How about That Smart Software?
January 3, 2022
In the short cut world of training smart software, minor glitches are to be expected. When an OCR program delivers 95 percent accuracy, that works out to five mistakes in every 100 words. When Alexa tells a child to put a metal object into a home electrical outlet, what do you expert? This is close enough for horse shoes.
Now what about the Google Maps of today, a maps solution which I find almost unusable. “Google Maps May Have Led Tahoe Travelers Astray During Snowstorm” quoted a Tweet from a person who is obviously unaware of the role probabilities play in the magical world of Google. Here’s the Tweet:
This is an abject failure. You are sending people up a poorly maintained forest road to their death in a severe blizzard. Hire people who can address winter storms in your code (or maybe get some of your engineers who are stuck in Tahoe right now on it).
Big deal? Of course not, Amazon and Google are focused on the efficiencies of machine-centric methods for identifying relevant, on point information. The probability is that most of the Amazon and Google outputs will be on the money. Google Maps rarely misses on pizza or the location of March Madness basketball games.
Severely injured children? Well, that probably won’t happen. Individuals lost in a snow storm? Well, that probably won’t happen.
The flaw in these giant firms’ methods are correct from these companies’ point of view in the majority of cases. A terminated humanoid or a driver wondering if a friendly forest ranger will come along the logging road? Not a big deal.
What happens when these smart systems output decisions which have ever larger consequences? Autonomous weapons, anyone?
Stephen E Arnold, January 3, 2021
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
