particular, brings back my memory of the heydays of quantitative investment management in the late 1990’s and early 2000’s, when many people believed in the power of econometrics to handle any investment decisions. While quant investing did make some headways and has become an important tool, it is a far cry from that rosy prediction. In my view, the main reason is because it is trying to use crude (and possibly wrong) statistical methods to model collective human psychology and behavior, of which we humans have little understanding to begin with.
Using young male mice, they applied relatively mild jolts, designed to result in a sudden, strong jerking of their heads, much as occurs during head-to-head tackles and other impacts. Afterward, some animals showed symptoms of a rodent version of a concussion, stumbling and performing poorly on memory tests.
The scientists then injected some animals with a dye that cannot cross a healthy blood-brain barrier and scanned the living animals’ brains. In about half of the mice, they saw signs of the dye in their brains, indicated that their blood-brain barriers had become permeable. Many of the mice also showed signs of leaky blood vessels and other damage, including inflammation and disruptions in the electrical activity within their brains.