Profile cover photo
Profile photo
Alejandro Weinstein
Alejandro's posts

Post has shared content

Post has shared content
VLOG from Valpariso Chile now live.

Post has shared content
No, you couldn't ride it either.

Post has shared content
Recent Uber troubles in Europe remind me of the history of the world's first steamship (from Why Nations Fail): 

In 1705 Papin, a professor of mathematics at the university of Marburg, used a rudimentary engine he constructed to build the world's first steamboat. He decided to steam the boat down the river Fulda to the river Weser. At that time, river traffic on the Fulda was the monopoly of a guild of boatmen. His friend Leibniz wrote to the Electrof of Kassel for permission to pass through, however that petition was rejected. Undeterred, Papin decided to make the journey anyway. When his steamer arrived at Munden, the boatmen's guild first tried to get a local judge to impound the ship, but was unsuccessful. The boatmen then set upon Papin's boat and smashed it and the steam engine to pieces.

Post has shared content
One sometimes hears political activists complain of "suspiciously convenient timing" when media coverage of an important and controversial event related to their cause is drowned out by some unrelated news event dominating the headlines.  For instance, in domestic US politics, there were suspicions raised recently in some quarters that the various votes to give fast-track authority to the US president to negotiate the Trans-Pacific Partnership (TPP) agreement were deliberately timed to coincide with other headline-grabbing events, such as the multiple shootings in Charleston, or the recent significant Supreme Court decisions on the Affordable Care Act and marriage equality.

But actually, the laws of probability predict such "convenient" coincidences to occur with a surprisingly high frequency, even in the absence of any conspiracy to co-ordinate timing (for reasons similar to those behind the birthday paradox, linked below).  To oversimplify things slightly, if

* a controversial event that the authorities would prefer not to draw attention to occurs every X days on average; and

* a spectacular event dominating the headlines occurs every Y days on average;

then just from random chance, we would have

* a coincidence in which a controversial event conveniently occurs on the same day as a spectacular event will occur every XY days on average.

The frequency of such coincidences also rapidly increases if one widens the window defining such a coincidence, for instance

* a controversial event will occur within a day of a spectacular event every XY/3 days on average;

* a controversial event will occur on the same week of a spectacular event every XY/7 days on average;

and so forth.

For instance, in this month of June in domestic US media, "controversial" events included two votes for TPP and one vote to reauthorise the Patriot act, while "spectacular" events included the Charleston shooting, two major Supreme Court decisions, a celebrity very publicly becoming transgender, and an extensively covered prison break, among others.  This suggests values of X and Y comparable to 10 and 5 respectively,  so that same-day coincidences should be expected about once every two months, and within-a-day coincidences once every two to three weeks.  This is admittedly a very back-of-the-envelope calculation involving a fairly short time interval of analysis and a rather subjective interpretation of what is "controversial" or "spectacular", but I expect these numbers to be of about the right order of magnitude.

One could in principle try to set up a more careful test to see if there is a statistically significant correlation between the timing of "controversial" and "spectacular" events, though given human nature I would expect that a negative result for such a test would do little to change the minds of someone who sees such a coincidence and is convinced that it is not purely due to chance.

EDIT: the same analysis suggests that a day with two or more spectacular events should occur about once every 2Y^2 days, on the average.  It seems that today could be one of these days, at least as far as the US media is concerned.

Post has shared content
The journal Nature just published a review paper on deep learning co-authored by myself, Yoshua Bengio​, and Geoff Hinton.

It is part of a "Nature Insight" supplement on Machine Intelligence, with two other articles on machine learning by Zoubin Ghahramani​ (probabilistic machines) and Michael Littman​ (reinforcement learning).


Insight supplement on AI:
Wait while more posts are being loaded